The Oxford Guide to Australian Languages 9780198824978, 0198824971

Table of contents :
The Oxford Guide to Australian Languages
Short contents
Detailed contents
Series Preface
Abbreviations and conventions
The Contributors
Language maps
Key to language maps
Australian language families and linguistic classifications
Chapter 1: Introduction
Part I: Background
Chapter 2: A history of the early description of Australian languages
Chapter 3: Documentation of Australian languages
Chapter 4: Australian languages and syntactic theory
Chapter 5: Australian languages and interdisciplinary approaches to the past
Chapter 6: Nineteenth-century classifications of Australian languages
Chapter 7: How many languages are and were spoken in Australia?
Chapter 8: Philological methods for Australian languages
Part II: Structures
A. Phonetics and phonology
Chapter 9: Articulatory and acoustic phonetics
Chapter 10: Segment inventories
Chapter 11: Phonotactics
Chapter 12: Morphophonology: Lenition and assimilation
Chapter 13: Nasal cluster dissimilation
Chapter 14: Lexical stress
Chapter 15: Intonation
Chapter 16: Sound change
B. Morphosyntax
Chapter 17: Word classes
Chapter 18: The noun phrase
Chapter 19: Noun classes
Chapter 20: Ergativity
Chapter 21: Semantic case
Chapter 22: Possession
Chapter 23: Demonstratives
Chapter 24: Pronouns
Chapter 25: Adjectives and adverbs
Chapter 26: Complex predication and serialization
Chapter 27: Conjugation classes
Chapter 28: Agreement morphology
Chapter 29: Suppletion
Chapter 30: Valency change and causation
Chapter 31: Reflexives and reciprocals
Chapter 32: Tense and aspect
Chapter 33: Modality and mood
Chapter 34: Negation
Chapter 35: Word order
Chapter 36: Questions
Chapter 37: Subordination
Chapter 38: Relative clauses
Chapter 39: Antipassives
Chapter 40: Morphological change
C. Semantics, pragmatics, and discourse
Chapter 41: Quantification
Chapter 42: Direction and location
Chapter 43: Kinship, marriage, and skins
Chapter 44: Toponyms
Chapter 45: Discourse and social interaction
Chapter 46: Narrative
Chapter 47: Interjections
Chapter 48: Insults and compliments
Chapter 49: Language names
Part III: Sociolinguistics and language variation
Chapter 50: The verbal arts in Indigenous Australia
Chapter 51: Sociolinguistic variation
Chapter 52: Australian Indigenous sign languages
Chapter 53: Gender-based dialects
Chapter 54: Multilingualism
Chapter 55: Code-switching
Chapter 56: Language contact
Chapter 57: Kriol
Chapter 58: Young people’s varieties
Chapter 59: Restricted respect registers and auxiliary languages
Chapter 60: Language input and child-directed speech
Part IV: Language in the community
Chapter 61: Language policy, planning, and standardization
Chapter 62: Indigenous children’s language practices in Australia
Chapter 63: Technology for Australian languages
Chapter 64: Language revival
Chapter 65: Language, land, identity, and wellbeing
Part V: Structural sketches of languages, subgroups, and families
Chapter 66: Contact language case studies
Chapter 67: The Gunwinyguan languages
Chapter 68: Anindilyakwa
Chapter 69: Languages of the Kimberley region
Chapter 70: The Maningrida languages
Chapter 71: Living languages of Victoria
Chapter 72: Lamalamic (Paman)
Chapter 73: The Bandjalangic languages and dialects
Chapter 74: Noongar
Chapter 75: The Wati (Western Desert) subgroup of Pama-Nyungan
Chapter 76: Ngumpin-Yapa languages
Chapter 77: Wajarri
Chapter 78: The revitalization of the sleeping Tasmanian Aboriginal languages palawakani
References
Language and Language Family index
Index of places mentioned
Subject index

Citation preview

the oxford guide to

AUSTRALIAN LANGUAGES

ox f or d g u ide s to t he w or l d ’s l a ng ua g e s

general editors Adam Ledgeway, University of Cambridge, and Martin Maiden, University of Oxford

advisory editors

Alexandra Y. Aikhenvald, James Cook University, Edith Aldridge, University of Washington, Stephen R. Anderson, Yale University, Bernard Comrie, University of California, Santa Barbara, Jan Terje Faarlund, University of Oslo, Alice Harris, University of Massachusetts, Amherst, Bernd Heine, University of Cologne, Paul Hopper, Carnegie-Mellon University,

Geoffrey Khan, University of Cambridge, Lutz Marten, SOAS, London, Marianne Mithun, University of California, Santa Barbara, Irina Nikolaeva, SOAS, London, Chris Reintges, CNRS, Paris, Masayoshi Shibatani, Rice University, David Willis, University of Cambridge

published The Oxford Guide to Australian Languages Edited by Claire Bowern The Oxford Guide to the Romance Languages Edited by Adam Ledgeway and Martin Maiden The Oxford Guide to the Transeurasian Languages Edited by Martine Robbeets and Alexander Savelyev The Oxford Guide to the Uralic Languages Edited by Marianne Bakró-Nagy, Johanna Laakso, and Elena Skribnik

in preparation The Oxford Guide to the Afroasiatic Languages Edited by Sabrina Bendjaballah and Chris Reintges The Oxford Guide to the Atlantic Languages of West Africa Edited by Friederike Lu¨pke The Oxford Guide to the Bantu Languages Edited by Ellen Hurst, Nancy Kula, Lutz Marten, and Jochen Zeller The Oxford Guide to the Languages of the Central Andes Edited by Matthias Urban The Oxford Guide to the Malayo-Polynesian Languages of Southeast Asia Edited by Alexander Adelaar and Antoinette Schapper The Oxford Guide to the Papuan Languages Edited by Nicholas Evans and Sebastian Fedden The Oxford Guide to the Slavonic Languages Edited by Jan Fellerer and Neil Bermel The Oxford Guide to the Tibeto-Burman Languages Edited by Kristine Hildebrandt, Yankee Modi, David Peterson, and Hiroyuki Suzuki

the oxford guide to

Australian Languages edited by

Claire Bowern

Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © editorial matter and organization Claire Bowern 2023 © the chapters their several authors 2023 The moral rights of the authors have been asserted All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2022936420 ISBN 978–0–19–882497–8 DOI: 10.1093/oso/9780198824978.001.0001 Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY Cover image: _Jamirlangu [husband and wife]. Jukuna (Mona J. Chuguna) [1933–2011], Walmajarri people, and Pijaju Peter Skipper [1929–2007]. © Mona Jukuna Chuguna and Estate of Pijaju Peter Skipper, courtesy Mangkaja Arts, image courtesy of the National Gallery of Australia, Accession Number: 2004.22 IRN: 127993 Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

We acknowledge the contributions of the First Peoples of the place now called Australia, who have never ceded ownership over this land and its languages.

Short contents Detailed contents Series preface Abbreviations and conventions The contributors Language maps Australian language families and linguistic classifications Foreword Kado Muir

xii xlvi xlvii liv lvi lxvii xcv

1. Introduction Claire Bowern

1

PART I: Background

7

2. A history of the early description of Australian languages Clara Stockigt

9

3. Documentation of Australian languages Nicholas Thieberger

23

4. Australian languages and syntactic theory Rachel Nordlinger

31

5. Australian languages and interdisciplinary approaches to the past Luisa Miceli and Claire Bowern

37

6. Nineteenth-century classifications of Australian languages Clara Stockigt

46

7. How many languages are and were spoken in Australia? Claire Bowern

56

8. Philological methods for Australian languages John Giacon and Harold Koch

65

PART II: Structures

77

A. Phonetics and phonology 9. Articulatory and acoustic phonetics Marija Tabain 10. Segment inventories Erich R. Round 11. Phonotactics Erich R. Round

81 96 106

vii

short contents 12. Morphophonology: Lenition and assimilation Erich R. Round

120

13. Nasal cluster dissimilation Erich R. Round

136

14. Lexical stress Kathleen Jepson and Thomas Ennever

145

15. Intonation Janet Fletcher

159

16. Sound change Barry Alpher

165

B. Morphosyntax

viii

17. Word classes Oliver Shoulson

187

18. The noun phrase Dana Louagie

196

19. Noun classes Amalia Skilton

205

20. Ergativity Vivien Dunn and Felicity Meakins

217

21. Semantic case Jane Simpson

226

22. Possession Maïa Ponsonnet

243

23. Demonstratives Dana Louagie

253

24. Pronouns Alice Gaby and Oliver Shoulson

268

25. Adjectives and adverbs Juhyae Kim

278

26. Complex predication and serialization David Osgarby and Claire Bowern

291

27. Conjugation classes Harold Koch

309

28. Agreement morphology Parker Brody

319

29. Suppletion Xavier Bach and Erich R. Round

328

30. Valency change and causation Stef Spronck

344

short contents 31. Reflexives and reciprocals Alice Gaby

360

32. Tense and aspect James Bednall

378

33. Modality and mood James Bednall

392

34. Negation Josh Phillips

411

35. Word order Magda Andrews-Hoke and Parker Brody

424

36. Questions Juhyae Kim and Claire Bowern

434

37. Subordination Marie-Elaine van Egmond

446

38. Relative clauses Rachel Hendery

457

39. Antipassives Jessica Denniss

468

40. Morphological change Barry Alpher and Claire Bowern

482

C. Semantics, pragmatics, and discourse 41. Quantification Margit Bowler and Ivan Kapitonov

499

42. Direction and location Dorothea Hoffmann

513

43. Kinship, marriage, and skins Patrick McConvell

520

44. Toponyms Katherine Rosenberg, Jane Simpson, and Claire Bowern

530

45. Discourse and social interaction Joe Blythe and Ilana Mushin

538

46. Narrative Francesca Merlan

548

47. Interjections Maïa Ponsonnet

564

48. Insults and compliments Michael Walsh

573

ix

short contents

x

49. Language names Katherine Rosenberg and Claire Bowern

579

PART III: Sociolinguistics and language variation

589

50. The verbal arts in Indigenous Australia Jennifer Green, Inge Kral, and Sally Treloyn

591

51. Sociolinguistic variation John Mansfield

601

52. Australian Indigenous sign languages Jennifer Green

612

53. Gender-based dialects John Bradley and Alice Gaby

628

54. Multilingualism Jill Vaughan

637

55. Code-switching Amanda Hamilton-Hollaway

645

56. Language contact Denise Angelo

656

57. Kriol Greg Dickson

667

58. Young people’s varieties Carmel O’Shannessy

680

59. Restricted respect registers and auxiliary languages Michael Walsh

689

60. Language input and child-directed speech Lucinda Davidson, Barbara Kelly, Gillian Wigglesworth, and Rachel Nordlinger

697

PART IV: Language in the community

705

61. Language policy, planning, and standardization Rob Amery

707

62. Indigenous children’s language practices in Australia Gillian Wigglesworth and Samantha Disbray

720

63. Technology for Australian languages Catherine Bow

728

64. Language revival Maryanne Gale

738

short contents 65. Language, land, identity, and wellbeing Rob Amery and Maryanne Gale

754

PART V: Structural sketches of languages, subgroups, and families

763

66. Contact language case studies Denise Angelo

765

67. The Gunwinyguan languages Nicholas Evans and Alexandra Marley

781

68. Anindilyakwa Marie-Elaine van Egmond

796

69. Languages of the Kimberley region Stef Spronck

812

70. The Maningrida languages Margaret Carew and David Felipe Guerrero Beltran

825

71. Living languages of Victoria K. Eira

845

72. Lamalamic (Paman) Jean-Christophe Verstraete

855

73. The Bandjalangic languages and dialects Margaret Sharpe

865

74. Noongar Denise Smith-Ali, Sue Hanson, George Hayden, Claire Bowern, Akshay Aitha, Lydia Ding, and Sarah Mihuc

876

75. The Wati (Western Desert) subgroup of Pama-Nyungan Sarah Babinski, Luis-Miguel Rojas-Berscia, and Claire Bowern

893

76. Ngumpin-Yapa languages Felicity Meakins, Thomas Ennever, David Osgarby, Mitch Browne, and Amanda Hamilton-Hollaway

918

77. Wajarri Doug Marmion

933

78. The revitalization of the sleeping Tasmanian Aboriginal languages palawa kani Annie Reynolds and Theresa Sainty

950

References Language and Language Family index Index of places mentioned Subject index

958 1061 1072 1074

xi

Detailed contents Series preface Abbreviations and conventions The contributors Language maps Australian language families and linguistic classifications Foreword Kado Muir 1. Introduction Claire Bowern 1.1 Introduction: the languages of Australia 1.2 Overview of the volume 1.2.1 Summary of parts 1.2.2 Coverage of topics 1.2.3 Authors’ approach 1.2.4 Supplementary materials and data archiving 1.3 Conventions and context 1.3.1 Terminological conventions 1.3.2 Referring to Australian languages 1.4 Overview of the languages of Australia 1.4.1 Language and descent groups 1.4.2 Language and European settlement 1.4.3 Language classification 1.4.4 Contemporary language use 1.5 Conclusion Acknowledgements PART I: Background 2. A history of the early description of Australian languages Clara Stockigt 2.1 Introduction 2.1.1 The first descriptive era: pre-1930 2.1.1.1 Description of non-Pama-Nyungan languages 2.1.1.2 Linguistics and ethnology 2.1.2 The second descriptive era: 1930–1960 2.1.3 The third descriptive era: 1960– 2.2 Early analyses of Pama-Nyungan languages 2.2.1 Phonology and orthography 2.2.2 Descriptive models 2.2.3 Schools of morphological description 2.3 Concluding comment

xii

xlvi xlvii liv lvi lxvii xcv 1 1 1 1 2 3 3 3 3 3 4 4 5 5 5 6 6 7 9 9 10 11 11 12 13 14 17 19 20 22

detailed contents 3. Documentation of Australian languages Nicholas Thieberger 3.1 Introduction 3.2 Why documentation? 3.3 What records of Australian languages are there and how much is there for each language? 3.4 Where to look for records of Australian languages 3.5 Accessing and presenting early sources on Australian languages 3.6 Loss of documentation 3.7 Conclusions Acknowledgements

23

4. Australian languages and syntactic theory Rachel Nordlinger 4.1 Introduction 4.2 Phrase structure and nonconfigurationality 4.3 Ergativity 4.4 Case stacking 4.5 Subordination 4.6 Conclusion Acknowledgements

31

5. Australian languages and interdisciplinary approaches to the past Luisa Miceli and Claire Bowern 5.1 Introduction 5.1.1 Overview 5.1.2 An aside: the Australian past as ‘deep history’ 5.2 Problems within linguistics 5.2.1 Missing information 5.2.2 Sound change 5.2.3 Language contact 5.2.4 Stasis 5.3 Linguistics and genetics and where they disagree: a case study of three trees 5.3.1 Domain specificity and assumptions about change 5.3.2 Are the language dates too young? 5.3.3 Are the genetic dates too old? 5.4 Conclusions

37

6. Nineteenth-century classifications of Australian languages Clara Stockigt 6.1 Introduction 6.2 The relatedness of Australian languages 6.3 Early grammatical typologies supporting the notion of relatedness 6.3.1 Dissemination of information about Australian languages 6.3.2 Development of terminology referring to linguistic entities 6.3.3 Northern and southern mainland languages 6.3.4 Subgroups of ‘southern’ languages 6.3.5 Early rankings of Australian subgroups 6.4 Concluding comment

46

23 23 25 26 28 30 30 30

31 31 32 33 34 35 36

37 37 37 38 38 38 39 40 41 41 42 44 45

46 47 47 50 51 52 54 55 55

xiii

detailed contents 7. How many languages are and were spoken in Australia? Claire Bowern 7.1 Introduction 7.2 The number ‘250’ 7.3 Calculating the number of languages 7.3.1 Sources of names 7.3.2 Variants in naming 7.3.3 Languages and dialects 7.3.4 Using phylogenetics and classifications to evaluate similarity 7.3.5 Languages with no sources 7.4 Problems with the language/variety model 7.5 Number of languages and sources of under- or over- counting 7.6 Changes in the number of languages over time 7.6.1 Pre-colonial period 7.6.2 The colonial period 7.7 Conclusion

56

8. Philological methods for Australian languages John Giacon and Harold Koch 8.1 What is philology? 8.2 Sources 8.2.1 Types of sources 8.2.2 Limitations of records and sources of errors 8.3 Practical steps in philological analysis 8.3.1 Background knowledge 8.3.2 General processes 8.3.3 Issues with wordlists 8.3.4 Phonology and the interpretation of spelling 8.3.5 Interpreting grammar 8.3.6 Interpreting texts 8.4 Philology and language revival 8.4.1 The completeness of philological descriptions 8.4.2 The products of philological analysis 8.4.3 The ‘philological spiral’ 8.5 Summary and conclusions

65

PART II: Structures

56 57 58 58 59 59 60 61 61 62 63 63 64 64

65 66 66 67 68 68 68 69 70 73 73 74 74 75 75 76 77

A. Phonetics and phonology 9. Articulatory and acoustic phonetics Marija Tabain 9.1 Introduction 9.2 Articulation of lingual consonants 9.3 A special consideration of apical consonants 9.4 Some acoustic correlates of coronal consonant production in Australian languages 9.5 Conclusion Acknowledgements

xiv

81 81 82 88 91 94 95

detailed contents 10. Segment inventories Erich R. Round 10.1 Main parameters of variation in consonant inventories 10.2 Additional consonantal segment types 10.2.1 Fricatives 10.2.2 Glides, laterals, and vibrants 10.2.3 Additional places of articulation 10.2.4 Labialized consonants 10.3 Uncommon absences from consonant inventories 10.4 Diversity in two-series stop systems 10.5 Vowels 10.6 Prospective 10.7 Sources Acknowledgements

96 96 98 98 99 99 99 100 100 102 103 103 105

11. Phonotactics Erich R. Round 11.1 Consonant phonotactics: main parameters of variation 11.1.1 Dixon’s and Hamilton’s generalizations 11.1.2 Laryngeal consonants 11.1.3 Superlaryngeal clusters across the continent 11.2 Statistical harmony of vowels in adjacent syllables 11.2.1 Statistical front–back harmony in non-low vowels 11.2.2 The *iCu constraint on non-low vowels 11.2.3 Statistical mid-vowel metaphony 11.3 Australian contour segments: at the edge of segment and sequence 11.3.1 Initial nasal+stop 11.3.2 Stop+lateral 11.3.3 Stop+nasal 11.4 Prospective 11.5 Sources Acknowledgements

106

12. Morphophonology: Lenition and assimilation Erich R. Round 12.1 Materials and methods 12.2 Lenition 12.2.1 Participation of stops in lenition alternations, by place of articulation. 12.2.2 Pairings between stops and lenis alternants 12.2.3 One-to-many and many-to-one pairings 12.2.4 Ranges of places of articulation for lenition 12.2.5 Phonological triggers 12.2.6 Morphological conditioning 12.2.7 Discussion: synchronic analysis as lenition or fortition 12.2.8 Summary 12.3 Assimilation 12.3.1 Assimilation between consonants 12.3.1.1 Hypotheses based on static phonotactics 12.3.1.2 Place assimilation 12.3.1.3 Manner assimilation 12.3.1.4 Assimilation of both place and manner

120

106 106 107 108 111 111 111 112 112 113 114 115 117 117 119

120 121 121 122 123 124 125 126 126 128 128 128 129 130 132 132

xv

detailed contents 12.3.2 Assimilation between vowels in adjacent syllables 12.3.2.1 Front–back assimilation between i and u 12.3.2.2 Assimilatory raising of a 12.3.2.3 Copying of the trigger vowel, irrespective of its quality 12.3.3 Summary and discussion 12.4 Conclusion 12.5 Language sample Acknowledgements

xvi

132 133 133 134 134 134 134 135

13. Nasal cluster dissimilation Erich R. Round 13.1 Progressive NCD nasal deletion in Gurindji 13.2 Dimensions of cross-linguistic variation 13.3 Triggers of NCD deletion 13.3.1 Heterorganic triggers 13.3.2 Nasal+nasal clusters as non-triggers 13.3.3 Morphological idiosyncrasy of triggers 13.3.4 Heteromorphemic triggers 13.4 Targets of NCD deletion 13.4.1 Restrictions on place and individual morphemes 13.4.2 Heteromorphemic targets 13.4.3 Optionality 13.5 Distant and multiple application 13.5.1 Distant application 13.5.2 Blockers 13.5.3 Chains of triggers and targets 13.6 Interaction with other processes 13.7 Summary and discussion 13.8 Language sample Acknowledgements

136

14. Lexical stress Kathleen Jepson and Thomas Ennever 14.1 Introduction 14.2 Previous surveys of Australian stress systems 14.3 Stress patterns and their variations 14.3.1 Location of stress 14.3.2 Quantity sensitivity 14.3.3 Morphological sensitivity 14.3.4 Variable systems 14.4 Lexical stress and segmental alternations 14.4.1 Lenition 14.4.2 Initial dropping 14.4.3 Contrastive vowel length restrictions 14.5 Impressionistic studies of lexical stress 14.5.1 Reported cues to stress 14.6 Acoustic studies of lexical stress 14.6.1 Fundamental frequency 14.6.2 Vowel formants and duration 14.6.3 Consonant related phenomena 14.7 Chapter summary

145

136 137 137 137 138 138 138 138 138 140 140 140 140 141 142 142 143 144 144

145 145 146 147 149 149 150 151 151 152 152 153 153 155 155 156 156 158

detailed contents 15. Intonation Janet Fletcher 15.1 Introduction 15.2 Prosodic structure and intonational categories in Australian languages 15.3 Sentence modality and grammatical structure 15.4 Alignment with grammatical constituents 15.5 Discourse and information structure 15.6 Conclusions

159

16. Sound change Barry Alpher 16.1 Introduction 16.2 Changes in consonant inventories: manners 16.2.1 Innovation of contrastive obstruent manners 16.2.1.1 Contrastive stop series possibly due to vowel-length merger 16.2.1.2 Stop manner contrasts possibly originating from loanwords 16.2.1.3 Manner contrast in non-apical stops 16.2.1.4 Manner contrast in apical stops only 16.2.1.5 A three-way obstruent manner contrast 16.2.1.6 Contrastive fortis stops conditioned by adjacent sonorants 16.2.1.7 Contrastive voiced stops originating from prestopped nasals 16.2.2 Other obstruent series: spirants and prenasalized stops 16.2.2.1 Spirants 16.2.2.2 Spirants originating from nasal-stop clusters 16.2.2.3 Prenasalized stops 16.3 Lenition conflating stops and liquids with corresponding glides 16.3.1 Lenition of initial stop 16.3.2 Lenition of medial stops 16.3.3 Lenition of liquids to glides 16.3.4 Chain shifts 16.4 Changes in vowel inventories 16.4.1 A couple of four-vowel systems 16.4.2 Development of some five-vowel systems 16.4.3 Separate development of structurally identical six-vowel systems 16.4.4 Loss and reinstatement of vowel-length contrasts 16.5 Vowel assimilation across consonants 16.5.1 Regressive 16.5.2 Progressive 16.5.3 Progressive assimilation preceding initial-dropping 16.6 Initial-dropping 16.6.1 Segmentally conditioned C-loss 16.6.2 Stress-conditioned C-Loss (Blevins 2001b: 484–5) 16.7 Alterations of consonant clusters 16.8 Loss of unstressed vowels, including rhythmic reduction 16.9 Mergers and other changes with laminals and apicals 16.10 Consonant fortitions not creating new manner series 16.10.1 Fortition of initial C 16.10.2 Fortition of intervocalic C 16.10.3 Fortition of a C adjacent to a C 16.10.4 Fortition of a word-final C 16.11 Creation of individual sounds not previously in the inventory

165

159 159 160 162 163 164

165 166 166 166 167 167 167 167 167 168 168 168 169 169 169 169 169 170 170 170 170 171 172 173 173 173 173 174 174 174 175 176 178 179 180 180 181 181 182 182

xvii

detailed contents 16.11.1 Apical stops with trilled release 16.11.2 Flapped laterals 16.11.3 Glottal closure (/ʔ/) 16.11.4 Voiced velar spirant (/ɣ/) in the Arandic languages 16.12 Conclusion Acknowledgements

182 182 183 183 183 184

B. Morphosyntax

xviii

17. Word classes Oliver Shoulson 17.1 Introduction 17.2 Major word classes 17.2.1 Verbal word classes 17.2.1.1 Matrix verbs 17.2.1.2 Auxiliaries and light verbs 17.2.1.3 Preverbs 17.2.2 Nominals 17.2.2.1 Nouns 17.2.2.2 Adjectives 17.3 Pro words 17.3.1 Pronouns 17.3.2 Pro-verbs 17.3.3 Demonstratives and articles 17.4 Minor word classes 17.4.1 Adverbs 17.4.2 Conjunctions 17.4.3 Negation particles 17.5 Changes in word class 17.5.1 Roots that belong to more than one word class 17.5.2 Derived classes 17.6 Conclusions

187

18. The noun phrase Dana Louagie 18.1 Introduction 18.2 Syntactic unithood 18.2.1 Word order in the NP 18.2.2 Case marking 18.2.3 Second position markers as diagnostics 18.2.4 Interaction of criteria: Different construals 18.2.5 Discontinuity 18.2.6 Questions 18.3 Determiners 18.3.1 Form of determiner slot 18.3.2 Elements that function as determiners 18.3.3 Optionality 18.3.4 Questions 18.4 Conclusion Acknowledgements

196

187 187 188 188 188 189 190 191 191 192 192 193 193 194 194 194 194 195 195 195 195

196 197 197 198 199 199 200 200 201 201 202 203 203 204 204

detailed contents 19. Noun classes Amalia Skilton 19.1 Introduction 19.2 Types of noun classification system 19.2.1 Noun class systems 19.2.2 Classifiers and generic-specific constructions 19.3 Noun class systems 19.3.1 Geographical and genetic distribution 19.3.2 Number of noun classes 19.3.3 Basis of noun class assignment 19.3.3.1 Semantic basis 19.3.3.1.1 Human nouns 19.3.3.1.2 Animal species 19.3.3.1.3 Inanimates 19.3.3.1.4 Identifying a default noun class 19.3.3.2 Phonological basis 19.3.4 Covert vs. overt noun class 19.3.5 Interactions between noun class and number 19.3.6 Targets of noun class agreement 19.4 Strict semantic noun class systems 19.4.1 Strict semantic noun class systems in non-Pama-Nyungan languages 19.4.2 Strict semantic noun class systems in Pama-Nyungan languages 19.5 Classifiers and generic-specific constructions compared to noun class 19.5.1 The classifier system of Murrinhpatha 19.5.2 Other intersecting systems of nominal classification 19.6 Further reading 19.7 Conclusion

205

20. Ergativity Vivien Dunn and Felicity Meakins 20.1 Introduction 20.2 Morphological ergativity 20.3 Syntactic ergativity 20.4 Split ergativity 20.5 Optional ergativity 20.6 Conclusion

217

21. Semantic case Jane Simpson 21.1 Introduction 21.1.1 Types of case 21.1.2 History of terminology 21.2 Forms 21.2.1 Allomorphy 21.2.1.1 Cardinal directions and place names 21.2.1.2 Case formatives built on other case formatives 21.2.2 Derivation 21.3 Syntactic properties 21.3.1 Position within the nominal phrase 21.3.2 Position within adpositional phrase

226

205 205 205 206 206 207 207 207 207 207 207 208 209 209 210 211 212 213 213 213 214 214 215 215 215

217 217 218 221 223 225

226 226 229 231 231 231 233 233 234 234 234

xix

detailed contents 21.3.3 Case-stacking 21.4 Meanings 21.4.1 Location in space at, to, and from 21.4.1.1 Location 21.4.1.2 Motion/direction/orientation towards 21.4.1.3 Motion from, source 21.4.2 Purpose 21.4.3 Association 21.4.3.1 Affixed to the possessor 21.4.3.2 Affixed to the possession or property 21.4.3.3 Affixed to the place 21.4.3.4 Accompaniment 21.4.3.5 Instrument 21.4.3.6 Lacking 21.4.4 Resemblance 21.5 Conclusion Acknowledgements

xx

234 235 235 236 237 237 238 238 238 239 239 240 240 240 241 242 242

22. Possession Maïa Ponsonnet 22.1 Introduction 22.1.1 Language sample 22.1.2 Definitions and labels 22.2 Proprietive and privative suffixes 22.3 The Pama-Nyungan profile 22.3.1 Dependent-marking constructions 22.3.2 The part–whole construction 22.4 Non-Pama-Nyungan trends 22.4.1 Adnominal strategies 22.4.1.1 Head-marking constructions 22.4.1.2 Dependent-marking constructions 22.4.1.3 Complex systems 22.4.1.4 Part–whole constructions and their historical traces 22.4.2 Clausal marking 22.4.3 Kin terms 22.4.4 Summary on inalienability in non-Pama-Nyungan languages 22.5 Inalienability and possession classes in Australian languages 22.5.1 Possession classes vs. noun classes 22.5.2 Trans-continental semantic patterns

243

23. Demonstratives Dana Louagie 23.1 Introduction 23.2 Morphosyntax 23.2.1 Free and bound forms 23.2.2 Distribution and categorial status 23.2.3 Inflectional possibilities 23.2.3.1 Case 23.2.3.2 Noun class 23.2.3.3 Number 23.2.3.4 Person 23.2.4 Relation to third person pronouns 23.2.5 Use in nominal expressions

253

243 243 243 244 244 245 245 247 247 247 247 248 248 248 250 250 251 251 251

253 253 253 254 256 256 257 257 258 258 259

detailed contents 23.2.5.1 Word order tendencies 23.2.5.2 Determiner slots 23.3 Semantics and use 23.3.1 Exophoric function: Spatial (and non-spatial) deixis 23.3.2 Anaphoric, cataphoric, and discourse deictic functions 23.3.3 Recognitional function 23.4 Signs of grammaticalization 23.5 Conclusion Acknowledgements

259 259 260 260 262 264 265 266 267

24. Pronouns Alice Gaby and Oliver Shoulson 24.1 Introduction 24.2 Paradigms of personal pronouns 24.2.1 Person 24.2.2 Number 24.2.3 Gender and class 24.2.4 Case 24.2.5 Tense, aspect, and mood 24.2.6 Kinship 24.3 Morphosyntax of personal pronouns 24.3.1 Bound pronouns 24.3.1.1 Verbal prefixes 24.3.1.2 Enclitic pronouns 24.3.1.3 Auxiliaries 24.3.2 Pronouns and the noun phrase 24.4 Other kinds of pronoun 24.4.1 Possessive pronouns 24.4.2 Intensifier pronouns 24.4.3 Reflexive and reciprocal pronouns 24.4.4 Demonstrative pronouns 24.4.5 Interrogative and indefinite pronouns 24.5 Pronouns in diachronic context Acknowledgements

268

25. Adjectives and adverbs Juhyae Kim 25.1 Introduction 25.2 Adjectives: A distinctive class? 25.2.1 Adjective identification criteria 25.2.1.1 Semantic criteria 25.2.1.2 Distributional criteria 25.2.1.3 Syntactic criteria 25.2.1.4 Morphological criteria 25.2.1.5 No distinction 25.2.2 Summary of adjective classifications 25.2.3 Adjective reduplication 25.2.3.1 Form 25.2.3.2 Semantic effect 25.2.4 Adjective derivations and their relation to other word classes

278

268 268 268 268 270 270 271 271 272 272 272 273 273 274 274 274 275 275 275 276 276 277

278 278 279 279 280 280 281 282 282 283 283 283 284

xxi

detailed contents 25.3 Adverbs 25.3.1 Forms of adverbs 25.3.1.1 Particles 25.3.1.2 Derivations 25.3.1.3 Word order 25.3.2 Semantics of adverbs 25.3.2.1 Locative and spatial 25.3.2.2 Temporal 25.3.2.3 Manner and degree 25.3.2.4 Miscellaneous categories 25.4 Conclusion

xxii

286 286 286 287 288 288 288 289 289 290 290

26. Complex predication and serialization David Osgarby and Claire Bowern 26.1 Introduction 26.1.1 Serial verb constructions and light verb constructions 26.1.2 Related structures 26.1.2.1 Multiclausal constructions 26.1.2.2 Peripheral modification 26.1.2.3 Auxiliaries (TAM, valency, associated motion) 26.1.2.4 Noun incorporation 26.1.2.5 Lexical compounds 26.2 Form 26.2.1 Composition 26.2.1.1 Symmetrical 26.2.1.2 Asymmetrical 26.2.1.3 Classes of major and minor members 26.2.2 Marking 26.2.2.1 Concordant marking 26.2.2.2 Single marking (major member) 26.2.2.3 Single marking (minor member) 26.2.2.4 No marking 26.2.3 Contiguity and wordhood 26.2.4 Multipartite complex predicates 26.2.4.1 Additional major member 26.2.4.2 Additional minor member 26.3 Function 26.3.1 Valency 26.3.2 Aktionsart 26.3.3 Vectorial configuration 26.3.4 Lexical semantics 26.4 Conclusion Acknowledgements

291

27. Conjugation classes Harold Koch 27.1 Introduction 27.2 Pama-Nyungan languages with robust conjugations 27.3 Languages with two classes 27.4 Languages with no classes 27.5 Languages with inadequate information 27.6 Historical-comparative discussion of Pama-Nyungan

309

291 292 294 294 294 294 295 295 296 296 296 296 296 297 298 298 299 299 300 301 301 302 302 303 304 305 306 307 308

309 309 314 314 315 316

detailed contents 28. Agreement morphology Parker Brody 28.1 Preliminaries 28.2 Anatomy of agreement 28.3 Agreement in the verbal domain 28.4 Agreement in the nominal domain 28.5 Conclusion

319

29. Suppletion Xavier Bach and Erich R. Round 29.1 Introduction, definitions 29.2 Materials and methods 29.3 Verbal suppletion 29.3.1 Tense and aspect 29.3.2 Mood (including imperatives) 29.3.3 Number 29.3.4 Other categorical types 29.3.5 Non-categorical suppletion 29.3.6 Lexical meanings 29.4 Suppletion beyond verbs 29.4.1 Pronouns, demonstratives, determiners, and directionals 29.4.2 Kin terms 29.4.3 Nominal and adjectival incorporation and compounding 29.4.4 Other nominal and adjectival roots 29.5 Conclusion

328

30. Valency change and causation Stef Spronck 30.1 Transitivity 30.1.1 Valency change 30.1.2 Causation 30.1.2.1 Internal causation 30.1.2.2 External causation 30.1.2.3 Other types of causation 30.1.3 Applicatives 30.2 Sampling 30.3 Causatives in Australian languages 30.3.1 Causative affixes 30.3.2 Periphrastic constructions 30.3.2.1 Light verbs and auxiliaries 30.3.2.2 Serial verb constructions 30.3.2.3 Indirect causation 30.3.3 Distribution of causatives 30.4 Applicatives and other valency extensions 30.4.1 Applicative functions 30.4.2 Distribution of applicatives 30.5 Anticausatives 30.6 Implications: Discussion and conclusion Acknowledgements

344

319 319 321 323 327

328 329 329 329 331 331 332 335 335 335 336 337 338 341 343

344 345 346 347 347 348 348 348 349 349 352 352 353 353 354 355 355 358 358 359 359

xxiii

detailed contents

xxiv

31. Reflexives and reciprocals Alice Gaby 31.1 Introduction 31.2 Coding strategies 31.2.1 Lexical reflexives and reciprocals 31.2.2 Transitivity frame shift 31.2.3 Combining strategies 31.2.4 Non-reflexive/reciprocal clauses 31.3 Semantics 31.3.1 Core reflexives 31.3.1.1 Partitive object 31.3.1.2 Autocausative 31.3.1.3 Oblique reflexives 31.3.2 Core reciprocals 31.3.2.1 Oblique reciprocals 31.3.2.2 Extended reciprocal senses 31.3.3 Absence/backgrounding of Actor 31.3.4 Foregrounding of Actor 31.3.5 Collective 31.3.6 Multiplicity of subevents 31.4 Conclusions Acknowledgements

360

32. Tense and aspect James Bednall 32.1 Introduction 32.2 The verbal complex 32.3 An overview of inflectional verbal tense/aspect marking 32.3.1 Inflectional tense/aspect marking in non-Pama-Nyungan languages 32.3.2 Inflectional tense/aspect marking in Pama-Nyungan languages 32.4 Selected topics in aspectuo-temporal expression 32.4.1 Temporal expression 32.4.1.1 Widespread temporal distinctions 32.4.1.2 Distinctions of remoteness 32.4.1.3 Scalar tenses 32.4.1.4 Non-verbal inflectional temporal marking 32.4.1.4.1 Case marking 32.4.1.4.2 Inflectional temporal marking on pronouns 32.4.2 Aspectual expression 32.4.2.1 Inflectional viewpoint aspect 32.4.2.2 Serialized verb constructions 32.4.2.3 Reduplication 32.4.2.4 Prosodic lengthening 32.5 Lexical, pragmatic, and discourse marking tense/aspect strategies 32.5.1 Lexical expression 32.5.2 Pragmatic and discourse-structural tense/aspect strategies 32.6 Conclusion Acknowledgements

378

360 360 361 362 363 364 365 365 365 367 368 368 369 369 370 372 373 374 376 377

378 379 381 382 384 384 384 384 385 385 385 385 387 387 387 389 389 390 390 390 390 391 391

detailed contents 33. Modality and mood James Bednall 33.1 Introduction 33.2 Modality and mood: Some preliminaries 33.3 Modal marking in Australian languages: A brief overview 33.4 Inflectional modal marking in Australian languages 33.4.1 Non-Pama-Nyungan languages 33.4.1.1 Form of inflectional modal marking in non-Pama-Nyungan languages

33.4.1.2 Irrealis and realis 33.4.1.3 Semantics of inflectional modal marking in

non-Pama-Nyungan languages 33.4.1.3.1 Open possibilities 33.4.1.3.2 Foreclosed (counterfactual) possibilities 33.4.2 Pama-Nyungan languages 33.4.2.1 Form of inflectional modal marking in Pama-Nyungan languages 33.4.2.2 Semantics of inflectional modal marking in Pama-Nyungan languages 33.4.2.2.1 Epistemic modality 33.4.2.2.2 Deontic modality 33.4.2.2.3 Dynamic modality (incl. ability modals) 33.4.2.2.4 Teleological-intentional and bouletic modality 33.4.2.2.5 Apprehensive modality 33.4.2.2.6 Foreclosed counterfactuals 33.4.2.2.7 Negation

33.5 Sentential mood 33.6 Conclusion Acknowledgements

34. Negation Josh Phillips 34.1 Introduction 34.2 ‘Standard’ (clausal) negation 34.2.1 The analytic standard negator 34.2.2 Fusion: The status of negation in the verbal paradigm 34.2.2.1 Fusion of tense and polarity 34.2.2.2 Tense marking under negation 34.2.2.3 The irrealis negator 34.2.2.4 Negation, ‘reality status’ and the verbal paradigm 34.2.3 Negation and ‘nonrealized’ moods 34.3 Negative imperatives 34.4 Nominal negation 34.4.1 Negative existentials 34.4.2 The privative 34.5 Suppletive/lexical negatives 34.6 Conclusion

392 392 392 393 395 396 396 398 399 399 404 406 406 406 406 407 408 408 409 409 409 410 410 410 411 411 411 411 413 413 413 414 415 417 417 419 420 421 422 423

xxv

detailed contents

xxvi

35. Word order Magda Andrews-Hoke and Parker Brody 35.1 Introduction 35.2 Word order at the sentence level 35.2.1 Typology of basic word orders 35.2.2 Preferred word order 35.2.3 Variability in constituent order 35.2.4 Nonconfigurationality 35.2.5 Newness and newsworthiness as determining factors on word order 35.2.6 Interrogatives 35.2.7 Adpositions 35.3 Word order in noun phrases 35.4 Conclusion

424

36. Questions Juhyae Kim and Claire Bowern 36.1 Introduction 36.2 Question strategies 36.2.1 Intonation in questions 36.2.2 Interrogative words 36.2.2.1 Indefinites and interrogative pronouns 36.2.2.2 Interrogative verbs 36.2.3 Clitics and particles 36.3 Syntax of questions 36.3.1 Yes/no polar questions 36.3.2 Content questions 36.3.3 Tag questions 36.3.4 Embedded questions 36.3.5 Sign language questions 36.4 Semantics and pragmatics of questions 36.4.1 Polarity in yes/no responses 36.4.2 Pragmatics of question words 36.4.3 Rhetorical questions 36.5 Conclusions

434

37. Subordination Marie-Elaine van Egmond 37.1 Introduction 37.2 Hale’s (1976a) ‘adjoined relative clause’ and Nordlinger’s (2006b) reply 37.3 Non-finite subordination 37.4 Switch-reference in non-finite clauses 37.5 Finite subordination 37.6 Complementizer case 37.7 Subordination and polysynthesis 37.8 Summary Acknowledgements

446

38. Relative clauses Rachel Hendery 38.1 Introduction

457

424 424 424 425 427 428 430 430 431 432 433

434 434 435 436 436 438 438 440 440 440 442 443 443 444 444 444 445 445

446 448 449 450 451 452 453 455 456

457

detailed contents 38.2 38.3 38.4 38.5 38.6 38.7 38.8 38.9

Defining relative clauses Relative clauses in typological perspective Adjoined relative clauses Nominalized and participial relative clauses Relative clause position and word order Relative clause markers Relative clauses In Kriol Conclusions

458 459 459 462 463 465 466 466

39. Antipassives Jessica Denniss 39.1 Introduction 39.2 Morphosyntax 39.2.1 Case marking and agreement 39.2.2 Verb marking 39.3 Semantic functions 39.3.1 Aspectual readings 39.3.2 Modal readings 39.3.3 Lexical shifts 39.3.4 Non-individuated objects 39.4 Syntactic functions 39.5 Grammaticalization patterns 39.6 Other semi-transitive constructions 39.7 Conclusion Acknowledgements

468

40. Morphological change Barry Alpher and Claire Bowern 40.1 Introduction 40.2 Changes involving clitics 40.2.1 Change in morphological marking and categories 40.2.2 Distribution of dative clitics in Eastern Daly 40.3 Change in the semantic organization of paradigms 40.3.1 Functional change with no linear change 40.3.2 Adaptation of morphemes from other paradigms 40.3.3 Introduction of a person-number form from one paradigm where it

482

is regular into another where it creates an irregularity

40.4 40.5 40.6

40.7 40.8 40.9

40.3.4 Reshaping of paradigms: Loss of agreement Nominal tense and aspect in Pama-Nyungan languages 40.4.1 Tense marking on pronouns 40.4.2 Development of tense-marking by ablaut of the root vowel Initial consonant mutation in Iwaidja Reanalysis 40.6.1 Reanalysis of a monomorphemic form as root plus derivational suffix 40.6.2 Reanalysis of an uninflecting form as an inflecting one 40.6.3 Reanalysis of affixes and the RR conjugation ‘Lost Wax’, a sudden morpheme-replacement process Morphemic constituent order Conclusion

468 469 469 469 470 471 473 474 475 477 478 478 481 481

482 482 482 483 484 484 485 485 486 486 486 487 488 489 489 489 490 492 494 495

xxvii

detailed contents C. Semantics, pragmatics, and discourse

xxviii

41. Quantification Margit Bowler and Ivan Kapitonov 41.1 Introduction 41.2 General morphosyntactic properties of quantificational expressions 41.3 Semantic findings 41.3.1 Expressing ‘many’/‘much’ 41.3.2 Expressing ‘all’/‘every’ 41.3.3 Expressing ‘several’/‘a small amount’ 41.3.4 Expressing partitive ‘some’ 41.3.5 Constituent (nominal) negation 41.3.6 Indefinite pronouns 41.3.7 Temporal quantifiers 41.3.8 Expressing ‘how many’/‘how much’ 41.3.9 Quantifier interaction 41.3.10 Lexical item for ‘to count’ 41.4 Conclusion and future directions Acknowledgements

499

42. Direction and location Dorothea Hoffmann 42.1 Introduction 42.2 Systems of Spatial Reference: Location 42.2.1 Topological relations 42.2.2 Frames of Reference 42.2.3 The role of toponyms 42.2.4 Spatial deixis 42.3 Motion in Australian languages 42.3.1 Lexicalization patterns: Verb-, satellite, and equipollent-framing 42.3.2 Spatial language in Dreamtime narratives 42.4 Multimodal representations of space 42.5 Conclusions Acknowledgements

513

43. Kinship, marriage, and skins Patrick McConvell 43.1 Introduction 43.2 Kinship systems of Australia 43.2.1 Classificatory systems 43.2.2 Radcliffe-Brown’s typology and ‘Kariera’ 43.2.3 Other system types: Aranda, asymmetrical, Aluridj 43.3 Marriage 43.4 Kinship in context 43.5 Rare kinship phenomena in Australia 43.5.1 Trirelational terminology 43.5.2 Kinship marked pronouns 43.6 ‘Skins’ (social categories) 43.6.1 Moieties 43.6.1.1 Descent moieties

520

499 499 500 500 501 503 504 505 506 508 509 510 511 511 512

513 513 513 514 515 515 516 516 517 517 518 519

520 520 520 520 521 521 523 523 523 524 524 524 524

detailed contents 43.6.1.2 Non-descent Moieties 43.6.1.3 Moiety Names

43.6.2 Sections 43.6.2.1 Structure of section systems 43.6.2.2 The Kariera (western) section system 43.6.2.3 Function of sections 43.6.3 Subsections 43.6.3.1 Structure of subsections 43.6.3.2 The function of subsections 43.6.4 The distribution of section and subsection systems 43.7 Prehistory of kinship in Australia 43.8 Prehistory of ‘skins’ in Australia 43.9 The significance of Australian kinship and skins in anthropology

525 525 525 525 525 525 526 526 526 527 527 528 529

44. Toponyms Katherine Rosenberg, Jane Simpson, and Claire Bowern 44.1 Introduction 44.2 Semantic and historical aspects of toponyms 44.2.1 Meaning 44.2.2 Types of features that are named 44.3 Structure of toponyms 44.3.1 Syntax 44.3.1.1 Bare nouns 44.3.1.2 Compounds 44.3.1.3 Relative clauses 44.3.1.4 Sentential names 44.3.1.5 Toponym-specific syntax 44.3.2 Morphology 44.3.2.1 Locatives 44.3.2.2 Comitatives and proprietives 44.3.2.3 Toponym-specific morphology 44.3.3 Transparency 44.4 Further discussion 44.4.1 Opacity and interpretation of place name meaning/etymology 44.4.2 Multiple names 44.5 Mapping place names 44.6 Aboriginal place names in Australian English 44.7 Conclusion

530

45. Discourse and social interaction Joe Blythe and Ilana Mushin 45.1 Introduction 45.2 Discourse analysis in grammatical description 45.3 Social interaction 45.3.1 Intercultural communication and ethnomethodology 45.3.2 Turn-taking and action sequencing 45.3.3 Grammar in interaction 45.4 Conclusion Acknowledgements

538

530 530 531 531 532 532 532 532 533 533 533 533 533 534 534 534 534 534 535 535 536 537

538 538 540 540 541 546 547 547

xxix

detailed contents 46. Narrative Francesca Merlan 46.1 Narrative and time 46.1.1 Australianist views of narrative 46.1.2 Indigenous-language scope of this chapter 46.2 Narrative vocabularies 46.3 The ‘emplaced story’ 46.4 Multimodality 46.5 Spacetime of Indigenous narrative 46.6 Character/s 46.7 Narrative moral content 46.8 Narrative means 46.9 Conclusions and prospects Acknowledgements

548

47. Interjections Maïa Ponsonnet 47.1 Introduction 47.1.1 Identifying interjections 47.1.2 Classifying interjections 47.1.3 Results 47.2 Classes of interjections 47.2.1 Conventional classes: Conative, phatic, and expressive interjections 47.2.2 Additional classes: Constative and social interjections 47.3 Conative interjections 47.3.1 Attention seekers 47.3.2 Frequent conative interjections 47.3.3 Others 47.4 Phatic interjections 47.4.1 ‘Yes’ and ‘no’ 47.4.2 Agreeing and seeking information 47.4.3 Epistemic interjections 47.4.4 Managing communication 47.5 Expressive interjections 47.5.1 Generic exclamations 47.5.2 Valence-oriented generic interjections (dissatisfaction and

564

satisfaction)

47.5.3 Specific states 47.5.3.1 Surprise and pain 47.5.3.2 Compassion 47.5.4 Marginal categories 47.5.5 Primary and secondary emotions 47.6 Conclusion 48. Insults and compliments Michael Walsh 48.1 Preamble

xxx

548 548 550 551 552 556 557 558 560 562 563 563

564 564 564 565 566 566 566 567 567 567 568 568 568 569 569 569 569 569 570 570 570 571 571 572 572 573 573

detailed contents 48.2 Mock insults and the joking relationship 48.3 Insults more generally 48.4 Nicknames 48.5 Compliments 48.6 Post-classical 48.7 Harvesting material from early sources 48.8 Summary Acknowledgements

574 574 575 576 577 578 578 578

49. Language names Katherine Rosenberg and Claire Bowern 49.1 Introduction 49.1.1 Language names 49.1.2 Data collection 49.2 General principles concerning language names 49.2.1 Wohlgemuth (2015) 49.2.2 McConvell (2006b) 49.2.3 Sutton (1979) 49.3 Types of language names 49.3.1 Hierarchical classification 49.3.2 Feature exclusivity 49.3.3 Non-exhaustivity 49.4 Naming conventions across Australian languages 49.4.1 Overall findings 49.4.2 Overlapping categories 49.4.3 Classifying by subfamily 49.5 Potential areal influences 49.6 Conclusion

579

PART III: Sociolinguistics and language variation

589

50. The verbal arts in Indigenous Australia Jennifer Green, Inge Kral, and Sally Treloyn 50.1 Introduction: The scope of verbal arts 50.2 Narrative practices 50.3 Children’s games 50.4 Speech play 50.5 Song 50.5.1 Early documentations of songs 50.5.2 Other song collections 50.5.3 Types of songs 50.5.4 Formal features of Indigenous songs in Australia 50.5.5 Songs for children 50.5.6 Sung and spoken language 50.5.7 Contemporary song and poetry 50.6 Conclusions Acknowledgements

591

579 579 580 581 581 581 582 582 582 585 585 585 586 586 586 587 588

591 591 593 594 595 595 595 596 597 597 598 598 600 600

xxxi

detailed contents

xxxii

51. Sociolinguistic variation John Mansfield 51.1 Introduction 51.2 Sociolinguistic variation in traditional society 51.2.1 Speech community and dialect 51.2.2 Individuals and identity 51.2.3 Kin relations and seniority 51.3 Sociolinguistic variation after town settlement 51.3.1 Yolngu 51.3.2 Western Desert 51.3.3 Murrinhpatha 51.4 Discussion

601

52. Australian Indigenous sign languages Jennifer Green 52.1 Introduction 52.1.1 Early documentations of sign 52.1.2 Renewed interest in Australian Indigenous sign 52.2 Contexts of Indigenous sign language use 52.3 Features of signs 52.3.1 Overview 52.3.2 Handshape 52.3.3 Sign lexicons and sign polysemy 52.4 Some examples 52.4.1 Signs for kin 52.4.2 New signs 52.4.3 Multi-sign utterances 52.5 Other features of signing practices 52.5.1 Relationships to speech 52.5.2 Signs and marking respect 52.5.3 Mixing and matching 52.6 Community sign resources 52.7 Discussion Acknowledgements

612

53. Gender-based dialects John Bradley and Alice Gaby 53.1 Introduction 53.2 Overview of Yanyuwa noun classes 53.3 Patterns of isomorphism 53.3.1 ♂ male/masculine = ♀ male 53.3.2 ♂ male/masculine = ♀ masculine 53.3.3 ♂ male/masculine ≠ ♀ male or masculine 53.3.4 ♀ male = ♀ masculine 53.3.4.1 ♂ male/masculine = ♀ male/masculine 53.3.4.2 ♂ male/masculine ≠ ♀ male/masculine 53.3.5 ♂ male ≠ ♂ masculine 53.3.5.1 ♂ male ≠ ♂ masculine ≠ ♀ male ≠ ♀ masculine 53.3.5.2 ♂ male ≠ ♀ male ≠ ♀ masculine = ♂ masculine 53.4 Stylistic differences

628

601 602 602 604 605 606 606 608 608 610

612 612 613 613 616 616 617 618 619 620 620 620 622 622 623 623 624 625 627

628 629 631 631 632 632 632 633 633 633 633 634 634

detailed contents 53.5 Discussion Acknowledgements

634 636

54. Multilingualism Jill Vaughan 54.1 Introduction 54.2 Work on multilingualism in Aboriginal Australia 54.3 Multilingualism in historical perspective 54.3.1 Pre-colonial language repertoires 54.3.2 Language, land, and kinship 54.3.3 The social lives of languages 54.3.4 Receptive multilingualism 54.4 Contemporary multilingualism 54.4.1 Contemporary language repertoires and changing language ecologies 54.4.2 Urbanization, mobility, and modernity 54.5 Conclusions

637

55. Code-switching Amanda Hamilton-Hollaway 55.1 Introduction 55.2 Defining code-switching 55.3 Switching multi-word phrases and clauses 55.3.1 Switching for social meaning 55.3.2 Switching for discourse structure 55.4 Switching single words and morphemes 55.4.1 Switching at the syntactic periphery 55.4.2 Switching within the clausal core 55.5 Conclusion

645

56. Language contact Denise Angelo 56.1 Overview 56.1.1 Australian contact languages old and new 56.1.2 Previous research 56.1.3 Notes on terminology 56.2 Historical background to Australian contact languages 56.2.1 Pastoral industries 56.2.2 Maritime industries 56.2.3 Other enterprises 56.2.4 Towns, settlements, and missions 56.3 Contemporary contexts of contact languages 56.3.1 Indigenous contact language ecologies and multilingualism 56.3.2 Speaker views of contact languages 56.3.3 Sporadic inclusion in policy 56.3.4 Mistaken identity: English-lexified contact languages and Standard

656

English

56.3.5 Employment, education, and training 56.3.6 Media and literature 56.4 Englishes spoken by Aboriginal and Torres Strait Islander peoples 56.4.1 Aboriginal English(es) and Torres Strait English(es) 56.4.1.1 Disparate definitions of Aboriginal English in research

637 637 639 639 639 641 641 642 642 643 644

645 645 646 647 649 650 650 651 655

656 656 656 657 657 657 658 658 659 660 660 660 660 661 661 662 663 663 663

xxxiii

detailed contents 56.4.1.2 Local vs. generic naming 56.4.1.3 Practical and research issues

664 664

dialect

665 665 666

56.4.2 Standard Australian English learned as an additional language or 56.4.2.1 The place of English in Indigenous multilingualism 56.5 Conclusion 57. Kriol Greg Dickson 57.1 Introduction 57.1.1 Kriol, its origins, emergence, and current distribution 57.1.2 Kriol, its speakers and ethnolinguistic vitality 57.2 Genetic affiliations 57.2.1 Relationship to other contact languages 57.2.2 Relationship to English 57.2.3 Relationship to traditional Australian languages (substrate languages)

57.3 Structural aspects 57.3.1 Phonology 57.3.2 Morphology and syntax 57.3.3 Lexicon and semantics 57.3.4 Pragmatics and discourse 57.3.5 Variation and dialects 57.4 Concluding remarks Acknowledgements

xxxiv

667 667 667 668 670 670 670 670 671 671 673 675 677 678 678 679

58. Young people’s varieties Carmel O’Shannessy 58.1 Introduction 58.2 Contexts of continuity with change 58.2.1 Pitjantjatjara 58.2.2 Bininj-Kunwok 58.2.3 Murrinhpatha 58.3 Emergence of new languages and varieties 58.3.1 Yolngu Matha 58.3.2 Mixed languages 58.3.3 Tiwi 58.4 Contexts of language obsolescence 58.5 Types and paths of change 58.6 Attitudes to change 58.7 Summary Acknowledgements

680

59. Restricted respect registers and auxiliary languages Michael Walsh 59.1 Introduction 59.2 Avoidance/‘mother-in-law languages’ 59.3 Initiate styles 59.4 Bereavement styles 59.5 Joking styles 59.6 Teenagers’ language

689

680 681 681 682 682 683 683 683 685 686 687 688 688 688

689 689 693 694 694 694

detailed contents 59.7 Baby talk 59.8 Conclusion Appendix I: Warlpiri avoidance registers (Laughren 2001: 223)

695 695 696

60. Language input and child directed-speech Lucinda Davidson, Barbara Kelly, Gillian Wigglesworth, and Rachel Nordlinger 60.1 Introduction 60.2 A brief overview of CDS study in Australian languages 60.3 CDS as a specialized register 60.3.1 Phonological features 60.3.2 Prosodic features 60.3.3 Lexical features 60.3.4 Semantic features 60.3.5 Morphosyntactic features 60.3.6 Caregiver ideas around CDS registers 60.4 Verbal routines and socialization practices 60.5 Situating CDS in Australian languages cross-linguistically 60.6 Conclusion Acknowledgements

697

PART IV: Language in the community

705

61. Language policy, planning, and standardization Rob Amery 61.1 Introduction 61.2 The language policy framework 61.2.1 National Indigenous Languages Policy 2009 61.2.2 Funding implications 61.2.3 State-based language policy 61.2.4 Indigenous language laws 61.2.5 Language planning and language revival 61.3 Language planning perspectives 61.3.1 Status planning 61.3.2 Place names 61.3.2.1 Dual naming 61.3.3 Corpus planning 61.3.3.1 Lexical expansion and new terminology 61.3.4 Coordination of language work 61.3.4.1 Language centres 61.3.5 Language-in-education planning 61.4 Kaurna case study 61.4.1 Kaurna status planning 61.4.1.1 Kaurna place names 61.4.1.2 Kaurna Language Week, 2014 61.4.2 Corpus planning 61.4.3 Education planning 61.4.4 Standardization of Kaurna 61.4.5 A Kaurna language planning authority—KWP and KWK 61.5 Conclusion

707

697 697 698 699 699 700 700 700 700 701 702 703 703

707 707 708 708 709 709 710 710 710 711 711 711 711 713 713 713 714 714 714 715 715 715 717 718 718

xxxv

detailed contents 62. Indigenous children’s language practices in Australia Gillian Wigglesworth and Samantha Disbray 62.1 Introduction 62.2 Language socialization studies 62.3 First language acquisition studies 62.4 Multigenerational studies of language 62.5 Summary and conclusion Acknowledgements

720

63. Technology for Australian languages Catherine Bow 63.1 Introduction 63.2 Language documentation practices 63.2.1 Tools for language documentation 63.2.2 Access to language data 63.2.3 Re-presentation of curated data 63.2.4 Dictionaries 63.3 Language in pedagogical practices 63.3.1 Online language teaching 63.3.2 Cross-cultural communication 63.4 Language in identity practices 63.4.1 Authority 63.4.2 Recognition 63.4.3 Cultural continuity 63.5 Challenges and opportunities 63.5.1 Text and literacy 63.5.2 Looking ahead 63.6 Conclusion Acknowledgements

728

64. Language revival Maryanne Gale 64.1 Introduction 64.2 Definitions 64.3 A brief history 64.4 Funds for language revival 64.5 Overview of revival programmes 64.5.1 New South Wales 64.5.1.1 Miromaa Aboriginal Language and Technology Centre 64.5.1.2 Muurrbay Cooperative and the Many River Aboriginal

738

Language Centre

64.5.2 Victoria 64.5.3 South Australia 64.5.3.1 Far West Languages Centre 64.5.3.2 Mobile Language Team (MLT) 64.5.4 Western Australia 64.5.5 Queensland 64.5.6 Tasmania 64.5.7 Australian Capital Territory 64.5.8 Northern Territory 64.5.9 National organizations and projects

xxxvi

720 721 722 725 726 727

728 729 729 729 730 731 731 731 732 732 732 733 733 734 735 735 736 737

738 738 739 741 742 743 743 743 743 744 744 744 745 745 746 746 746 748

detailed contents 64.5.9.1 Resource Network for Linguistic Diversity (RNLD) – now Living Languages

748

Studies (AIATSIS)

748 748 748 749 750

64.5.9.2 Australian Institute of Aboriginal and Torres Strait Islander 64.5.9.3 First Languages Australia 64.6 Approaches and methods of revival 64.7 Ngarrindjeri case study 64.8 Conclusion Appendix: Community language programmes which won funding from Indigenous Languages & The Arts (ILA) from 2015 onwards New South Wales Victoria South Australia Western Australia Queensland Northern Territory National programmes

750 751 751 751 751 752 752 753

65. Language, land, identity, and well-being Rob Amery and Mary-anne Gale 65.1 Introduction 65.2 Language, land, and identity 65.3 Defining the concept of ‘wellbeing’ 65.4 Closing the gap on Indigenous disadvantage 65.5 The health and well-being of first language speakers of Indigenous languages 65.6 What promotes wellbeing? 65.7 Health, wellbeing, and language revival 65.8 Examples of Kaurna language revival aiding wellbeing 65.9 Conclusion

754

PART V: Structural sketches of languages, subgroups, and families

763

66. Contact language case studies Denise Angelo 66.1 Australia’s shifting langscapes 66.2 North-eastern affiliated creoles—a language contact and diffusion zone: Yumplatok, Cape York Creole, and Lockhart River Creole 66.2.1 Language contact and contact languages in the north-east 66.2.2 Yumplatok (Broken, Torres Strait Creole) 66.2.3 Cape York Creole 66.2.4 Lockhart River Creole 66.2.5 Language features of the north-eastern creoles 66.3 Superdiverse Indigenous settlements—multiple languages in contact over time: Woorie Talk and Yarrie Lingo 66.3.1 Language contact in the Superdiverse Indigenous Settlements in

765

Queensland 66.3.1.1 Early investigations 66.3.1.2 Re-positioning as contact languages: Reasons and methods 66.3.2 Woorie Talk

754 755 756 757 758 759 760 761 762

765 765 765 766 766 767 767 769 769 769 770 770

xxxvii

detailed contents 66.3.3 Yarrie Lingo 66.4 Mixed languages—language contact with input from one traditional language: Light Warlpiri, Gurindji Kriol, and Modern Tiwi 66.4.1 Mixed languages and their language ecologies 66.4.2 Light Warlpiri and Gurindji Kriol 66.4.3 Modern Tiwi 66.4.4 Novel developments not replicas 66.5 The Kriol sphere of influence: Language contact on the periphery of the Kriol Sprachraum 66.5.1 Examining Kriol influence 66.5.1.1 The unclear extent of the Kriol Sprachraum 66.5.1.2 Nomenclature 66.5.2 To the north 66.5.3 To the south 66.5.3.1 Southern Barkly contact languages 66.5.3.1.1 Wumpurrarni English: A style drawing on multilingual resources

66.5.3.1.2 Alyawarr English: Shifting and blending

66.5.3.2 In the centre 66.5.4 To the east 66.5.4.1 Mornington Island in the Gulf of Carpentaria 66.5.4.2 Far western Queensland 66.5.4.3 Western Cape York: Kowanyama 66.5.5 To the west: Northern Western Australia Acknowledgements

xxxviii

771 772 772 773 773 774 775 775 775 775 775 776 776 776 777 777 778 778 779 779 780 780

67. The Gunwinyguan languages Nicholas Evans and Alexandra Marley 67.1 Introduction 67.2 Classification 67.3 Typological features 67.3.1 Phonology 67.3.1.1 Phonological features 67.3.1.2 Syllable structure 67.3.2 Polysynthetic verbal morphology 67.3.3 Genders 67.3.4 Case 67.3.5 Pronominal categories 67.4 Outstanding questions Acknowledgements

781

68. Anindilyakwa Marie-Elaine van Egmond 68.1 Introduction 68.2 Phonology 68.2.1 Phonological changes 68.2.1.1 Sound changes shared with Wubuy 68.2.1.2 Sound changes unique to Anindilyakwa

796

781 783 784 784 784 784 785 787 788 789 794 795

796 797 798 798 799

detailed contents 68.3 Noun classification systems 68.3.1 Noun classes and genders 68.3.2 Incorporated body parts and classifiers 68.3.2.1 Suppletion 68.3.2.2 Polysemy 68.4 Nominal derivation 68.4.1 Inalienable possession 68.4.2 Alienable possession 68.5 Subordination strategies 68.5.1 Nominalization 68.5.2 Complementizer case 68.6 Conclusions Acknowledgements

803 803 805 805 805 806 806 807 808 808 809 810 811

69. Languages of the Kimberley region Stef Spronck 69.1 History and current status of languages of the Kimberley region 69.1.1 First contacts 69.1.2 Missions, mines, cattle stations, and boarding schools 69.1.3 Language vitality 69.1.4 Reclamation 69.2 Overview of previous research 69.2.1 Linguistic research 69.2.2 Archival materials 69.3 Classification 69.3.1 Internal classification 69.3.1.1 Introduction 69.3.1.2 Nyulnyulan 69.3.1.3 Worrorran 69.3.1.4 Bunuban 69.3.1.5 Jarrakan family 69.3.1.6 Pama-Nyungan 69.3.2 Language contact in the Kimberley 69.4 Structural features 69.4.1 Verbal morphology 69.4.2 Nominal morphology 69.4.2.1 Syntactic features 69.4.2.2 Speech styles and avoidance languages 69.5 Open questions Acknowledgements

812

70. The Maningrida languages Margaret Carew and David Felipe Guerrero Beltran 70.1 Introduction 70.2 Genetic affiliation and dialectology 70.2.1 Classification 70.2.2 Lects 70.3 Phonology 70.3.1 Phoneme inventory 70.3.2 Phonotactics 70.4 Nominals

825

812 812 813 813 814 815 815 816 817 817 817 817 817 818 818 818 818 820 820 821 823 823 824 824

825 825 825 827 827 827 828 829

xxxix

detailed contents

70.5

70.6

70.7 70.8

xl

70.4.1 Nominal words 70.4.1.1 Simple nouns 70.4.1.2 Part classifiers 70.4.1.3 Prefixed nominals 70.4.2 Possession classes 70.4.3 Nominal classification and gender 70.4.4 Case-marking 70.4.4.1 Local case 70.4.4.2 Direct case 70.4.4.3 Oblique and possessive 70.4.5 Noun phrases and nominal groups Simple predicates 70.5.1 Argument-indices 70.5.1.1 Person and number 70.5.1.2 Person and clusivity 70.5.2 Tense, aspect, mood, and modality 70.5.3 Nominal predicates 70.5.4 Verbal predicates Complex predicates 70.6.1 Serial verb constructions 70.6.2 Light verb constructions 70.6.3 Multiclausal constructions Ethnolinguistic vitality Conclusions

829 829 829 830 832 833 835 835 836 836 837 837 838 838 839 840 840 841 841 842 842 843 844 844

71. Living languages of Victoria K. Eira 71.1 Introduction 71.2 Baselines 71.2.1 Language families 71.2.2 Language sources 71.2.3 Typological comments 71.3 Language in Victoria in the present 71.3.1 Language context 71.3.2 Language revival activities 71.3.3 Systems and structures supporting language 71.3.4 Language characteristics 71.4 Closing remarks

845

72. Lamalamic (Paman) Jean-Christophe Verstraete 72.1 Introduction 72.2 Phonology 72.2.1 Root structure 72.2.2 Phoneme inventories 72.2.3 Phonotactics 72.3 Morphology 72.3.1 Morphophonological alternations 72.3.2 Inflectional morphology 72.4 Syntax Acknowledgements

855

845 847 847 848 849 851 851 851 852 853 854

855 856 856 857 860 861 861 862 863 864

detailed contents 73. The Bandjalangic languages and dialects Margaret Sharpe 73.1 Name and location of the language 73.1.1 The BNJ origin story 73.1.2 Social organization 73.1.3 Initiation marking and the Huchet line 73.2 Phonology 73.3 Verbs 73.3.1 ‘Old’ verbs 73.3.2 Aspect in BNJ verbs 73.3.3 The verb forms and suffixes 73.3.3.1 First order suffix -ba 73.3.3.2 Second order suffix -ndi 73.3.3.3 Third order: The reflexive/reciprocal and antipassive suffix 73.3.3.4 73.3.3.5 73.3.3.6 73.3.3.7

-li Fourth order past action -dja Fourth order -ma ‘causative’ Fourth order -wa ‘repetitive’ Fifth order suffixes: The aspect suffixes 73.3.3.7.1 -hn ‘imperfect’ 73.3.3.7.2 -hny ‘imminent’ or ‘potential’ 73.3.3.7.3 -h ‘imperative’ 73.3.3.7.4 -hla ‘progressive’ 73.3.3.7.5 -nyun ‘synchronous action’ 73.3.3.7.6 -nah/-hna ‘antechronous action’ 73.3.3.7.7 -luhr/-lugu ‘progressive in the past’ 73.3.3.7.8 -De ‘potential’ 73.3.3.7.9 -yan ‘progressive participle’ 73.3.3.7.10 -n ‘permissive’ 73.3.3.7.11 -niban 73.3.3.7.12 -ni ‘perfect, completive’

73.4 Other grammatical points 73.4.1 Pronouns 73.4.2 The demonstrative distinctions 73.4.3 Tense marked noun suffixes 73.4.4 Gender in BNJ

74. Noongar Denise Smith-Ali, Sue Hanson, George Hayden, Claire Bowern, Akshay Aitha, Lydia Ding, and Sarah Mihuc 74.1 Introduction 74.2 Varieties of Noongar 74.2.1 Previous work 74.2.2 Three dialects? 74.3 Phonetics and phonology 74.3.1 Consonant inventory 74.3.2 Voicing 74.3.3 Places of articulation 74.3.4 Vowel inventory 74.3.5 Vowel length 74.3.6 Centralization (or not) under stress 74.3.7 Orthography 74.3.8 Phonotactics

865 865 865 866 866 867 867 868 868 870 870 870 870 871 871 871 872 872 872 872 872 872 873 873 873 873 873 873 873 874 874 874 874 875 876 876 876 876 877 878 878 879 879 879 879 880 880 880

xli

detailed contents 74.3.9 Stress 74.4 Word classes 74.4.1 Nominals 74.4.1.1 Nouns 74.4.1.2 Adjectives 74.4.1.3 Pronouns 74.4.1.4 Interrogative pronouns 74.4.1.5 Numerals 74.4.2 Demonstratives 74.4.3 Verbs 74.4.4 Adverbs 74.4.5 Interjections 74.5 Nominal morphology 74.5.1 Core case marking 74.5.1.1 Nominative case 74.5.1.2 Accusative case 74.5.1.3 Ergative–absolutive marking on first person pronouns 74.5.2 Oblique case marking 74.5.3 Variable application of case marking 74.5.4 Differential object marking 74.6 Verbal morphology 74.6.1 Aspect 74.6.2 Mood 74.7 Clause structure 74.7.1 Attested word orders 74.7.2 Factors influencing word order 74.7.3 Copular clauses 74.7.3.1 Predicate nominals 74.7.3.2 Locatives and existentials 74.7.3.3 Possessives 74.7.4 Negation 74.7.4.1 Non-imperative clausal negation 74.7.4.2 Imperative clausal negation 74.7.4.3 Privative constituent negation 74.7.4.4 Double negation 74.7.5 Coordination 74.8 Clause combinations 74.8.1 Subordination 74.8.2 Coordination 75. The Wati (Western Desert) subgroup of Pama-Nyungan Sarah Babinski, Luis-Miguel Rojas-Berscia, and Claire Bowern 75.1 Introduction 75.2 Western Desert varieties 75.2.1 Previous discussions 75.2.2 Speaker awareness of dialect differences 75.3 Phonology 75.3.1 Phonemic inventory 75.3.2 Phonotactics 75.3.3 -pa suffix usage across varieties

xlii

881 881 881 881 881 882 882 882 883 883 883 884 884 884 884 885 885 885 886 887 887 887 888 888 888 889 889 889 890 890 890 890 891 891 891 891 892 892 892 893 893 893 893 895 895 895 896 896

detailed contents 75.4 Sound change 75.5 Case morphology 75.5.1 Ergative 75.5.2 Absolutive 75.5.3 Dative, purposive, and genitive 75.5.4 Locative 75.5.5 Ablative 75.5.6 Allative 75.6 Pronouns 75.6.1 Free pronouns 75.6.2 Bound pronouns 75.7 Verbs 75.7.1 Verb classes 75.7.2 Imperative 75.7.3 Present/progressive 75.7.4 Past perfective/completive 75.7.5 Habitual/characteristic 75.7.6 Future 75.8 Conclusions

897 899 899 900 902 902 904 905 907 907 909 912 912 913 913 915 915 916 916

76. Ngumpin-Yapa languages Felicity Meakins, Thomas Ennever, David Osgarby, Mitch Browne, and Amanda Hamilton-Hollaway 76.1 Introduction 76.2 Genetic affiliation 76.3 Structural properties 76.3.1 Phonology 76.3.1.1 Phoneme inventory 76.3.1.2 Phonotactics 76.3.1.3 Nasal cluster dissimilation 76.3.2 Argument structure 76.3.2.1 Case marking 76.3.2.2 Bound pronouns 76.3.3 Complex predicates 76.4 Interesting questions 76.4.1 Nonconfigurationality 76.4.2 Bound pronoun clitic placement 76.4.3 Language contact, old and new 76.5 Conclusion Acknowledgements

918

77. Wajarri Doug Marmion 77.1 Introduction and overview 77.1.1 Background: The language and its speakers 77.1.2 Linguistic type 77.2 Phonology 77.2.1 Consonant inventory 77.2.2 Vowel inventory 77.2.3 Phonotactics 77.2.4 Orthographic conventions

933

918 919 920 920 920 921 922 922 923 924 926 927 927 929 930 932 932

933 933 933 933 933 934 934 935

xliii

detailed contents 77.3 Nominal morphology 77.3.1 Overview 77.3.2 Nominal suffixes 77.3.2.1 Ergative 77.3.2.2 Instrumental usage of ergative 77.3.2.3 Nominative 77.3.2.4 Accusative 77.3.2.5 Purposive/Genitive 77.3.2.6 Locative 77.3.2.7 Allative 77.3.2.8 Ablative 77.3.2.9 Proprietive -nyuwa 77.3.2.10 Privative -gurru 77.3.2.11 Semblative -nguny 77.3.2.12 Resultative -ganu 77.3.2.13 Associative -manu 77.3.2.14 Side -gardi 77.3.2.15 Habitat -jirri ~-jiyi 77.3.2.16 Disapproving -wily 77.3.2.17 Pair of kin -yarra 77.4 Pronouns and demonstratives 77.4.1 Free personal pronouns 77.4.2 Bound personal pronouns 77.4.3 Demonstrative pronouns 77.4.4 Interrogative/indefinite pronouns 77.5 Verbal morphology 77.5.1 Conjugation classes 77.5.2 Intransitive verbs 77.5.3 Transitive verbs 77.5.4 Semi-transitive and ditransitive verbs 77.5.5 Inflectional categories 77.5.5.1 Overview 77.5.5.2 Present tense 77.5.5.3 Past tense 77.5.5.4 Future tense 77.5.5.5 Imperative 77.5.5.6 Imperative continuous 77.5.5.7 Past habitual 77.5.5.8 Habitual 77.5.6 Verb deriving morphology 77.5.6.1 Inchoative suffix -yi ~ -wi ~ -i 77.5.6.2 Collective subject -ji 77.5.6.3 Associated motion -gayi 77.5.6.4 Put -ju 77.5.6.5 Causative -ma 77.5.6.6 Reciprocal/reflexive -marri 77.5.7 Subordinate clauses 77.5.7.1 Overview 77.5.7.2 Circumstantial same subject 77.5.7.3 Circumstantial different subject 77.5.7.4 Verbal purposive

xliv

935 935 935 935 936 936 936 936 937 937 937 937 937 938 938 938 938 938 938 938 939 939 940 940 941 942 942 942 942 942 943 943 943 943 943 943 943 944 944 945 945 945 945 945 945 946 946 946 946 946 947

detailed contents 77.5.7.5 Apprehensive 77.5.7.6 Origin -thanu 77.6 Adverbs, particles, and interjections 77.6.1 Overview 77.6.2 Adverbs 77.6.3 Particles 77.6.4 Interjections Acknowledgements 78. The revitalization of the sleeping Tasmanian Aboriginal languages palawa kani Annie Reynolds and Theresa Sainty 78.1 palawa kani means ‘Tasmanian Aborigines speak’ 78.2 Tasmanian Aboriginal languages 78.3 Tasmanian Aboriginal Language Project 78.3.1 palawa kani Language Program 78.3.2 Community language workshop 1994 78.4 Sources of recorded language 78.4.1 Glossaries 78.4.2 The sounds of the original Tasmanian Aboriginal languages 78.4.3 Assessing recorders’ value 78.4.4 Was Robinson a good recorder? A linguistic analysis of Robinson’s value as a recorder

78.5 Regions, speakers, and places 78.5.1 Language regions 78.5.2 Aboriginal speakers 78.5.3 Names of places 78.6 Reconstructing words 78.7 Grammar and conventions 78.7.1 Plurals 78.7.2 Word order 78.7.3 Affixes and suffixes 78.7.4 Tense 78.7.5 Adaptation and new words 78.7.6 Stress 78.7.7 Capital letters 78.8 Domains of use References Language and Language Family index Index of places mentioned Subject index

947 947 947 947 947 948 948 949 950 950 950 951 952 952 952 953 953 954 954 954 954 955 955 955 956 956 956 956 956 956 957 957 957 958 1061 1072 1074

xlv

Series Preface We know that the close study of individual language families and linguistic areas is vital both to the synchronic and diachronic study of language and to cognitive science more widely. Comparative investigations of this type stimulate exciting synergies between different subdisciplines of linguistics, such as language change, contact linguistics, sociolinguistics, linguistic typology, textual philology, and microvariation in grammar, sound, and meaning within and across languages. Besides reflecting and encouraging the links between these subdomains, the fundamental goal of the series is to publish high-quality, substantial reference works which represent a set of theoretically informed and systematic guides to what is known about the world’s languages. Each Guide focusses on a particular language family, subfamily, or areal grouping, and is edited by leading authorities, who bring together contributions from the best international scholars in the field. The Guides aim to show the more general theoretical significance of the languages’ history, linguistic and sociolinguistic characteristics, and overall to provide an indispensable reference tool both to specialist scholars and students and to professional linguists. The approach adopted in all the Guides is systematic and comparative, informed by the latest research and theoretical and methodological perspectives, and, where appropriate, the authors draw on relevant work in such fields as anthropology, archaeology, and cognitive science. Adam Ledgeway and Martin Maiden University of Cambridge and University of Oxford

xlvi

Abbreviations and conventions < > 1 1” 1|2 12 12” 1+2 1.foc 1pes 1ss 1st 2 3 3 3po 3ps a A A A a a a/p AA ABC abl abs acc acs act act add ades adj adj admon agm AIATSIS all

acted on by acting on first person first person non-minimal exclusive person first person inclusive first person inclusive non-minimal first person inclusive first person focalization marker first person plural exclusive subject first person singular subject verbal inflection class second person third person transitive subject a acting over object b third person third plural object third plural subject augmented agent alveolar Assimilation animate subject of transitive clause antipassive AusPhon-Alternations (dataset) Australian Broadcasting Corporation ablative absolutive accusative accessory active actual additional adessive adjective adjacent demonstrative admonitive augmented number Australian Institute of Aboriginal and Torres Strait Islander Studies allative

amb ambiph ana anaph and anim apas app aor ap appl applic art asp ass ass assert assoc att aug aux auxc aver awy BBA ben bgn C c c/t C1 card cataph caus cc cfact char cl class2verb clc clf cm

ambulative ambiphoric anaphor anaphoric connector, conjunction ‘and’ animate object antipassive apprehensive aorist antipassive applicative applicative article (Heath 1980b) aspectual suffix associative case assertive assertive associative attenuative augmented number auxiliary auxiliary (continuous aspect) (Heath 1980b) averative motion away Bouckaert, Bowern, and Atkinson (2018) benefactive Burarra/Gun-nartpa language catalyst contemporary tense contemporary? initial consonant of a word cardinal pronoun cataphoric causative circumstantial clause counterfactual characteristic classifier class 2 verb stem forming affix clitic classifier conjugation marker

xlvii

abbreviations and conventions CNSW CofR coll com comit comp con con conj cont contr cp cs csl ctf ctr ctrp cts cv CYP D d d d.b. dat dc dec def deic dem dem(n) der des desid detr di direc dist dl dm do dpf ds d.term dtr 1 2 3

Central New South Wales change of referent collective comitative comitative complementizer concomitative contemporary tense conjunction continuative aspect contrastive focus compulsional changed state causal counterfactual contrast centripetal continuous, non-moving coverb suffix Cape York Peninsula dental distal (deictic) directional dative/benefactive dative direct case deceased definite deictic suffix demonstrative demonstrative (neuter) derivational1 desiderative desiderative detransitivizer desiderative-intentional suffix directional distal dual demonstrative marker direct object ‘departing from’ affix different subject direction, terminative de-transitivizer

du dub dur e e ed edge eff el ela EMA emp emph en Eng ep erg ex exc excl f fact fem FF f.kposs FM foc foc fore fut G g gen genevt gent gentil ger gg gun GUR GW H h hab

dual dubitative durative extended argument English edible gender edge, side effector elative elative electromagnetic articulography emphatic clitic emphatic epenthetic nasal English borrowing epenthetic morpheme ergative exclusive exclusive exclusive feminine gender factive feminine gender father’s father morpheme cross-referencing female propositus of a kin relationship father’s mother focus contrastive focus foregrounding clitic future Gugada goal genitive generic event gentilic gentilic gerund Gurr-goni gun noun class Gurindji Gunwinyguan high tone2 higher object3 habitual

In Chapter 70 this is glossed as ‘derivative affix’, which it is assumed is identical to ‘derivational’. For additional annotations associated with ToBI transcription conventions, see Beckman et al. (2006). In interlinear glossing

xlviii

abbreviations and conventions habit have hesit hist hith hyp I I IE id ideo I-IV im imm imm imp imperfSS impf impipfv impl implic inc incep inch incho incl ind ind indf inf ins inst instr int intens inter interr intr inv io ipfv ir irr itrv 4 5 6 7

habitual have suffix hesitation particle historical past hither hypothetical inanimate Iwaidjan Indo-European identifiable ideophone numbered noun classes immediacy immediate (past) immediacy imperative imperfective tense/aspect same subject imperfective imperative imperfective implicated implicative inclusive inceptive inchoative inchoative inclusive indicative individuation indefinite infinitive instrumental instrumental instrumental interrogative intensifier interrogative interrogative intransitive inverse indirect object imperfective irrealis irrealis iterative aspect

jin kin kposs Kr Ku L L L L l/i lct lig lim ll loc loc/i loct lots lvc M m m ma m.alp masc masc med med MF min mj m.kposs mloc MM mn mobl motiv mov Mp m.poss mprop msc

jin-marked noun class kinship marker kinship possession Kriol borrowing Kukatja labial lenition4 local5 low tone6 locative instrumental locative possessor ligative limitative land gender locative local/instrumental case locational case grammatical morpheme meaning ‘lots’ light verb constructions Manyjilyjarra minimal (number)7 masculine gender masculine gender (Singer 2006a) masculine alienable possessive masculine gender masculine noun class medial medial demonstrative mother’s father minimal (number) major member morpheme cross-referencing male propositus of a kin relationship marked locative mother’s mother minor member modal oblique motivative continuous, moving manipulative morpheme cross-referencing a male class possessor of a body part modal proprietive masculine class

In Chapter 12 only Languages attesting cognates of a particular etymon but lacking coherence as a phylogenetic group. In Chapter 14 In Chapter 70

xlix

abbreviations and conventions msc.kposs msc.poss mu mut N n nact nader nar NC nc NCD ndj neg neu neut nf nf nfut Ngaany Ngal NITV nk nlz nm nmlz nmr nmzr nom nomlsr now np nPN npp nprox nprs npst ns nsg num nvis Ny Ny2 o

l

morpheme cross-referencing masculine class propositus of a kin relationship morpheme cross-referencing a masculine class possessor of a body part mu-class marker of symmetrical information access from speaker perspective noun neuter gender non-actual (irrealis) mood noun class marker narrative nasal + stop cluster noun classifier nasal cluster dissimilation Ndjébanna negative neuter gender neuter noun class non-feminine non-future non-future Ngaanyatjarra Ngalia National Indigenous Television (Australian TV network) Na-Kara nominalization non-masculine (Nordlinger 1998a) nominalizer nominalizer nominalizer nominative nominalizer temporal/contrastive clitic ‘now’ non-past non-Pama-Nyungan non-past progressive non-proximal demonstrative non-present non-past non-subject non-singular number numeral non-visible Nyungic Nyulnyulan object

obj objcomp obl oblig obliq OLAC op opt or orig orig OVV P p p p pa pair PARADISEC part part partic pass past past pun past.punct pauc pc pc pc pcon perf perl pers pes pf(v) phab pi pi pimp Pitj P-L pl plup plur PN PNy

object object complementize oblique obligative oblique Open Language Archives Community object-promoting suffix optative mood originative (case) originative origin case O’Grady, Voegelin, and Voegelin (1966) Paman past tense patient-like argument plural pronoun paucal number enclitic (in the vp) pair suffix Pacific and Regional Archive for Digital Sources in Endangered Cultures participial partitive participle passive past tense past punctual past punctual paucal past completive past continuous pre-contemporary tense past continuous perfective perlative personalizer plural exclusive subject perfective past habitual prior information past imperfective past imperfective Pitjantjatjarra Pintupi-Luritja plural pluperfect plural Pama-Nyungan Pama-Nyungan

abbreviations and conventions poss possd pot pot pp p.pfv pPN pPNy pr pr pre preccomp pres priv prm pro prog proh prom prop prox prox prs prt pst pst.compl pst.cont pst.def pst.indef ptcp punct pur purp purpcomp purps=o q qual qual’r ques R r/a rdp re real rec recog

possessive possessed noun potent potential past perfective past perfective Proto-Pama-Nyungan Proto-Pama-Nyungan present present (Harvey 2002) precontemporary tense preceding event complementizer present privative prominence marker (personal) pronoun progressive prohibitive prominence proprietive proximal demonstrative proximate present part suffix past tense past completed past continuous past definite past indefinite participle punctiliar (Birk 1976) purposive (Heath 1980b) purposive purposive [subsequent] event complementizer purpose subordinate subject is main clause object interrogative quality nominalizer, Jaminjung qualifier question particle (Sharp 1998) retroflex realis/assertive reduplication realis realis recognitional recognitional

redup refl reflex refr rel rel.ds rel.ss rem rememb rem.pst resum rls rm round rpa rr s s s sa sbj sel seq seq.ss ser(ial) sg side1 side2 sp ss stat stem sub subj subord subsect SVC tag TAM texd th th then to

reduplication reflexive reflexive referential relative relative (different subject) relative (same subject) remote item previously mentioned remote past resumptive realis reflexive/reciprocal rounded form remote past reflexive/reciprocal single argument of canonical intransitive verb (subject) singular subject intransitive/transitive subject subject selective enclitic sequential sequential, same subject serial singular locating referent in small locational context (cardinal) locating referent in large locational context (cardinal) ‘speed’ form (action undertaken before departing) same subject state of affairs (contrast action) stem-forming affix (McKay 1975) subordinate subject subordinate subsection (skin) serial verb construction tag question tense/aspect/mood text deictic thematic temporal clitic temporal/contrastive clitic ‘then’ motion towards

li

abbreviations and conventions too top tpa tr tr trans tre trvz twd ua ua uap uaugm unf unm unmkd up usp V

‘too’ clitic (Simpson 1991) topic today past (inflection) transitive transitive conjugation marker transitive temporal relative transitivizer towards (directional) unaugmented unit augment unit augmented Past unit augment unfamiliar unmarked [verb form] unmarked verticality underspecified, phonologically null TAM velar

V V1 val vb vblz vcomp veg W w WALS Wang Warn with WL wu Ya yk Yul

vowel initial vowel of a word valence increaser (Reid 1990) verbalizer verbalizer sentential complement index vegetable noun class wife celestial class World Atlas of Language Structures Wangkajunga Warnman associative Wafer and Lissarrague (2008) wu-gender Yankunytjatjara ‘you know’ clitic Yulparija

Abbreviations of kin terms These abbreviations are used widely in anthropology. F Fa M Ma S D B

‘father’ ‘father(’s)’8 ‘mother’ ‘mother(’s)’ ‘son’ ‘daughter’ ‘brother’

Z H W e y m f

‘sister’ ‘husband’ ‘wife’ ‘elder’, e.g. eB ‘elder brother’ ‘younger’ e.g. yZ ‘younger sister’ ‘male propositus’ e.g. mZ ‘male’s sister’ ‘female propositus’ e.g. fB ‘female’s brother’

Symbols are concatenated to form complex kin terms e.g. MBDS ‘mother’s brother’s daughter’s son’

Other conventions: =

morpheme boundary clitic boundary

Ages of children are expressed in the format years;months

.

‘when a single object-language element is rendered by several metalanguage elements (words or abbreviations)’

8 This (and ‘Ma’ for ‘Mother’) are from Carew and Beltran, Chapter 70. Since they use F for ‘feminine gender’ and M for ‘minimal’ noun marking, non-standard abbreviations are used for kinship terms.

lii

abbreviations and conventions

Transcription conventions The individual papers in this volume use several sets of conventions in spelling language examples (International Phonetic Alphabet, standardized language orthographies, and ad hoc researcher conventions). These are explained in the chapters themselves. However, there are a number of conventions that apply across many chapters, which are detailed here.

Consonants Transcription of consonants, listed left-to-right in the order bilabial, lamino-dental, apicoalveolar, apico-postalveolar (retroflex), lamino-alveopalatal, and dorso-velar, are notated as follows. Notational variants are separated by a comma. See Round, Chapter 10, this volume, for further details. Stops

p

th

t

rt

c, ty

k

b

dh

d

rd

j, dy

g

Fricatives

ß

ð

Nasals

m

nh

n

rn

ñ, ny

lh

l

rl

ʎ

R, r

y

Liquids

γ ng, ŋ

rr Glides

w

yh

n.g is a heterorganic nasal+stop cluster, as distinguished from the digraph ng. h and ɦ are glides, respectively without and with voicing. rr is a tap or a trill; where the two are distinguished rrh is used for the trill. R is glide (IPA /ɹ/), reserving single lowercase r for diacritic use to indicate retroflection of stops, nasals, and laterals. ʔ is glottal closure, classed as a stop, glide, or prosodic boundary marker as appropriate for individual languages. This is also represented as 7 or ’ in some languages. lf and rlf are flapped laterals tr , rtr , dr , and rdr are trill-released stops n p, n t etc. are prenasalized stops *T is an initial apical non-nasal (*t,*n,*l,*r, *rr) in a cited reconstruction in which the initial consonant is not material to the discussion. | (vertical stroke) separates material in a daughter-language that is additional to the reconstructed form of an etymon. I use the template C1 V1 C2 V2 C3 V3 …, where C2 and C3 can represent any intervocalic consonant or cluster, to show syllabic structure, omitting subscripts as appropriate.

Vowels: i e

a

u o

v (IPA /ə/) is an unstressed mid-to-high central vowel, not the same as 3, a (primary or secondary) stressed mid-to-high central vowel. For the Arandic languages, orthographic is used for the stressed non-low vowel, as in the sources. The front-rounded vowel symbols o¨ and œ are used as in the original transcription sources.

liii

The Contributors Akshay Aitha University of Chicago Barry Alpher independent researcher Rob Amery University of Adelaide Magda Andrews-Hoke independent researcher Denise Angelo Australian National University Sarah Babinski Yale University, University of Zurich Xavier Bach University of Oxford James Bednall Australian National University Joe Blythe Macquarie University Catherine Bow Charles Darwin University and Australian National University Claire Bowern Yale University Margit Bowler University of Manchester John Bradley Monash University Parker Brody Yale University Mitchell Browne University of Queensland Margaret Carew Batchelor Institute Lucinda Davidson University of Melbourne Jessica Denniss University of Toronto Greg Dickson Australian National University Lydia Ding Carleton College Samantha Disbray University of Queensland Vivien Dunn University of Queensland K. Eira independent researcher Thomas Ennever Monash University Nicholas Evans Australian National University Janet Fletcher University of Melbourne Alice Gaby Monash University Mary-Anne Gale University of Adelaide John Giacon Australian National University Jennifer Green University of Melbourne David Felipe Guerrero-Beltran Universidad Nacional de Colombia, Université de Paris, and University of Melbourne Amanda Hamilton-Hollaway University of Queensland Sue Hanson Goldfields Aboriginal Language Centre George Hayden Noongar Boodjar Language Cultural Aboriginal Corporation

liv

Rachel Hendery Western Sydney University Dorothea Hoffmann University of Oregon and The Language Conservancy Kathleen Jepson University of Melbourne Ivan Kapitonov University of Cologne Barbara Kelly University of Melbourne Juhyae Kim Cornell University Harold Koch Australian National University Inge Kral Australian National University Dana Louagie KU Leuven John Mansfield University of Melbourne Alexandra Marley Australian National University Doug Marmion Australian Institute of Aboriginal and Torres Strait Islander Studies Patrick McConvell Australian National University Felicity Meakins University of Queensland Francesca Merlan Australian National University Luisa Miceli University of Western Australia Sarah Mihuc McGill University Ilana Mushin University of Queensland Rachel Nordlinger University of Melbourne David Osgarby University of Queensland Carmel O’Shannessy Australian National University Josh Phillips Yale University Maïa Ponsonnet University of Western Australia Annie Reynolds Tasmanian Aboriginal Centre Luis Miguel Rojas-Berscia University of Queensland Katherine Rosenberg independent researcher Erich R. Round University of Surrey and University of Queensland Theresa Sainty Tasmanian Aboriginal Centre Margaret Sharpe University of New England Oliver Shoulson Yale University Jane Simpson Australian National University Amalia Skilton Cornell University Denise Smith-Ali Noongar Boodjar Language Cultural Aboriginal Corporation Stef Spronck University of Helsinki

the contributors Clara Stockigt University of Adelaide Marija Tabain La Trobe University Nicholas Thieberger University of Melbourne Sally Treloyn University of Melbourne Marie-Elaine van Egmond University of Greifswald and Martin Luther University Halle-Wittenberg

Jill Vaughan University of Melbourne Jean-Christophe Verstraete University of Leuven and Australian National University Michael Walsh University of Sydney Gillian Wigglesworth University of Melbourne

lv

Language maps The following maps were based on a compilation of sources made over the period 2008–2020. The maps show language family locations to the extent known. The drawing of the language maps and decisions about which varieties and names to represent is described in more detail in Chapter 7. The following notes should be taken into account especially. As should be clear, the placement of languages in physical space is an abstraction and an idealization. Languages are used by people, who are often multilingual. While some Indigenous groups view language as being particularly associated with geographical regions (that is, that tracts of land belong to certain languages and people speak those languages by virtue of being in that space; see Rumsey 1993, Rigsby 2005, Sutton 1978, for example; see further discussion in Chapters 1 and 49 of the current volume), not all groups view language in that way. These maps are also ‘atemporal’; that is, they are not a specific snapshot of a particular point in time. Language locations change over time. All Indigenous language locations have been affected, to varying degrees, by colonial settlement (cf. Department of Infrastructure, Transport, Regional Development and Communications (DoITRDC), Australian Institute for Aboriginal and Torres Strait Islander Studies (AIATSIS), and Australian National University (ANU) 2020). This is not a map of contemporary language use, but it’s probably also not a map of traditional language locations for some parts of the country. At best, it could be a reconstruction for some areas, contemporary continuation for others, and contemporary change in yet other areas. These maps should be considered as a way to relate linguistic groups to one another in space. It should help to visualize the approximate distances between groups, abstracting away from multilingualism, population density, and Indigenous settlement patterns. This map represents my best attempt to reconcile the myriad of conflicting information about Australian languages and their traditional and contemporary locations. It is important to note that even ‘reliable’ sources are sometimes irreconcilable. Reasons for conflicting information include a difference in the time period represented, working with different groups or individuals, different ideas of what constitutes a ‘language’, different choices of which named linguistic varieties to group together, vagueness in prior maps, and aesthetic mapping choices that change information. For these and other reasons, I may have come to different decisions than others about how to represent language–land relationships in Australia, although for the most part, this map is broadly similar to others. Perhaps the biggest difference is the number of languages represented: 490, rather than the ‘250’ that linguists have tended to work with, as further discussed in Chapter 7. The language map is keyed to the classification beginning on page lxi. Since that is a classification of languages (rather than all named varieties), only languages are shown on the map. This means that groups where the language name and the name of the group are different, show the language names rather than the group names. For some parts of the country, the ethnonyms and the language names are identical (this is true for most, though not all, of the Western and Central blocs of Pama-Nyungan, for example). In parts of Eastern Australia, however, language names were not identical to names for social groups. Maps 1 and 2 include classification information, such as families and subgroups. PamaNyungan subgroup boundaries are represented by dashed lines; language families have

lvi

language maps solid lines. These subgroups of Pama-Nyungan are the ones most familiar to Australianists. They descend from the earlier classifications of roughly 30 groups that were originally not classified more closely together (cf. the Pama-Nyungan rake model described in Bowern and Atkinson 2012). The classification presented on the map is somewhat conservative. For example, it does not show substructure within Paman or Maric in Map 2. The map was designed with historical linguistic uses in mind (for example, units for reconstruction). It does not show groups with cultural affinities to one another (for example, the languages of the Victoria River district) or other ways of grouping languages. Map 3 does not present boundaries between individual languages. This decision was not undertaken lightly. It was done because it is impossible to represent boundaries accurately in too many cases. In some places, the boundaries themselves are unknown; single fixed boundaries are not the best way to represent the way languages in particular regions shaded into one another; boundaries are disputed; boundaries have changed over the last few hundred years; and other reasons. Of course, the boundaries given in Maps 1 and 2 are derived from language locations presented in Map 3 and are subject to some of the same types of considerations. They should be treated as indicative rather than definitive. The process of compiling information was as follows. The starting point for this project was the compilation of a language map and list of ‘standardized’ language names for use in the Chirila database (Bowern 2016a). The language map included both centroid (point) locations and polygons showing the approximate boundaries of languages. In order to be as complete as possible, data were compiled from all available sources on language locations, as well as the language names used in general sources about Australian languages. Numerous compilations exist in the prior literature, with the most important continent-wide surveys including O’Grady, Wurm, and Hale (1966), Tindale (1940; 1974), Wurm and Hattori (1981), and Dixon (2002). Regional surveys were also important, and included both published sources (e.g. McGregor 2004; Wafer and Lissarrague 2008) and unpublished or locally published maps and pamphlets from regional Aboriginal language centres (for example, the Pama Language Centre, VACL, Wangka Maya, the Goldfields Language Centre, and Noongar Boodja Aboriginal Culture Language Centre). The maps were overlaid and geocoded in Google Earth to facilitate comparison. These were checked against the Austlang website from the Australian Institute of Aboriginal and Torres Strait Islander Studies. Some decisions about mapping are given in the notes to the classification, starting on page lxi. In general, I privileged information from Language Centres and from specific reference grammars over the older maps such as Tindale (1974). For the most part, sources were in broad agreement, as might be expected since they were not compiled independently. However, there were substantial minor differences, particularly around boundaries.

lvii

language maps 1. Non-Pama-Nyungan language families Legend

Tiwi

Family name

N

Darwin region Iwaidjan Tiwi Maningrida

Family boundary

Kungarakany Anson Bay Northern Daly Eastern Daly Western Daly Southern Daly

Giimbiyu Gaagudju Wagiman Mangarrayi Wardaman

Gunwinyguan Warndarrang Marran

Alawa

Wororan Western Mirndi

Eastern Mirndi

Garrwan

Tangkic

Jarrakan

Nyulnyulan Bunuban

0

200

400 km

100

0

200

300 miles

Area covered by this map

Scale

2. Major subgroups of Pama-Nyungan Legend Subgroup name

Western Torres

Yolngu

N

Subgroup boundary

Yolngu Paman

Warluwarric

Dyirbalic

NgumpinYapa

KanyaraMantharta

Mayi Warumungu

Ngayarta

Marrngu

Dharumbal

Arandic

Kalkatungic Maric Karnic

Wati

Yardli

Kartu

ThuraYura

Southwest

CentralNSW Paakintyi Nganyaywana

Kulin

Bunganditj 500

0

Scale

lviii

200

1000 km 400

600 miles

Durubalic

Bandjalangic

Yorta Yorta Lower Murray

0

Bigambalic WakaKabi Muruwari

Gumbaynggiric Yuin-Kuri

Eastern Victoria

language maps 3. Language locations

218

135 136 227 180 257

460

N

352 253 261 9 207 108 330 181 357 236 54 36 89 435 199 46 124 105 198 288 50 366 214 335 400 370 358 87 85 436 431 60 237 154 219486 143 221 271 376 223 206 212 269 264 59 217 398 268 222 250 220 483 138 393 216 5 297 452 226

To

p

En

266

d

114

144

31 305 289

429

360

137

Ki m

364

160 438 277 209 43 247 123 391 141 395

156

211

146

394

485

383 274

387

399

325

213

161

396

177

203 299

453

328

259

354

265 193

27

58

151 29

379 300 324 412 155 111

159 432

242 263

306

239

39 500 km

0 200

Scale of main map

69

150

234

Yo C a p e in s u Pe n

77 63 73

400 miles

466

Legend Language number Subgroup boundary

75 80

433 406 401 82 488 303 381 116 104 45 397 115 231 153 166 467 25 47 283 385 477 7 244 451 30 425 445 448 126 16 132 14 446 32 68 386 84 473 443 17 90 228 329 392 411 389 55 465 97 96 129 100 373 374 95 38 162 182 449 37 243 388 157 110 327 41 65 201 459 272 120 461 434 112 122 15 83 331 215 267 121 468 380 35 128 101 450 292 74 18 444 355 298 127 28 390 42 351 139 251 372 56 106 490 326 168 130 238 131 1 33 489 356 57 210 314 184 482 26 202 377 457 317 273 424 93 464 339 345 313 319 295 119 287 163 260 421 341 320 133 48 131 423 66 164 481 61 427 262 286 167 225 479 256 94 254 278 195 408 255 290 67 86 458 410 64 284 430 409 275 480 441 322 71 140 334 456 52 353 76 62 70 81 312 248 407 403 426 183 349 49 485 346 40 296 173 405

302

270

447 107 78 72 79

152

454

158 301 455 281 224 469 318 316 323 185 148 344 472 145 332 348 350 280 402 333 471 347 378

0

338

103

b e rle y

8

240

205

109

293

208

484

91

53

304

343 291

371

428

10

279

321

165

102

241

276

462

404

294 147 463

194

437

142 363

197

384

92

422

13

44 282

113 442 365 20 440 487 252 258 246 192 12 6 204 475 19 22 235 361 474 416 149 200 417 359 419 415 340 420 191 414 88 34 175 413 418 24 362 315 125 188 337 196 189 311 178 2 367 476 369 23 478 170 3 308 172 368 309 134 179 171 310 307 190 174 169 118 186 249 187 336 117 98 233 21 375 342 176 382 4 245

11

r la k

99

439

470

230 229

51

232 285

lix

language maps 4. New Indigenous languages (see p xciii)

Legend

25

New language

1

17

State border State capital city

Lake

21 20

5a

16

5b 15

42

18

Darwin

River

3

7

6

5c

N

10 24 13

9

NOR TH E RN

8

TERRITORY

22

19

W E ST ER N

QUEENSLAND

11

AUS TR A LIA

12

14

SOUTH

Brisbane

AUS TRA LIA NE W SOUTH

Perth

23

WA L E S

Adelaide

Canberra

VIC TORIA

Melbourne

500

0 0

Scale

lx

200

AC T

1000 km 400

600 miles

TA S MA NIA

25

Hobart

Sydney

Key to language maps Name

Map No.

Name

Map No.

Adnyamathanha Aghu Tharnggala Agu Aloja Agwamin Alawa Alngith Alyawarr Amangu Amurdak Andajin Angkamuthi Anguthimri Anindilyakwa Anmatyerr Antakirinya Antekerrepenhe Ara-ara Arabana Aritinngithigh Atampaya Athima Awngthim Awu Alaya Ayapathu Ayerreyenge Baanbay Badimaya Badjiri Balardung Barada Bardi Barna Barranbinya Barrow Point Barunggam Batjjamalh Batyala Bayali Bibbulman Bidawal Bidjara Bigambal

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42

Bilinarra Binbinka Bindal Bininj Kunwok Biri Birrpayi Boonwurrung Bukurnidja Bularnu Bunganditj Bunuba Burarra Central Arrernte Condamine-Upper-Clarence Copmanhurst Language Cundeelee Wangka Dagoman Dalabon Darrkinyung Daungwurrung Dhangu Dharawal Dharawala Dharuk Dharumba Dharumbal Dhay’yi Dhudhuroa Dhurga Dhuwal Dhuwala Diyari Djabugay Djadjawurung Djangu Djinang Djinba Djirbal Djirringany Dungaloo Duungidjawu Eastern Arrernte

43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84

lxi

key to language maps

lxii

Name

Map No.

Name

Map No.

Emmi Eora Erre Flinders Island Gaagudju Gabalbara Gajirrebeng Gamberre Gamilaraay Gandangara Gangulu Ganulu Gara-gara Garandi Garig Garingbal Garlali Garrwa Gija Giya Giyug Gold Coast Pimpana Golpa Gonbudj Gooniyandi GoorengGooreng Goreng Gubbi Gubbi Gudang Gudanji Gudjal Gugu Badhun Gugu Djangun Gugu Wakura Gumbaynggir Gungabula Gunggari Gunya Gurindji Gurr-goni Guugu-Yimidhirr Guwa Guwamu Guwar Guweng Guyambal Guyani Guynmal Hunter River Lake Macquarie

85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133

Ikarranggal Ilgar Iwaidja Jabirrjabirr Jaminjung Janday Jardwadjali Jaru Jawi Jawoyn Jingulu Jiwarli Jiwarliny Jukun Jurruru Kaanju Kala Lagaw Ya Kaalamaya Kalaw Kawaw Ya Kalkatungu Kamu Kaniyang Karajarri Karangura Kariyarra Karlaaku Karranga Kartujarra Karuwali Katthang Kaurna Kayardild Kaytetye Keramin Kokatha Kokiny Koko Bera Koko Dhawa Koko-Babángk Kolakngat Koogobatha Kugu Nganhcara Kukatj Kukatja Kuku Thaypan Kuku Yalanji Kunbarlang Kungarakany Kungkarri

134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182

key to language maps Name

Map No.

Name

Map No.

Kurnai Kurnu Kurrama Kurtjar Kuthant Kuuk Thaayorre Kuuk Yak Kuuk-Narr Kuuku Yani Kuuku-Ya’u Kuwarra Kwini Ladji Ladji Lamalama Lardil Larrakia Limilngan Linngithigh Lower Aranda Lower Richmond Luthigh Luthigh Mabuiag Magati Ke Malak Malak Malkana Malngin Malyangapa Mangala Mangarrayi Manjiljarra Maranunggu Margany Marra Marramaninjsji Marrgu Marri Ammu Marri Dan Marri Tjevin Marri Ngarr Marrithiel Martuthunira MathiMathi Matngele Mawng Mayawali Mayi-Kulan Mayi-Kutuna Mayi-Thakurti

183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231

Mayi-Yapi Mbabaram Mbara Mbeiwum Mengerrdji Menthe Middle Clarence Minang Minkin Miriwoong Mirniny Mithaka Miyan Mpakwithi Mpalityan Mudburra Muk-Thang Muluriji Murrinhpatha Muruwari Muthanthi Nakara Nari Nari Narrungga Nauo Ndjébbana Ndra’ngith Ngaanyatjarra Ngadjuri Ngaduk Ngaiawang Ngajumaya Ngalakgan Ngalia Ngaliwuru Ngamini Ngan’gikurunggurr Ngan’gityemerri Nganakarti Ngandi Ngantangarra Nganyaywana Ngardi Ngarigu Ngarinyin Ngarinyman Ngarkat Ngarla Ngarlawangga

232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280

lxiii

key to language maps

lxiv

Name

Map No.

Name

Map No.

Ngarluma Ngarnka Ngaro Ngarrindjeri Ngawun Ngintait Ngiyambaa Ngomburr Ngumbarl Ngunawal Nhanda Nhirrpi Nhuwala Nimanburu Nukunu Nulit Nungali Nunukal Nyaanyatjarra Nyaki Nyaki Nyamal Nyangumarta Nyawaygi Nyikina Nyulnyul Nyunga Ogh Awarrangg Ogh-Alungul Ogh-Angkula Ogunyjan Olkola Omeo language Paakantyi Paaruntyi Pakanh Palyku Pantyikali Panyjima Parnkala Parrintyi Payungu Peramangk Pinikura Pinjarup Pintupi-Luritja Pirlatapa Pirriya Pitjantjatjara Pitta-Pitta

281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329

Pungupungu Punthamara Purduna Putijarra Raminyeri Rembarrnga Ribh Rimanggudinhma Ritharrngu Southern Anaiwan Southern Kaantju Southern Paakantyi Takalak Thaagurda Thalanyji Thanggatti Thangguai Tharrayi Tharrgari Thawa Thiin Thirarri Tiwi Tjapwurrung Tjupan Turrbul Tweed-Albert Tyerraty Umbugarla Umbuygamu Umiida Umpila Umpithamu Unggarranggu Unggumi Uradhi Urningangk Uw El Uw Oykangand Uw-Olgol Uwinymil Waanyi Wadikali Wadjabangayi Wadjigu Wagaman Wagiman Wailwan Wajarri

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378

key to language maps Name

Map No.

Name

Map No.

Wajuk WakaWaka Wakaya Walangama Walmajarri Wambaya Wanamara Wangan Wangkajunga Wangkangurru Wangkayutyuru Wangkumara Wanyjirra Wapabara Wardaman Wardandi Warlmanpa Warlpiri Warluwarra Warndarrang Warnman Warray Warrgamay Warriyangga Warrnambool Warrwa Warumungu Warungu Wathawurrung WathiWathi Waywurru Wemba Wemba Western Arrarnta Wiilman Wik Iyanh Wik Me’anh Wik Mungkan Wik-Alkan Wik-Ep Wik-Ngathan Wik-Ngatharr Wik-Paacha Wilyakali Winjarumi Wiradjuri Wirangu Wirri Woiwurrung Worimi

379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427

Worla Worrorra Wotjobaluk Wubuy Wudjari Wulguru Wuli Wuli Wulna Wulwulam Wunambal Wurlayi Wurrugu Wuthathi YabulaYabula Yadhaykenu Yagalingu Yagara Yalarnnga Yambina Yan-nhangu Yanda Yandjibara Yandruwandha Yangga Yangman Yankunytjatjara Yanyuwa Yapurarra Yaraldi Yardliyawarra Yari-Yari Yarluyandi Yatay Yawarrawarrka Yawijibaya Yawuru Yaygirr Yetimarala Yidiny Yilba Yiman Yindjibarndi Yindjilandji Yingkarta Yinhawangka Yiningayi Yintyingka Yinwum Yir Yoront

428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476

lxv

key to language maps

lxvi

Name

Map No.

Name

Map No.

Yirandhali Yirrk Thangalkl YithaYitha YortaYorta Yu-Yu Yugambal Yugul

477 478 479 480 481 482 483

Yukulta Yulparija Yunggurr Yupngayth Yuru Yuwaalaraay Yuwaliyaay

484 485 486 487 488 489 490

Australian language families and linguistic classifications This material presents classification information for the Indigenous languages of Australia. They include languages spoken in the area now known as Australia prior to 1788, as well as subsequent languages which have arisen through language contact. Languages are grouped by relationship, using the non-Pama-Nyungan/Pama-Nyungan distinction for convenience. I compare the current ‘Bowern’ classification with three others: Dixon (2002), Wurm (1972), and O’Grady, Voegelin, and Voegelin (1966). Other classifications could be used, and others also vary somewhat in the details, but as Walsh (1997b) has noted, the overall number of languages and broad principles of classification have not changed substantially since O’Grady, Voegelin, and Voegelin (1966). That is apparent here, too: though some details vary, and some individual languages are placed in different subgroups or families, there is broad consensus across the classifications. The main difference is in the number of languages, discussed in more detail in Chapter 7 of this volume. This comparison has both some historical value, in seeing how classifications of Australian languages have changed over time and should orient the reader to how the current classification compares to others established in the literature. The internal structure of some of the subgroups is simplified for ease of display. Some of the names have been adapted to make comparisons easier across sources (e.g. I do not list the Yibian Subgroup in OVV, using simply ‘Wardaman’ as the name). I have rearranged the order of families from previous classifications so as to align the classifications more clearly. I have kept Dixon’s names, though I have abbreviated them in places so as to make the items clearer, since Dixon does not follow the naming conventions of previous literature in many cases. Note that while groups within cells are comparable, the languages may occur in different orders within cells. The ‘Bowern’ classification presented here is based in part on results published in Bouckaert et al. (2018), which in turn is one publication from approximately 12 years’ work on language relationships, particularly in Pama-Nyungan (and between Pama-Nyungan and non-Pama-Nyungan families). That is, the current classification was compiled from a combination of prior sources (evaluated for reliability), inspection of primary materials, including cognate coding, and the results of phylogenetic work published in Bouckaert et al. (2018). However, here I present a fairly conservative classification, especially as regards to groupings between Pama-Nyungan subgroups. I do not list all internal structure within Bouckaert et al. (2018), except where it is well established from previous classifications. Though the Bowern classification is ‘flat’ (in not showing relationships within, for example, the Northern languages in Pama-Nyungan), this is for ease of reference and because the details of the more remote relationships within Pama-Nyungan are tentative, not because I believe they don’t exist. This handbook does not focus on the classificatory work done on Australian languages in the 20th century, mostly because there are already substantial reviews of these classifications in the work of Koch (2014a; Koch and Nordlinger 2014) and Bowern and Koch (2004: Chapter 2). The main classifications of the second half of the 20th century—O’Grady, Voegelin, and Voegelin (1966), Wurm (1970), Oates and Oates (1970), Oates (1975), Tindale (1974), Dixon (1980), Wurm and Hattori (1981), and Dixon (2002)—build on one another and

lxvii

australian language families and linguistic classifications are based, to some extent, on similar methodologies. This sets them apart from the earlier classifications, such as Schmidt’s (1919a) and others described by Stockigt (Chapters 2 and 6, this volume), on the one hand, and the phylogenetic classifications of Bowern and colleagues on the other. The phylogenetic classifications of Bowern (Bowern and Atkinson 2012, and Bouckaert et al. 2018) for Pama-Nyungan are based on inferring a tree from basic vocabulary from sources from across the country. Details on the methods of phylogenetic classification have been published elsewhere (not least, in the publications themselves, though see Bowern and Atkinson 2012 in particular). In brief, the classification is based on vocabulary, that is, the cognacy (or presumed cognacy) of vocabulary on a 200 item word list. The cognate codings produce approximately 6,500 character sets (as well as approximately double that number of character sets where the word in the language has no known cognates; these are uninformative for subgrouping). We then use an evolutionary framework to model changes in characters over time, which gives probabilistic language classifications, approximate branching times, and (through Bayesian MCMC) an estimate of the confidence of subgroupings. Because the classification is based on 6,500 cognate sets, it is relatively robust to individual data errors. There are four situations where this classification method performs poorly. One is where loan rates are very high. In such cases, classifications may be swamped by similarities due to loans. This is possibly the case with the Bowern and Atkinson (2012) internal classification of Ngumpin-Yapa, for example. The second is where the languages are very similar. In that case, there is little to choose between conflicting hypotheses (that is, if the varieties differ in only a few words, they will clearly group together, but their internal subgrouping may be poorly resolved). The third situation is where the number of overall cognates is very low. For the Western Torres group, for example, two-thirds of the etyma under consideration were either unique or words shared with Papuan languages, which aren’t part of the classification (and so are coded as unique for the purposes of classification within PamaNyungan). In such cases, classification tends to be contingent on very few etyma. The final issue is languages where the only good cognates are those which are also shared by a large number of other languages (such as Western Torres and Anaiwan). In the case of Anaiwan, the classification is poorly resolved, because there are no or few solid innovations by which to group the languages. The language is clearly Pama-Nyungan, but shares small numbers of words with several subgroups. All these problems result in low posterior probabilities. These points are noted in the classification below. Nonetheless, Bowern and Atkinson (2012) recovered all but four of the main subgroups of Pama-Nyungan, as established in the prior literature and classifications such as Wurm’s and Dixon’s. It should be noted that there is a fair amount of agreement between the four classifications discussed here. There is some difference for non-Pama-Nyungan languages, which is to be expected given that, at the time the Wurm (1972) and OVV classifications were developed and published, there were relatively few in-depth studies of linguistic relationships within non-Pama-Nyungan families, and many of the resources relied on here were published after 1975. The classifications differ primarily in the number of languages (see Chapter 7), as well as in the assignment of some individual languages to subgroups, particularly among the Eastern languages. Finally, discussion is needed of Proto-Australian, Pama-Nyungan’s relationship to other language families, and relationship between non-Pama-Nyungan families. I consider ProtoAustralian ‘not proven’. This is not the majority view of Australianists, it must be said: Australianists have presumed that all languages of Australia are ultimately related to one another, even if at too great a time depth to be demonstrable at this point. Another point I consider beyond investigation (at this point) is whether the (putative single) ancestor of

lxviii

australian language families and linguistic classifications all current Australian families (including Tasmanian) is the same as the language(s) of Australia’s original settlers, more than 60,000 years ago. In Australia, as elsewhere, languages change sufficiently fast that they are a crucial source of information for the Holocene, but not for further back in time. Thus I treat the 30 traditional language families of Australia (five from Tasmania, and 25 from the mainland) as distinct entities, at least for the purpose of this classification. Some speculation on language family relationships—between Pama-Nyungan and other families, as well as between non-Pama-Nyungan families—is given in Bowern (2020). Australianists have tended to rely heavily on morphological evidence for demonstrating strengths of relationship between language groups. For example, Harvey and Mailhammer (2017) discuss non-Pama-Nyungan nominal classification markers as evidence for a single Australian family, and Evans (1995c) and Blake (1990b) classify Tangkic as non-PamaNyungan (that is, not a subgroup within Pama-Nyungan) primarily (though not exclusively) on the basis of pronouns and verbal morphology. But as I point out in Bowern (2020), the morphological evidence isn’t absolutely solid. After all, Yolngu pronouns also show some changes from Proto-Pama-Nyungan (e.g. 1sg ŋarra as compared to Proto-Pama-Nyungan *ŋayu, conflation of 2dl and 2pl, replacement of 3pl *thana with walal(a) in some varieties, for example). Verb morphology and lexicon show Yolngu to be solidly Pama-Nyungan, but across the family there are enough morphological differences that on those grounds alone Tangkic is difficult to exclude. The problem deepens further when we consider lexicon. Australianists, as noted by Campbell (2004), tend to distrust lexical arguments for classification because lexicon is susceptible to borrowing. Yet as Bowern et al. (2011) showed, Australian loan rates are comparable to the rest of the world: some languages have a lot of loaned basic vocabulary, while others do not. Almost all languages have some identified loans, but few languages have many loans. Therefore lexical information should be, like all evidence, interpreted with caution, but that caution also applies to morphological data. After all, morphology is also borrowable, especially where language contact is extensive. And the lexicon, because it contains both stable and borrowable items, is a good place to evaluate contact claims. For example, if two languages do not share lexical items which are stable, but they share other vocabulary, that is good evidence that the shared vocabulary is due to language contact. The converse, on the other hand (sharing of basic but not non-basic vocabulary) is potential evidence of shared (if remote) genetic relationship. Some clearly Pama-Nyungan languages nonetheless have few lexical items in common with other languages in different parts of the family. And yet there are non-Pama-Nyungan families with apparent conservative cognates with widely found Pama-Nyungan items. Consider the vocabulary in Nyulnyulan that is shared with Pama-Nyungan languages. Various languages of the Nyulnyulan family have words that are clearly loans from Pama-Nyungan languages. Bardi yagoo ‘brother-in-law’, for example, is a recent loan because it does not undergo either the lenition sound change or the loss of the initial glide that inherited words in Bardi undergo. If this word were truly cognate with *yaku (e.g. Nyangumarta ‘wife’s brother, sister’s husband’, Wardandi yaku ‘wife’, Pitta-Pitta yaku ‘elder sister’, Yulparija yakurti ‘mother’, Mayi-Yapi yakurti ‘mother’, etc.), its form in Bardi should be o /ɔ/ or awoo /awu/. Likewise, Nyikina kampi ‘egg’ is probably a loan, even though there are no diagnostic sound changes to identify it as such, simply because, although it is widespread in Pama-Nyungan, Nyikina is the only one of the Nyulnyulan languages to show it, and it is securely reconstructed within Pama-Nyungan but not Nyulnyulan. Other words are probably loaned into Proto-Nyulnyulan (or at least a stage of the family that predates the breakup of all the languages we see today). Yet others are widespread in Nyulnyulan but are also regional Wanderwoerter, such as lungkurta ‘blue-tongue lizard’ and baarni ‘goanna’ (see Haynie et al. 2014). However, there are yet other words that are

lxix

australian language families and linguistic classifications inherited into Proto-Nyulnyulan, in that they show the expected sound changes in the daughter languages and are well integrated into Nyulnyulan morphology, and yet are similar or identical to words in similar meanings which are widespread and reconstructible in the Pama-Nyungan family, either to Proto-Pama-Nyungan or to an intermediate (but high-level) branch of the family. Some cases from basic vocabulary are given in (1) below. (1)

Proto-Nyulnyulan reconstructions *ni-lirr ‘3sg.poss-mouth’ *ngayu ‘1sg’ *ngamana ‘breast’ *kapali ‘father’s mother’ *ni-marla ‘hand’ (*mara9 ) *tyamu ‘mother’s father’ (*tyamu, *tyami) *kamarta ‘mother’s mother’ (*kamarta) *waalka ‘sun’ (*walngka) *kutyarra ‘two’ (*kutyarra) *ma-kunbira-n ‘urinate’ (*kunpi) *mayi ‘vegetable food’ *ma-ni-n ‘sit’ (*nhi-) *ma-wa-n ‘give’ (*wa-)

Nyulnyulan shares more basic vocabulary with Pama-Nyungan languages than it does with Wororan. There are also loans between Wororan languages and Nyulnyulan languages, as would be expected given the history of the region (cf. Bowern 2018a). However, there are no clear recurrent or systematic similarities between Nyulnyulan and other non-Pama-Nyungan families of the region, as discussed in Bowern (2004b). Now, no one to my knowledge has ever proposed that Nyulnyulan is closely related to Pama-Nyungan, and, to be clear, this is not a claim I’m making here. Grammatically and morphologically, Nyulnyulan languages are not at all similar to Pama-Nyungan languages, though one should also note that they are rather different typologically from other non-Pama-Nyungan Kimberley languages too. I discuss this case not as a claim that Nyulnyulan is Pama-Nyungan, but as a way to evaluate the context of claims for Tangkic relationships, both lexically and grammatically. Nyulnyulan shares as many Pama-Nyungan words as Tangkic does; therefore any investigation of language relationships between PamaNyungan and other families needs to look more broadly. Better criteria are needed for evaluating claims of language relationship when the evidence is sparse.

9

lxx

Though in Ngumpin-Yapa, r > rl in this position; cf. Warlpiri marla (see further McConvell and Laughren 2004).

australian language families and linguistic classifications

Australian Families apart from Pama-Nyungan Bowern

Dixon (2002)

Wurm (1972)

O’Grady, Voegelin, and Voegelin (1966)

Tiwi

Tiwi

Tiwi (Wanuk)

Iwaidjan Amurdak4 Iwaidjic Iwaidja Mawng Garig Ilgar Wurrugu Marrgu

North-west Arnhem Land subgroup Mawung-Iwaydja Mawung Iwaydja Amurdag Marrgu Popham Bay (=Iyi)

Iwaidjan Wargbi Iwaidji Maung Maung Amarag Margu Garik

Iwaidjan Amaragic Iwaidjic Iwaidji Maung Margu

Gunwinyguan5 Gunwinygic Kunbarlang Bininj Kunwok6 Dalabon Jawoyn Warray Uwinymil Wulwulam RembarrngaNgalakgan Rembarrnga Ngalakgan East Arnhem Wubuy Anindilyakwa7 Ngandi

Arnhem Land8 Rembarrnga/Ngalakan Rembarrnga Ngalakan Far east Arnhem Ngandi Nunggubuyu Aninhdhilyagwa Dalabon Gunwinjgu Gunwinjgu Gunbarlang Jarroyn/Warray Jawoyn Warray Uwinjmil

Gunwingguan Gunwinggic Gunwinggu Gunbalang Dangbon Dalabon Ngalagan Ngandi Rainbarngo Djawan Yangman Yangman Wardaman Wageman Warrai

Gunwingguan Gunwinggic Bininy Gunwiggu Gundangbon Muralidban Gunbalang Boun-Dalabon Ngalakgan Ngandi Rembarrnga Jawoyn Yangmanic Yangman Wardaman

Nunggubuyu

Nunggubuyu

Andilyaugwa

Groote Eylandtan

Tiwi2 3

9

Kungarakany 10

Mangarrayi 11

Gungarakanj

Kungarakany

Mangarrayi

Mangarai

Mangarai

Gaagudju

Gaagudju

Kakadju

Kakaduan

Maningrida12 Ndjébbana Nakara Gurr-goni Burarra

Maningrida Burarra Gurrgoni Nakkara Ndjebbana

Gunavidjian Gunavidji13

Gunavidjian (Gunabidji)

Nagaran (Nagara)

Nagaran (Nakara)

Bureran Burera Gorogone

Bureran Burarra-Gunnartpa Gorogone

Wagiman

Wagiman-Wardaman

14

15

Wagiman

lxxi

australian language families and linguistic classifications

Wardaman 16

Alawa

Alawa

Marran Yugul Marra

Marra/Warndarrang Marra Warndarrang

Alawa

Maran Mara Nawariyi Wandarang Alawa

Maran Mara Wandaran

18

19

Warndarrang Wardaman Wardaman Yangman Dagoman17 Garrwan20 Garrwa Waanyi

Waanji/Garrwa subgroup21 Waanji Garrwa

Karawan Karawa Wanyi

Karwan Karawa Wanji

Darwin Region22 Larrakiyan Wulna23 Larrakia Limilngan Umbulgarlan Umbugarla Ngomburr Gonbudj Bukurnidja Ngaduk24

Darwin Region Umbugarla Limilngan-Wulna Limilngan Wuna Larrakiya

Larakian Larakia Wuna

Larakic Larakia Wuna

Jarrakan Gija Miriwoong Gajirrebeng

Kitja/Miriwung subgroup (ND*) Kitja Miriwung

Djeragan Gidjic Gidja Guluwarin Lungga Miriwunic Miriwun Gadjerong

Djeragan Gidjic Gidja Guluwarin Lungga Miriwunic Miriwun Gadjerong

Bunuban Bunuba Gooniyandi

South Kimberley Subgroup (NF*) Bunuba Guniyandi

Bunaban Bunaba Gunian

Bunaban Bunaba Gunian

Wororan25 Wunambalic Kwini Wunambal

North Kimberley Areal Group Wororan (NG) Wunambalic Worrorra Wunambal Ungarinjin Gambre

lxxii

Wororan Wunambalic Wunambal Gambre

australian language families and linguistic classifications

Gamberre Worrorric Unggumi Umiida Unggarranggu Worrorra Yawijibaya Winjarumi Ngarinyinic Worla Ngarinyin Andajin Nyulnyulan26 Western Jawi27 Bardi Nyulnyul JabirrJabirr Nimanburru28 Eastern Yawuru Ngumbarl29 Nyikina Djukun Warrwa

Wunambal

Fitzroy River Subgroup (NE*) Njigina Baardi

[Pama-Nyungan]30

Bagu/Gwini Wororic Worora Mailnga Unggumi Umida Unggarangi Yaudjibara Ngarinyinic Ngarinyin Munumburu Manungu

Bagu/Gwini Wororic Worora Mailnga Unggumi Ungarinyinic Ungarinyin Munumburu Wolyamidi

Nyulnyulan Nyulnyul/Bardi Yawuru Nyigina Warwa

Nyulnyulan Nyulnyul/Bard Jauor Nyigina Warwa

Yanyulan

Yanyulan 31 (Yanyula) Pama-Nyungan

Tangkic33 Lardil Yukulta Kayardild Minkin

Tangkic subgroup (NA*) Lardil Kayardild/Yukulta Kayardild Yukulta Minkin

[Pama-Nyungan]32

Minkinan (Minkin)

Minkinan (Minkin)

Mirndi34 Western Ngaliwuru Nungali Jaminjung Eastern Ngarnka Binbinka Gudanji

NCa Mindi West Mindi

Djingili-Wambayan Djingili Wambayic Wambaya Ngarndji

Tjingilu

Djamindjungan Djamindjung Jilngali Ngaliwuru

Djamindjungan Djamindjung Jilngali Ngaliwuru

Djamindjung/Ngaliwuru Nungali East Mindi Djingulu Ngarnga Wambaya

Tangic Lardil Gayardilt

Tangkic Lardil Gayardilt

Wambaya

lxxiii

australian language families and linguistic classifications

Wambaya Jingulu Southern Daly43 Murrinhpatha44 Ngan’gikurunggurr45

Nungali

Daly River areal group46 Southern Daly Murrinhpatha Ngan.gi-tjemerri

Murinbatan Murinbata

Nungali

Garaman Garaman Murinbata Nangumiri

Western Daly Marramaninyshi Marrithiyel47 Marri Ngarr48 Maranunggu49

Western Daly Emmi Marrithiyel Marri Ngarr

Eastern Daly Kamu Matngele50

Eastern Daly Matngele Kamu

Anson Bay Batjjamalh Pungu Pungu Kiyuk51

Patjtjamalh

Northern Daly Malak Malak Tyerratj

Malak-Malak

Giimbiyu53 Erre Mengerrdji Urningangk 10

Giimbiyu

Daly Moil Ngangikurrungur Brinken Maramanandji Maredan Marengar

Brinken Marithiel Maramanindji

Daly Yunggor-Matngala

Wagaty Wadjiginy52 Maranunggu Ame Mulluk Mullukmulluk Djeraity

Mullukmulluk

Urningangk

Uningank

Mangerian Mengerai

Mangerian Mengerai

See, for example, (Osborne 1974; Wilson 2013). Per Evans (2000a). 12 See further Handelsmann (1991); Mailhammer (2008). 13 See Evans (2003a) and much subsequent work. 14 Bininj Gunwok (or simply Gunwok) includes Kunwinjku, Kune, Mayali, and other varieties, per Evans (2003a). I treat these as varieties rather than listed languages for reasons of consistency with other areas of the country, but a case could be made for including more varieties here. 15 See van Egmond (Chapter 68, this volume) and for more detail on Anindilyakwa’s genetic position, van Egmond (2012) and van Egmond and Baker (2020). 11

lxxiv

australian language families and linguistic classifications 16

Though Dixon (2002) groups many of the languages of Arnhem Land together in a single group, I do not consider there to be sufficient evidence of genetic relationship to do so at this point. Dixon (2002) gives both genetic and ‘areal’ groupings. I stick with genetic relationships as I do not consider Dixon’s groupings either consistent or sufficiently supported by evidence. 17 See Parrish (1983); for this and other isolates of North Australia, see also Bowern (2017a). 18 The position of Mangarrayi is uncertain—as well as considering it an isolate, some have assigned it to Gunwinyguan (e.g. Alpher, Evans, and Harvey 2003), some to Marran (e.g. Merlan 2003). 19 See Harvey (2002). 20 See Carew and Beltran (Chapter 70, this volume). See Green (2003) for possible arguments around the relationships between the Maningrida languages and others in Arnhem Land. 21 This is the same language as Ndjébbana. 22 Included in Gunwinyguan. 23 Included in Gunwinyguan. 24 Alawa and Marran (including Marra, Yukul, and Warndarrang) was proposed in OVV and followed by Heath (1978a) and Sharpe (1976b). Harvey (2012) shows that the relationship between Marra and Warndarrang is not close, and the similarities are most likely due to language contact. Sharpe (1976b) is agnostic about the relationship between Alawa and Marra. I follow Harvey (2012) here in treating these languages as three separate families (two isolates, and Marra related to Yukul, following Harvey 2012 and Baker 2010). 25 Tentative as separate language; could also be considered a variety of Wardaman. 26 Subgroup of Gunwinyguan and listed there. 27 Subgroup of Gunwinyguan and listed there. 28 See Harvey (2009). I consider this ‘not proven’ (see also notes on Tangkic below) and thus conservatively keep Garrwan as a separate group, distinct from Pama-Nyungan for now. 29 Dixon’s (2002) intent is unclear here. Dixon calls this a ‘subgroup’, but does not say which higher-order group it belongs to. In terms of alpha-numeric classification, this group is included with groups which other authors include as Pama-Nyungan; however, given that Dixon (2002) does not accept the unity of Pama-Nyungan, is it unclear on what basis we should assign Garrwan to that grouping. 30 Harvey (2001: 9) considers this group a Sprachbund, and should be considered very tentative. 31 Also known as Wuna (cf. Harvey 2001). 32 Treatment as a distinct language is tentative. 33 Classification for Wororan broadly follows (Clendon 2014; McGregor and Rumsey 2009). That there are three branches is not in doubt, but the divisions within each group are unclear. See also Spronck (Chapter 69, this volume), who includes more varieties. 34 See Bowern (2012a). 35 Jawi is listed as distinct from Bardi. Though the two varieties are now close, there is considerable evidence that the two have partially merged over the 20th century, given extensive language contact on Sunday Island and the great loss of Jawi people, particularly to influenza and diphtheria. That is, the languages were likely much more different, both lexically and morphologically, than the most recent records suggest. 36 Treated as distinct by Bowern (2023) on the basis of material collected by A. R. Peile, which is the only extensive material for the language. The material in Nekes and Worms (1953) for Nimanburru appears to be much more similar to Nyulnyul. 37 Ngumbarl is sometimes said to be a variety of Djukun (cf. Nekes and Worms 1953), but work with Bates’s Ngumbarl material recorded from Billingee (cf. Coyne 2005) suggests that it is sufficiently different from Djukun and Yawuru that it should be treated as a distinct language. 38 See below for Pama-Nyungan groups. 39 In Pama-Nyungan in subsequent classifications; see Yanyuwa (Warluwarric).

lxxv

australian language families and linguistic classifications 40

Tangkic is Pama-Nyungan in this classification but Minkin is non-Pama-Nyungan and a distinct family. Bowern (2020) discusses the variable placement of Tangkic as a sister to Pama-Nyungan or a subgroup within it. While early classifications (e.g. O’Grady, Wurm, and Hale 1966) placed Tangkic as one of the primary subgroups of Pama-Nyungan on primarily typological grounds, since Evans (1990; 1995c) and Blake (1990), it has been customary to treat Tangkic as non-Pama-Nyungan, albeit a close relative (the closest family apart from Garrwan; though see Harvey (2009) for discussion). Evans’s basis for classification was the pronominal system, with the reconstructions of Tangkic forms being rather different from those typically reconstructed for Pama-Nyungan (e.g. by Blake 1990). That in itself might not be evidence against shared genetic relationship however, since, after all, pronouns do change, and other Pama-Nyungan subgroups (e.g. Yolŋu, Karnic) are reconstructed with at least some forms other than those reconstructed to Proto-Pama-Nyungan (Schebeck 2001; Bowern 1998). Bouckaert, Bowern, and Atkinson (2018) include Tangkic among the groups discussed in their phylogeny of Pama-Nyungan. In their tree, Tangkic is a subgroup within Pama-Nyungan, not a sister to the Pama-Nyungan family. This classification is based on sparse lexical cognates, but ones which might a priori be thought to be indicative of shared genetic relationship. We might therefore say that Tangkic is a ‘sparse evidence’ question—that is, there are sufficient differences between the reconstructed pronominal systems to cast doubt on the genetic affiliation. Lexical data is historically distrusted in Australia (see e.g. Alpher 2004a). The best we might say at this point is that the claim is ‘not proven’. BBA (2018) also includes Nguburindi, Yangkaal/Yangarella, and Ganggalida as additional varieties. The placement of Minkin is doubtful; see Evans (1990). 42 See Harvey (2008b). 43 Daly language families are per dalylanguages.org (by Ian Green and Rachel Nordlinger). Please note the language/dialect classification given there: ‘Language names listed vertically are considered by us, as linguists, to be in a dialectal relationship with each other. However, it is important to remember that from a community perspective, all of the varieties listed here are different languages, with those listed vertically just being “close” to each other’. For consistency with the rest of the map I have used the ‘language’ level relationships and listed the first of the names on the map; other names are given in the notes. 44 Also includes Murrinh Kura. 45 Also includes Ngen’giwumirri and Ngan’gimerri, per dalylanguages.org. 46 Dixon (2002) groups all languages of the Daly together as a single family but I follow Green and Nordlinger’s extensive research on the languages in question. 47 This also includes Marri Tjevin, Marri Ammu, and Marri Dan, per dalylanguages.org. One language name is given for reference in the classification but others are included on the language map. 48 Also includes Magati Ke, per dalylanguages.org. 49 Also includes Emmi, Menthe, per dalylanguages.org. 50 Also includes Yunggurr. 51 The status of this language is unclear. The language name is solely known from other groups, who say that it was different from the closest languages to the mainland. It is therefore tentatively included. See further Ford (1990) and Bowern, Chapter 7, this volume, for discussion. Note that Kiyuk is not listed in dalylanguages.org, which is otherwise the main source both classification and mapping of languages of the Daly region (see also Tryon 1974). 52 This is the name of the tribe/group of which Batjjamalh/Bachamal is the language name. Wakac is a Batjjamalh word for ‘beach’, hence ‘Wagaty’ (Ford 1990). 53 See Campbell (2006). 41

lxxvi

australian language families and linguistic classifications

Tasmanian Classifications generally do not discuss Tasmanian along with other Australian languages (though they do include Meryam Mir, which is Papuan, which is something of an inconsistency in my view). See Bowern (2012c) and discussion of sources in that paper for prior classifications. See also Chapter 77, this volume. Tasmanian languages are not classified in Dixon (2002) or O’Grady, Voegelin, and Voegelin (1966); Wurm (1972: 169) provides two classifications, one based on material by Schmidt and Capell; the other attributed to Geoffrey O’Grady and Rhys Jones. The two differ in geographic boundaries and in higher-level classifications. Both recognize the groups listed here; but O’Grady/Jones contrasts Northern vs. the other groups, whereas Schmidt/Capell group Northern with Western. Note that this classification is tentative and simply repeats information published elsewhere. Glottolog 4.3 (Hammarstro¨m et al. 2020) also includes specific classification of Tasmanian languages, based on Bowern 2012c and other sources. Here we include classifications based on published sources but emphasise that they do not take into account the substantial amount of unpublished work completed by the Tasmanian Aboriginal Centre. Bowern (current volume)

Dixon (2002) Wurm (1972) O’Grady, Voegelin, and Voegelin

Southeastern Tasmanian Bruny Island Southeastern mainland Tasmanian

Southeastern

Oyster Bay Oyster Bay Swanport

Mideastern

Northeastern Tasmanian Ben Lomond Northeastern

Northeastern

Northern Tasmanian Port Sorrell Northern Tasmanian

Northern

Western Tasmanian Northwestern Southwestern

Western

Glottolog 4.3 South-Eastern Tasmanian Bruny Island Southestern Tasmanian Hinterland Oyster Bay Big River Little Swanport North-Eastern Tasmanian Ben-Lomond-Cape-Portland NE Tasmanian Dialect Chain

Western Tasmanian Port Sorell Western Coastal Tasmanian

Pama-Nyungan language subgroups Macro groups are given as per Bouckaert, Bowern, and Atkinson (2018), but should be regarded as tentative, as some of these groupings are not well supported in the tree. The ‘Western’ node, for example, is well supported, except for whether Yolngu and Warluwaric are included.

lxxvii

australian language families and linguistic classifications

Bowern (current volume)

Dixon (2002)

Wurm (1972)

O’Grady, Voegelin, and Voegelin

Yolngu Southern Dhuwal Dhay’yi Ritharngu Northern Nhangu Dhangu Djangu Western

Murngic47 Yulngu Gubabuyngu Djambarr-puyngu Riraidjangu Ritarngu Waramiri Dalwongo Yulngo Yarnango

Murngic Yulngu Gobabingo Riraidjangu Ritarungo Wan’guri Dalwongo Yolngo Jarnango Djariwidji

Western Yolngu46 Northern Djinang Djinba Yan-nhangu Djangu Dhangu/Rirratjingu

Golpa Southern Ritharrngu Central Dhay’yi Dhuwal Dhuwala

Djinang Djinba

Djinang Djinba

Jandjinung Yulngi Djinba

Warluwaric48 Yanyuwa Southern Warluwarra Bularnu Wakaya Yindjilandji

Ngarna Yanyuwa Southern Ngarna Wagaya Bularnu Warluwara

Yanyula [non-PamaNyungan]

Yanyula [non-Pama-Nyungan]

Wakaya-Warluwaric Wakaya Warluwara

Warluwaric Warluwara Wakayic Wakaya

Warumungu49

Warumungu

Waramungic (Waramungu)

Warumungic (Warumungu)

Ngayarta Yapurarra Ngarluma Ngarlawangga50 Nhuwala Martuthunira Panyjima Palyku, Nyiyaparli51 Nyamal Kurrama Jurruru Kariyarra Ngarla Yindjibarndi Yinhawangka Kanyara-Mantharta Kanyara Thalanyji Payungu Purduna Pinikura

lxxviii

Gasoyne River to Pilbara Pilbara/Ngayarta Nhuwala Martuthunira Panyjima Yintjiparnrti Ngarluma Kariyarra Tjurruru Palyku Nyamal Ngarla

Mantharta Kanjara Payungu Thalantji

52

Southwest Ngayarda Ngarla Nyamal Bailko Kurrama Kariera Mardudunera Binigura Noala

Kanyara Talandji Bayungu Wadiwangga Mantharda Warienga

Southwest Ngayarda Ngarla Nyamal Bailko Kurrama Kariera Mardudunera Binigura Noala

Kanyara Talandji Bayungu Targari Wadiwangga Mantharda

australian language families and linguistic classifications

Thiin Mantharta Warriyangga Tharrgari Jiwarli Tharrayi

Tenma Djiwali Thargari

Warienga Tenma Djiwali

Kartu Nhanda cluster Malgana Nhanda Amangu53 Wajarri Kaalamaya54 Badimaya Thaagurda55 Yingkarta56

Moore River to Gascoyne Watjarri Watjarri Parti-maya Cheangwa Nana-karti Natingero Witjaari Nhanda Malkana Yingkarta

Kardu Maia Inggarda Malgana Nanda Muliara Wadjeri

Kardu Maia Inggarda Malgana Nanda Muliara Wadjeri

Southwest Ngatju Cluster Ngatyumaya Galaagu57 Mirniny58 Nyungar Cluster59 Southern: Nyunga Wudjari Goreng Minang Southwestern : Bibbulman Wardandi Kaniyang Pinjarup Wiilman Northern: Wajuk Nyaki Nyaki Balardung Nganakarti

Western Bight Mirning Kalaaku Karlamay

Nyungar Watjari Mirniny Mirniny

Nyunga Mirniny Mirning Kalamai

Wati

Wati (Western Desert) Wanman Kardutjara and others

Western Desert (Wati)60 Warnman Northern: Manjiljarra Kukatja Wangkajunga Yulparija Southeastern:

Nyungar

Western Desert language

Wanman Kardutjara and others

lxxix

australian language families and linguistic classifications

Southeastern: Antakirinya Kokatha Kartujarra69 Putijarra Ngalia Ngaanyatjarra Nyaanyatjarra PitjantjatjaraYankunytjatjara Pintupi-Luritja Tjupan Kuwarra Cundeelee Wangka Marrngu70 Karajarri Nyangumarta Mangala

Mangunj Marrngu subgroup Njangumarta Karatjarri Mangala

Marngu Nyangumarda Karadjeri Mangala

Marngu Nyangumarda Karadjeri Mangala

Ngumpin-Yapa71 Ngumpin Walmajarri Bilinarra Karranga Ngarinyman Wurlayi Gurindji Mudburra Jaru Malngin Wanyjirra Jiwarliny Ngardi Yapa Warlmanpa Warlpiri

Northern Desert Fringe Edgar Range to Victoria River Walmatjarri Djaru Gurindji Mudburra Yapa Warlpiri Ngardi Warlmanpa

Ngarga Walmanba Walbiri Ngardi Wanayaga Ngalia72 Ngumbin Mudbura Gurindji Djaru Malngin Ngarinman Bunara Tjiwarliny-Wolmeri Nyangga73 Wirangu Yura subgroup32

Ngarga Walmanba Walbiri Ngardi Wanayaga Ngalia33 Ngumbin Mudbura Gurindji Djaru Malngin Ngarinman Bunara Tjiwarli-Wolmeri Nangga32 Wirangu Yura subgroup32

Northern Western Torres74 Kala Lagaw Ya Kalaw Kawaw Ya

A Torres Strait group75 A1 West Torres A2 East Torres76

Mabuiagic Kaurareg Dauan-Saibai

Maguiagic Kaurareg Dauan-Saibai

Kukatj77

[in Paman]

Kalibamu

Kalibamu

78

Kukatyi Greater Maric79 Dharambalic80 Dharumbal81 Guynmal Mbabaram

lxxx

Lower Burdekin87 Cunningham Gorton O’Connor

Pama-Maric Mari Mandandanji Koa Ilba

Pama-Maric Mari Mandandanji Koa Ilba

australian language families and linguistic classifications

Wulguru Wulguru Coonambella74 Bindal North Maric Gudjal Gugu Badhun Warungu East Maric Giya Ngaro Yuru Biri75 Biri Yilba Yangga Miyan Wirri Barna Yetimarala Gabalbara Garinybal Barada Yambina Wangan Yiman Ganulu Gangulu Gayiri Dhungaloo76 South Maric Wadjabangayi Yandjibara Badjiri Gunggari77 Guwamu Wadjigu Gungabula Bidjara Gunya Margany Yiningay Yagalingu Yirandhali78 Paman82 Northern Paman Linngithigh

Rockhampton/Gladstone Darambal Bayali80 Greater Maric Maric proper Bidjara Biri Warungu Ngaygungu Yirandhali Mbabaram Mbabaram Agwamin Proserpine Ngaro Giya Guwa/Yanda81 Guwa Yanda Kungkari Kungkari Pirriya

B* North Cape York Northern Paman Gudang

Central Pama Oykangand Okunjan

Yara Nawagi Atherton Pama

lxxxi

australian language families and linguistic classifications

Yinwum Yatay Gudang Luthigh Anguthimri91 Ndra'ngith Alngith Aritinngithigh Yupngayth Muthanthi Mpakwithi Mpalityan Mbeiwum Uradhi group: Yadhaykenu Uradhi Atampaya Wuthathi Angkamuthi Middle Paman Ump.-Yinty. Umpithamu92 Yityingka Northeastern Kuuku-Ya’u Kaanju Umpila Kuuku Yani Wik Wik Alkan Wik Paacha Wik Ep Wik Ngathan Wik Me’anh Wik Ngatharr Wik Mungkan Wik Iyanh Kugu-Muminh Ayapathu Pakanh Kugu Nganhcara Alaya-Athima93 Thaypanic Aghu-Tharnggala Kuku Thaypan94 Others Ikarranggal

lxxxii

Uradhi Wuthati Luthigh Yinwum Anguthimri Ngkoth Aritinngithigh Mbiywom Andjingith Umpila Wik Wik-Ngathan Wik-Me’nh Wik-Mungknh Kuku-Muminh Bakanha Ayabadhu Western CYP Upper SW Thaayorre Oykangand Ogh-Undjan Coastal SW Yir-Yoront Koko Bera Kok Thaw(a) Kok Narr Norman Pama Kurtjar Kuthant

Aghu Tharnggala North-eastern Pama Ompila Kandju Koko Ya’o Northern Pama Uradhi Mpalitjan Yinwun Awngthim Alngith Nggoth Aritinngithig Mbeiwum Middle Pama Wik Munkan Wik Muminh Wik Mean Wik Epa Wik Ngatara Bakanha Western Pama Yir Yoront Thaayorr Coastal Pama Koko Pera Gulf Pama Kunggara

Mbabaram Karantic Karanti99 Cairns Djabugay Yidinj

Yidinic Yidin Dyabugay

Kuku-Yalanji

Yalanjic Gugu Yalanji Koko Yimidhir Koko Buyundji Koko Yawa Lamalamic Umbuykamu Umbindhamu Umbuykamu

Umbindhamu Southeast Cape York Lama Morroba-Lama

Tjapukai Idinji Eastern Pama Koko Imudji Muluridji Bay Pama Lamalama Northern Pama Otati Mpalitjan Jinwun Awngthim Nggoth Aritinngithig Mbeiwum Middle Pama Ompila Kandju Wikmunkan Wik Muminh Wikmean Wikepa Wikngatara Taior Western Pama Jir Joront Koko Pera Gulf Pama Kunggara Southern Pama Ogondyan Aghu Tharnggala

australian language families and linguistic classifications

Ogh-Alungul Ogh-Angkula Agu Aloja Koogobatha Takalak Ogunyjan Koko Dhawa Ogh-Awarrangg Kokiny Athima Awu Alaya Southwest Paman RR past group Kuuk Thaayorre Uw-Oykangand Uw-Olgol Uw-Olkola Kuuk-Narr NT past Koko Bera Yir Yoront Yirrk-Mel Norman Pama Kuthant Garandi Kurtjar Ribh Walangama87 Lamalamic Umbuygamu Lamalama

Lama-Lama Rimang-Gudinhma Rimang-Gudinhma Kuku-Wara Bathurst Head Flinders Island Marrett River Guugu Yimidhirr Guugu Yimidhirr Barrow Point Thaypan Kuku-Thaypan Kuku-Mini Takalak Walangama Mbara Kukatj

Wurangung Parimankutinma Wurangung Lamalama Coastal Lamalama Tablelands Lamalama

Walangama ?

Rimanggudinhma 88

Flinders Island89 Barrow Point Guugu-Yimidhirr Mbara Kuuk Yak Agwamin Yalanjic90 Kuku Yalanji Muluriji Gugu Wakura Gugu Djangun Wagaman Yidinyic Yidiny Djabugay

lxxxiii

australian language families and linguistic classifications

Mayi92 Wanamara Ngawun Mayi-Kulan Mayi-Kutuna Mayi-Yapi Mayi-Thakurti

Mayi Ngawun Mayi-Kutuna

Mayapic Maykulan Wanamara Mayapi

Mayapic Maykulan Wanamara Mayapi

Dyirbalic Dyirbal Nyawaygi Warrgamay

Herbert River Dyirbal Warrgamay Nyawaygi Manbara93

Nyawigic Nyawigi Wulgurukaba

94

Kalkatungic95 Yalarnnga Kalkatungu

Kalkatungu Areal group Kalkatungu Yalarnnga

Kalkatungic Kalkatungu

Kalkatungic Kalkatungu

Yalarnngic96 Yalarnnga

Yalarnngic Jalanga

Central Eastern Waka-Gabi Dappil Gureng-Gureng Gabi-Gabi Waga-Waga Bigambal Yugambal Bandjalang Gumbaynggirr

Waka-Kabic Miyan Dungidjau DjakundaKorenggoreng Than Dalla-Batjala Taribeleng Kingkel Wadja Darambal

Waka-Kabic Miyan Keinjan (Dungidjau) DjakundaKorenggoreng Than Dalla-Batjala Taribeleng Kingkel Wadja Darambal

Durubulic Gowar Djendewal Yagara(bal)

Durubulic Gowar Djendewal

Marawari

Muruwari

Wiradjuric Main Wiradjuri Wonggaibon Kamilaroi Yualyai Wiriwiri101

Wiradjuric Main Wiradjuri Wonggaibon Kamilaroi Yualyai Wiriwiri

Dyirbalic Dyirbal Wargamay Bandyin ?

Southeastern Waka-Kabi Wuli Wuli Barunggam Bayali Waka Waka Duungidjawu Batyala Guweng Gubbi Gubbi Gooreng Gooreng Turubulic97 Nunukul Turrubul Yagara Janday Guwar98 Muruwaric99 Muruwari Barranbinya Central-NSW Ngiyambaa Wailwan Wiradjuri Gamilaraay100 Yuwaliyaay Yuwaalaraay

lxxxiv

Yagara Guwar

australian language families and linguistic classifications

Nguri ? Barunggama ? Bigambal110 Bigambal Guyambal Yugambal

Bandjalangic111 Gold Coast Pimpama Southport Burleigh Heads

Miyal Bigumbil

Bandjalangic Gidabal Giabalic Giabal Keinyan

Nguri Barunggam Miyal Bigumbil

Bandjerangic Bandjarang Bandjalangic Giabal Yugumbal

Condamine/Upper-Clarence

Geynyan Gidhabal Dinggabal Galibal Middle Clarence Waalubal Biriin Wudjehbal Lower Richmond Wiyabal Nyangbal Bandjalang Tweed-Albert Minyangbal Yugambeh Nerang Ck Ngahnduwal Copmanhurst Nganyaywana112 Nyanyaywana Southern Anaiwan

Aniwan

Lower Murray113 Yitha Yitha Keramin, Kureinji Ngintait, Ngarkat, YuYu Ngaiawang Ngarrindjeri, Yaraldi Peramangk

Lower Murray Yaralde Ngayawang Yuyu Keramin Yitha-Yitha

Narrinyeric Korni Tanganekald Mirili NgangurukuNgaiawang Ngult Maraura-Ngintait

Narrinyeric Korni Tanganekald Mirili NgangurukuNgaiawang Ngult Maraura

Yuin-Kuric Yuin Dhawa Dharawal Ngarigu Ngunawal

Yuin-Kuric Yuin Thaua Thurawal Ngarigo Ngunawal

Yuin-Kuri N Central New South Wales119 Awabakal/Gadjang Dhanggatti Sydney-Central Coast114 Awabagal Darkinyung Dharug

Kattang Djan-gadi/Nganjaywana Thangatti

lxxxv

australian language families and linguistic classifications

Kuri HRLM115 Awabakal Wanarruwa Lower North Coast

Birrpayi Worimi Katthang Yuin116 Northern Inland Ngun(n)awal Gandangara Northern Coastal Dharawal Dhurga Dharumba117 Southern Inland Ngarigu Omeo language118

Nganjaywana Central Inland NSW Gamilaraay Wiradhurri Ngiyambaa Muruwari Barranbinja O Sydney Dharuk Darkinjung P Southern NSW Southern tablelands Gundungurra Ngarigo NSW Coast Dharawal Dhurga

Gandangara

Kuri Gandangara Darkinung Wanarua Kattang Ngamba (Dangadi) Yukambal120

Kuri Darkinung Wanarua Worimi Ngamba (Dangadi) Jukambal121

Djirringani Thawa

Southern Coastal Thawa Djirringany Gumbaynggiric Gumbaynggir Yaygirr Baanbay

Eastern Victoria Kurnai group122 Kurnai Muk-Thang Nulit Thangguai Bidawal Dhudhuroa123 Waywurru124

(Q) Muk-thang (Kurnai)

Kulin126 Bunganditj Kolakngat127 Wathawurrung Western128 Tjapwurrung Jardwadjali Djadjawurung Mathi129 Ladji Ladji Mathi-Mathi

West Victorian areal group Kulin Wemba-Wemba Wadha-wurrung Wuy-wurrung Buwandik Buwandik Kuurn-Kopan-Noot Kolakngat

lxxxvi

(R) Upper Murray Pallanganmiddang Dhudhuroa

Gumbainggaric Gumbainggar Yegir

Kumbainggaric Kumbainggar

Yaithmathangic Yaithmathang Duduruwa Pallangamiddah

Kurnic Bratauolong125

Kulinic Kulin Wergaia Wudjawuru Woiwuru Narinari Drual Marditjali Gulyan Kurung Kurnic132

Kulinic Kulin Djadjala Wudjawuru Taungurong Drual Marditjali Tjapwurung Gulyan Kurung

Yegir Yegir

Yaitmathangic Jaitmathang

australian language families and linguistic classifications

Yari-Yari Wathi-Wathi122 Wemba WembaWemba123 Nari Nari Warrnambool Wotjobaluk Eastern Woiwurrung Boonwurrung Daungwurrung Yortayortic Yorta Yorta YabulaYabula

Brataolung

Yota/Yabala Yota-Yota Yabala-Yabala

Yotayotic Yotayota Eastern Banygarany

Yotayotic Jotijoti Baraparapa Jabulajabula

Central125 Yardli126

Yalyi

Yalyi 127

Malyangapa Yardliyawarra Wadikali

Nadikali-Malyangapa

Karenggapa/Wadikali

Thura-Yura128 Ngadjuri Nauo Parnkala Nukunu Adnyamathanha Kaurna Narrungga Guyani Wirangu

Spencer Gulf Kadli Yura Parnkalla Adjnjamathanha

Paakantyi129 Kurnu Paaruntyi Pantyikali Paakantyi Southern Paakantyi Parrintyi Wilyakali

Baagandji

Darling Kurnu

Darling Kula/Kurnu

Arandic Kaytetye Aranda Anmatyerr Alyawarr Antekerrepenhe Ayerreyenge Central Ar rernte

Arandic Arrernte Kaytetj

Arandic Artuya Kaititj Urtwa Alyawarra Lower Aranda

Arandic Artuya Kaititj Urtwa Iliaura Lower Aranda

[Southwest] Nangga Wirangu Yura Nawu Pangkala Kuyani Wailpi/ Adhnyamathanha Jadliaura/Nukuna

Wirangu

lxxxvii

australian language families and linguistic classifications

Western Arrarnta Eastern Arrernte Lower Aranda138 Karnic139 Northern Pitta-Pitta Wangkayutyuru Arabana Wangkangurru Central Mayawali Karuwali Mithaka Karangura Western Pirlatapa Thirarri Diyari Ngamini Yarluyandi Yawarrawarrka Yandruwandha Nhirrpi Kungkarric Kungkarri Pirriya Eastern Wangkumara140 Ngantangarra Garlali/Kullilli Punthamara Yandic141 Guwa Yanda 54

Lake Eyre Basin (WA) North and West Pitta-Pitta Wangka-yutjuru Arabana/Wangkangurru Central Yandruwandha Diyari Ngamini Midhaga South-west Wangkumara Galali Badjiri

Pittapittic Ulaolinya Wangkadjera-Pittapitta

Pittapittic Ulaolinya Wangkadjera-Pittapitta

Mitakudic Mitakudi

Mitakudic Mitakudi

Arabanic Arabana

Arabanic Arabana

Yandic Yanda

Yandic Janda

Dieric Karna Dieri Pilatapa Jauraworka Karendala Ngura Punthamara Badjiri142

Dieric Karna Dieri Pilatapa Jauraworka Karendala Ngarna Punthamara Kalali Badjiri56

This grouping is tentative. Central and Southern Yolngu are clear (and group together), but it is not clear whether Northern Yolngu is a group of its own, or rather a set of languages that branch off sequentially from Proto-Yolngu, and therefore do not form a discrete subgroup of their own. Or rather, it is clear that some ‘Northern’ languages group together, but it is not clear that all of them do. BBA (2018), for example, find no single Northern group, but rather a series of splits. Note further that the position of Yolngu, Warluwarric, Kalkatungic, and Arandic are poorly supported in the BBA tree and so the classification of those groups with respect to other Pama-Nyungan groups should be regarded as tentative. 55 Wurm (1972: 149) also gives a classification based on work by Bernard Schebeck, but it’s not clear how the clan lect mapping relates to the classification given here, since Wurm does not make it clear which names he considers equivalent to one another. 56 BBA (2018) strongly supports earlier classifications where Warluwarra and Bularnu are a group, Wakaya and Yindjilandji are a group, and those two groups go together as a sister to Yanyuwa (with the groups of similar time depths).

lxxxviii

australian language families and linguistic classifications 57

Warumungu is usually treated as a family-level isolate within Pama-Nyungan. Both Bowern and Atkinson (2012) and Bouckaert, Bowern, and Atkinson (2018) found that Warumungu was grouped within Ngumpin-Yapa; the possibility of loans influencing the classification means that I treat this as tentative and here use the more conservative classification. 58 Doug Marmion (pers. comm.) comments that classification of this language is very uncertain and should perhaps be in a group on its own with Nyiyaparli. Austlang suggests that Ngarla, Ngarlawangga, and Yinhawangka are either the same language or closely related varieties. Since there appears to be some uncertainty here, I retain distinct names but do so tentatively. Classifications of ‘Yinhawangga’ have varied substantially (between Wati, Ngayarta, and Wajarri group, per Austlang). On the placement of Nyiyaparli and the names Nyiyaparli and Palyku, see also Battin (2019). 59 Equivalent to Nyiyaparli per Austlang, following Dench (1998b,c). 60 Both OVV and Wurm have a large ‘South-west’ group which covers most of the Pama-Nyungan languages of Western and South Australia. It is not exactly equivalent to the ‘Western’ group in BBA (2018), though Wurm and OVV’s South-west, plus Warumungu, is a monophyletic group in BBA. 61 Amangu is sometimes classified as a Noongar variety; it also appears to be an alternative name for Nganakarti (which is clearly Noongar). The grouping within Kartu is per Blevins (2001a). 62 Classification is tentative. 63 It is not clear whether Thaagurda is closely related to Malgana or is a distinct language (cf. Austlang W15). 64 Yingkarta is a sister to Kanyara-Mantharta in Bouckaert, Bowern, and Atkinson (2018). Kartu and Kanyara-Mantharta are not particularly close relatives; Kartu is sister to the South-west languages, while Kanyara-Mantharta is sister to Ngayarta in a ‘Pilbara’ group. Those two clusters form a group. I retain the Kartu classification here pending further research. 65 Some group Galaagu and Marlpa together with Ngatjumaya as a single language (or use Marlpa as the cover term for the language). 66 There seems some disagreement about whether the name is spelled Mirniny or Mirning. I have used Mirniny per the Goldfields Language Centre. 67 Within the Nyungar/Noongar group, it is clear that there are at least three distinct groups. However, it is not clear which varieties go with which, and sources differ on how the varieties are divided. These groups follow Dench (1994); Douglas (1976) also uses three groupings, but not the same three (the differences concern where the South-western languages group). See Chapter 74, this volume, for further discussion. BBA (2018) used ten word lists across the south-west and recovered the three-way Natju-Mirniny-Nyungar grouping. Within Nyungar/Noongar, subgroupings were less clear, in part because south-western was over-represented compared to the other varieties. Amangu appears in Noongar classifications but per Blevins (2001a) is classified as Kartu. Amangu is also an alternative name for Nganakarti, a Noongar variety. Language locations and named varieties are based on the Noongar learner’s guide, using their map Noongar boodja wongki – Noongar dialect map (Noongar waangkiny 2014), but have been adapted slightly given other information from surrounding regions (including the relationship between the names Amangu and Nganakarti, for example). 68 See Babinski et al., Chapter 75, this volume, for a note on the subgroup name. There is more structure in the group than is presented here, but as with Nyungar/Noongar above, sources disagree. There is, for example, agreement that the northern Wati languages are different from the south-eastern ones. Others have discussed Western Desert as a dialect chain; sources also agree that Warnman is different from the other languages in the group. The Goldfields Language Centre gives four regions for Western Desert varieties. Babinski et al. (Chapter 75, this volume) discuss data from 11 varieties and show how subgrouping within the Wati group is difficult. Language placement for the map should be considered approximate, since sources provided conflicting information. 69 The remaining Wati/Western Desert varieties are not yet classified. 70 See Weber (2009). 71 See McConvell and Laughren (2004). 72 Classified as Western Desert in Bowern. 73 See Thura-Yura in the Central group. 74 See Alpher et al. (2008) for discussion surrounding views over whether the Western Torres Strait group is Pama-Nyungan. 75 Dixon (2002): ‘These are Papuan languages, not closely related to each other. A1 has a significant Australian substratum’. 76 Listed below in ‘Other’ languages in Bowern’s classification, since per Piper (1989) and Alpher et al. (2008), Eastern Torres is a Papuan language related to other Trans Fly languages, and Western Torres is Pama-Nyungan. In Bouckaert, Bowern, and Atkinson (2018) and Bowern and Atkinson (2012), Western Torres’s closest relative is Kukatj (see below), but the lexical evidence is very slim, so that is not followed here.

lxxxix

australian language families and linguistic classifications

77

Breen’s (1976: 151) contrastive study of Gog-Nar and Gugadj has some discussion of classification. Wurm (1972) places Kukatj in a group on its own; Breen suggests it is possibly part of Pama-Maric (based on lexical comparison); he suggests that its closest neighbours in terms of classification are not the Norman Paman languages which are geographically closest. 78 Said in Wurm (1972) to be of doubtful existence, along with Walangama and Karanti. See Breen (1976) for other discussion. 79 The comments made above for Western Desert/Wati and Nyungar/Noongar about complexities of classification also apply here. For the term ‘Greater Maric’, see Barrett (2005), though my composition of ‘Greater Maric’ is not the same as his. He includes Pirriya, Kungkarri, and Guwa as ‘Greater Maric’, whereas I class them within an expanded Karnic group. Barrett’s (2005) classification of Maric has three groups: North, East, and South; but there are numerous other varieties which are not included in his classification, particularly in the north-west of the Maric area where data are sparse. Note that both OVV and Wurm (1972) only include three Maric languages (though with other dialects listed), whereas even Dixon (2002), who normally ‘lumps’ rather than ‘splits’, includes five in his ‘Maric proper’ and eight others in other affiliated groups. Sutton (1973: 12) includes 34 Maric languages. BBA has South and East Maric as a group, but North Maric groups with Dyirbalic and the (previously unnamed) group including Bindal, Wulguru, and Coonambella. I consider this as tentative and flag this part of the classification as needing further clarification and detailed work with all available sources, and I note that even in this area, linguists differ about the number of varieties according to a ‘linguistic’ classification. Note that Bowern’s classification includes many more Maric varieties than previous classifications, part of the cause of differences in the number of languages recognized overall (see Chapter 7, this volume, for some discussion). 80 The grouping of Dharumbal within Maric is tentative. As Terrill (2002) notes, Dharumbal appears to have a number of Pama-Nyungan retentions which are not otherwise found in Maric languages, implying it should be a primary group within Pama-Nyungan, rather than a group within Maric. 81 According to Terrill (2002), the language of Great Keppel Island, Wapabara, has considerable differences from Dharumbal. It could therefore be tentatively considered a distinct language. 82 Austlang quotes Dixon (2002) and Donohue (2007) as treating Coonambella as a dialect of Wulguru; however, there appear to be sufficient lexical differences in the sources to treat them as different languages here. 83 I am agnostic at present on the number of languages here. Clearly the dialectology of this area was complex. Breen (2009) includes the ‘Biri dialects’ within a single language, but treats the rest as distinct (though with similar caveats to those given here). 84 This name Dungaloo or Dhungaloo is known only, as far as I know, from a word list of 76 items by Hatfield at AIATSIS (PMS 4902). It is clearly a Maric language, and is here tentatively grouped with Biri and varieties. It is, however, sufficiently distinct to be included here, at least tentatively. 85 There is doubt about whether Mandandanyi is a distinct language (see further Chapter 7, this volume). It is used as a language name in Barrett (2005), and is listed in Tindale (1974); Tindale gives Kogai as a language name and Mandandanji, Kunggari, and Barrungam as group names. However, Barunggam (per Kite and Wurm 2004 and earlier Holmer 1983) spoke Waka-Waka. Given this uncertainty, I retain Mandandanyi and Gunggari as distinct (though tentatively). Note that there are two language varieties known as ‘gunggari’ (which is a word for ‘east’ in part of Queensland; Gunggari is therefore most likely an exonym). The language spelled Kungkarri (more closely related to Karnic) here is clearly distinct from the Maric language known as Gunggari. Map information for this area is based primarily on Breen (2009). Breen uses different spellings of several varieties (e.g. Barana for Barna); I retain the more commonly used ones. 86 Beale (1975) also includes a language Maŋira with Yirandhali. The name does not appear in Austlang; presumably it is the same as the ‘Mungerra’ source for Yirandhali given in Curr (1886). See also Breen (2009). 87 Bowern finds no concrete evidence for a distinct Burdekin group. The three word lists in Curr (1886), from which Dixon derives the three languages of the group, appear to be Yuru (see Terrill 1998), a Maric language, possibly including words from Bigambal. 88 Bowern places Bayali in Waka-Kabi. See also Breen (2009). 89 This is a separate group within Karnic in Bowern’s classification. 90 See Black (1980); Hale (1976c,d,e); Hale (1964). Thanks to J. C. Verstraete, Peter Sutton, and Barry Alpher for discussion of language placements and classification for Paman. 91 Same as Awngthim. 92 For Middle Paman, see Verstraete and Rigsby (2015) and Verstraete (2020) for evidence. The grouping of Umpithamu and Yintyingka is given in Verstraete and Rigsby (2015: 192–4), they also include Umpila with many of the changes that characterize Yintyingka and Umpithamu within Middle Paman.

xc

australian language families and linguistic classifications 93

Gugu Mini was included in an earlier version of this classification but is not included here since it appears to be a cover term for a number of groups, including Kokiny and Athima. 94 Also known as Awu Alaya. 95 Placement within the group is unclear. 96 There is also material known as Kuku-Warra (lit. ‘bad language’) but this appears to be an exonym for the languages of the Princess Charlotte Bay region (Barry Alpher pers. comm.). 97 Also known as Oko Wurrima. 98 Note that I have fewer language distinctions than the Pama Language Centre, who also includes Kuku Jakandji and Wulpuru. 99 Kuthant in Bowern’s classification, following Black’s work on Norman Paman. 100 See Breen (1981b). 101 This term is equivalent to Wulguru, which is probably a cover term for several varieties spoken on Palm Island, Cleveland Bay, and Magnetic Island (per Austlang). 102 OVV place Dyirbalic within the Paman subgroup. 103 See Blake (1979a). 104 ‘The language is structurally similar to Kalkatungu which constitutes the Kalkatungic Group’ (Wurm 1972: 131). 105 Per information provided by the Yugara-Yugarapul Aboriginal Corporation, they consider there to be a single Yuggara language, spoken by groups who identify as Miguntyun and Chepara. Gowar (Moreton Island language) is considered separate, and Turrubul is possibly based on a misnaming in the 1850s. Austlang gives Yuggera/Yagara as the cover term for three languages: Moondjan (spoken by Nunukul people); Jandai (spoken by Goenpul people); and a third variety spoken by Turrbul people. However, Austlang also gives an (unsourced) alternative: Yuggera and Koopenul; Nunukal and Goenpul; and Ngugi (Moreton Island). These groups are together known as Quandamooka. 106 Jeffries (2011) treats Guwar as a Bandjalangic language which has been heavily influenced by Yagara. 107 See Oates (1988), and (Oates 1988: 198–9) for the classification of Barranbinya as closer to Muruwari than Central New South Wales. 108 WL (2008: 215) has Gamilaraay and five other varieties as a single language, called ‘Darling Tributaries language’. 109 Wiriwiri, Nguri, and Barunggama are Waka-Kabi in the current classification. See Breen (2009) and Terrill (2002) for discussion. 110 See Bowern, Chapter 7, this volume, and Wafer and Lissarrague (2008: 333–4) for some discussion of the variation in classification of this subgroup. These languages have been variously classified as Yuin-Kuri, Central NSW, Waka-Kabi, and Bandjalangic. BBA shows that they are a distinct subgroup. 111 See Sharpe (1985; 2005; Chapter 73, this volume). The classification of this dialect chain is complex. Crowley (1978: 158, 196) gives maps of the Bandjalang languages which are redrawn and adapted by Sharpe (1985: 103). This classification is followed by WL (2008: 352ff.). Note that the two maps in Sharpe (1985) are not reconcilable to one another, as the shape of the area given as Bandjalangic differs from one map to the other. The dialect labelled Galibal is bisected by the line that divides the Upper Clarence from Middle Clarence. I use the dialect placements from Map 2 imposed on the classification of Map 3, as per WL and Crowley (1978). Shaun Davies (at the 2022 Australian Languages Workshop) provided good evidence that prior classifications have over differentiated. 112 I follow Wafer and Lissarrague (2008) and Crowley (1976) in recognizing two languages here. 113 Classification based on the materials in Horgen (2004) and Wafer and Lissarrague (2008). Blake (2011b), Dixon (2002), Wafer and Lissarrague (2008), and Horgen (2004) have five Lower Murray Languages. I also include Peramangk. This name is not mentioned by Wafer and Lissarrague and Horgen. Austlang says that classification is uncertain (AIATSIS code S5). Note that Horgen’s (2004) map is misleading, since it labels the area coloured as Yitha Yitha as Mathi Mathi. I take location information primarily from Wafer and Lissarrague (2008) rather than Blake (2011b). Blake places Mathi Mathi on the Lachlan River and Yitha Yitha around Mildura, while Wafer and Lissarrague (2008) has Yitha Yitha on the Lachlan River and Yari Yari (a Mathi language) near Mildura. This also accords more closely with Horgen (2004). I give several dialect names for each language so as not to privilege one name above another. 114 Thanggatti is grouped with the Nganyaywana languages in Wafer and Lissarrague (2008). However, in the BBA (lexical) classification, Thanggatti is grouped solidly within Yuin-Kuri, while Nganyaywana did not clearly group with any particular subgroup or language. For now I retain grouping of Thanggatti within Yuin-Kuri, pending further research. Per members of the Dharug Dhalang project, comparative research of the archival sources and of more contemporary linguistic writings shows that varieties termed Dharug/Dharuk, Eora/Yura, ‘Sydney Language’ (Troy 1993), and Steele (2005) represent data of one language, named here ‘Dharug’, with some dialectal variations. 115 That is, Hunter River-Lake Macquarie, per WL. 116 I follow Wafer and Lissarrague (2008: 105) and Koch (pers. comm.) in using a four-way distinction between inland and coastal, and northern and southern to describe this complex dialect chain. 117 This is the language of the group known as Wandandian (cf. Eades 1976: 4).

xci

australian language families and linguistic classifications 118

See Wafer and Lissarrague (2008: 67) for the uncertainty surrounding the names and classifications of the Omeo vocabularies. I follow Wafer and Lissarrague in using the name ‘Omeo language’ and Koch (pers. comm. to Wafer and Lissarrague) in classifying it as Yuin. 119 Note that the name Central NSW here in Dixon’s formulation does not refer to the same languages as the Central New South Wales group established in other classifications. 120 Treated with Bigambal as own primary subgroup by Bowern. 121 Treated with Bigambal as own primary subgroup by Bowern. 122 Classification per Hercus (1987) and Clark (2005); see also Fesl (1985). 123 Map placement of these languages is uncertain. Blake (2011b) clearly treats both Pallanganmiddang (Waywurru here) and Dhudhuroa as only extending as far as the Murray River. Bowe (2002: 134–8) says that Waywurru extended north of the Murray River (‘east of Albury’). See Wafer and Lissarrague (2008: 64–7) for issues around the naming of Dhudhuroa, Waywurru, Pallanganmiddang, and Yaitmathang. 124 This is the name of the language spoken by the Pallanganmiddang people. 125 Kulin in Wurm (1972) 126 See (Blake 2011a; Blake and Reid 1994). Note that the Victorian Aboriginal Corporation for Languages (VACL) has a somewhat different classification, organized partly by geography. See also Eira (Chapter 74, this volume). 127 Also called Gulidjan. 128 Werkaya is another name also used for these languages. 129 See Blake et al. (2011) for discussion. Hercus (1992b) gives three groupings of dialects for this area. Wafer and Lissarrague (2008) have different language placements in some areas. For example, Hercus (1992b) is clear that Wadi-Wadi is downstream of Swan Hill, whereas Wafer and Lissarrague’s (2008: 63) map places the name upstream. Hercus’s (1986: 5) map of Baagandji (Paakantyi) and neighbours has Southern Paakantyi country extending as far south as Kureinji (that is, almost to the Murray River), whereas Horgen (2004) extends Mathi Mathi into that area. I have followed Hercus here, except that I also follow Blake et al. (2011) in extending Mathi territory north of the Murray. 130 There are two different Wathi-Wathi varieties: Swan Hill and Piangil, cf. in Blake et al. (2011). 131 Also Baraba-Baraba. 132 In East Victoria in Bowern’s classification. 133 Note that this is not a monophyletic group in BBA (2018) but is rather a set of subgroups which are not a stable phylogenetic group. However, they do share some features; for example, all the central subgroups have lost verb conjugation classes (cf. Brody 2020), though it is not clear whether this is a shared innovation, since other languages in the family have also lost them. 134 See Hercus and Austin (2004). While many language placements are (at least approximately) according to Wafer and Lissarrague (2008), some placements were irreconcilable with other information (for example, concerning the placement of Wadikali and Pirlatapa). 135 Nadikali is presumably a typographical error for Wadikali. 136 See Simpson and Hercus (2004). 137 Wafer and Lissarrague’s (2008) map of Paakantyi and Darling River varieties varies substantially from the map in Hercus (1986: 5). The locations given on this map are closer to Wafer and Lissarrague’s in most (but not all) respects. Hercus (1986: 8ff.) gives five main divisions between varieties, which would seem to correspond to closely related languages (or different dialects). Wafer and Lissarrague (2008: Chapter 10) give two main divisions. I have probably over-differentiated varieties here. 138 See Hale (1962): classified as a branch coordinate with other ‘Aranda’ (i.e. all except Kaytetye); see also Koch (Koch 1997a,b; Koch 2004b). 139 For the difficulties of classifying Karnic, see Breen (2007), responding to Bowern (2001). 140 Wangkumarra ‘southerners’ is both a specific name and a more general name for the people of this area. The geographical placement of Wangkumarra here is subject to some uncertainty. Hercus (1986) says that the Wangkumarra, Punthamarra, and Kungadityu were the northern neighbours of Paakantyi people. 141 See Breen (1990b). 142 Classified as Maric by Bowern.

xcii

australian language families and linguistic classifications

Papuan Language Meryam Mir (Piper 1989); Eastern Trans-Fly (Papuan).

New Indigenous Languages143 • Creole Languages • Yumplatok and affiliated north-eastern creole languages 1. Yumplatok; also known in the literature as Torres Strait Creole, Broken (Shnukal 1988; 1991) 2. Cape York Creole; from the Northern Peninsula Area of Cape York (Crowley and Rigsby 1979, with some commentary in Harper 2001) 3. Lockhart River Creole (Mittag 2016) 4. Napranum Creole (Carter, Angelo, and Hudson 2019) • Kriol and the Kriol sphere of influence144 5. Kriol (Schultze-Berndt et al. 2013; 5a. Fitzroy Valley Kriol: Hudson 1985; 5b. Bamyili/Barunga and Ngukurr Kriol: Sandefur 1979; Barkly: Graber 1988) Eastern periphery 6. Mornington Island Creole (Nancarrow 2014; Community negotiated project, Language Perspectives 2015a) 7. Kowanyama Creole (Community negotiated project, Language Perspectives 2018) Southern periphery 8. Alyawarr English (potentially also described as a blended language; Dixon 2017; 2018) 9. Wumpurrarni English (potentially also described as a blended language; Disbray and Simpson 2004; Disbray 2008a; 2016) • Superdiverse Aboriginal settlements145 10. Yarrie Lingo (Angelo Fraser and Yeatman 2019; Mushin, Angelo, and Munro 2016; Community-negotiated project, Language Perspectives 2009) 11. Woorie Talk (Munro and Mushin 2016; Community-negotiated project, Language Perspectives 2015b) 12. Cherbourg Lingo (Mushin, Angelo, and Munro 2016; Community-negotiated project, Language Perspectives 2009) 13. Palm Island (Dutton 1964; 1965) 14. Murdi Language (south-western Queensland; Community-negotiated projects, Language Perspectives 2009) 143 This classification is reproduced (with additions) from Angelo, D., O’Shannessy, C., Simpson, J., Kral. I., Smith, H., and Browne, E. (2019). Well-being & Indigenous Language Ecologies (WILE): A strengths-based approach. Literature Review for the National Indigenous Languages Report, Pillar 2. Canberra: The Australian National University. doi 10.25911/5dd50865580ea. Available from http://hdl.handle.net/1885/186414: Appendix 2, pp. 115–16. There are no doubt other contact languages which are not in this classification, but which are yet to be documented. 144 See also Dickson (Chapter 57, this volume) and Angelo (Chapter 66, this volume). 145 See Angelo (Chapter 66, this volume) for more information.

xciii

australian language families and linguistic classifications • Mixed languages 15. Light Warlpiri (O’Shannessy 2005; 2009; 2013, and much other work) 16. Gurindji Kriol (McConvell 1988a, 2002; McConvell and Meakins 2005; Meakins 2008b; 2010, and much other work) 17. Modern Tiwi (Lee 1987; McConvell 2010; Wilson, Hurst, and Wigglesworth 2018) • Languages not otherwise classified 18. 19. 20. 21. 22. 23. 24. 25.

Dhuwaya (Amery 1985) Areyonga Teenage Pitjantjatjara (Langlois 2004) Broome Pearling Lugger Pidgin (Hosokawa 1995) Broome Aboriginal Englishes (Hosokawa 1995)146 Pidgin Ngarluma (Dench 1998a) Contemporary/New Noongar (Douglas 1968 [1976]; Rooney 2011) Jambun English (Schmidt 1985) Palawa Kani

Sign Languages After much consideration, Sign Languages are not further differentiated in this classification. This is because it is currently very unclear how many Sign Languages are and were present in Aboriginal Australia, and how best to delineate them. See Green, Chapter 52, this volume, for discussion.

Acknowledgements Numerous people have assisted with this classification in various ways over almost 15 years. In particular, I thank Barry Alpher, Peter Sutton, and J. C. Verstraete for discussion of Paman; Barry Alpher for discussion on the relationship between Paman and Maric; David Nash and Harold Koch for innumerable points; Denise Angelo, Nick Evans, Jenny Green, Mark Harvey, Luise Hercus, Amanda Lissarrague, Patrick McConvell, Doug Marmion, Rachel Nordlinger, Kazuko Obata, Erich Round, and Jane Simpson. They are not responsible for any errors that remain here, and in some cases we have no doubt come to different conclusions and would not necessarily agree with the classification and maps presented here.

146 Aboriginal English is shown on the map in North-Western Australia but it should be noted that there are many varieties of Aboriginal English.

xciv

Foreword It is with a great sense of humility and pride that I write the foreword for this important work. The survival and continuation of Australian languages, the Aboriginal and Torres Strait Islander languages of Australia, is a matter of global significance. In my lifetime I have personally witnessed the agonizing death of my language family and my language group, simply through the forces of interaction and engagement with modern society, represented by the colonial settler society of Australia. In my language community, Ngalia, my mother and her parents were the first generation to have contact with white Australia. As each of my elders passed, I witnessed the progressive disappearance of the language of my childhood—language that carried all the meanings of my connection to country, language that held the wisdom and insight of countless generations before me, language that allowed us to flourish in a hostile environment. Our language united us through family, kinship, and knowledge of country. It enabled us to communicate not just to survive but to flourish as a society. It is a long, slow, and painful journey with each passing, as the realization dawns that you can no longer speak to those around you, you can no longer share your understandings and insights of the world around you in the same language. Instead you take on the words, ideas, and mannerisms of the languages spoken around you. Sometimes this means speaking other Aboriginal languages, but more commonly in my region it’s the adoption of English, or the Aboriginal English dialect, as the primary means of communication. When you arrive at that accursed destination of being the last speaker, you can never truly express the joy, the pleasure, of hearing and speaking your mother tongue again. Every sound triggers the synapses in your brain. Pleasure courses through your body as you hear the words of your mother tongue. They roll off your tongue, bringing a sensation of immense relief at speaking and connecting with another soul. This is usually the pleasure of a rare and temporary interaction with a linguist or language worker, coming to frame your language into the archive of the world’s lost treasures. It’s not ideal, but it’s an important contribution. This significant work focussing on Australian Indigenous languages carries the stories of individuals, families, and countries and the histories of spirit, art, and culture. The history of First Nations languages in Australia is mired in genocide, soaked with the blood of the frontier. It survives under the hammer of in-built bias and prejudice and the socially perpetuated oppression of language communities. The numbers speak for themselves: multiple hundreds of languages were spoken prior to contact—some say 200, others up to 700—and yet today only 20 remain as living languages. With the passing of each year, that number moves closer to the precipice. This work commits to record an insight into those languages once spoken and those that still survive today. I hope it will inspire the survival of Australia’s First Nations language communities into the future. Pick it up, read it. Wonder, be inspired, and treasure each word. Palunya. Ngula nyaku. Kado Muir Ngalia language speaker

xcv

chapter 1

Introduction Claire Bowern

1.1 Introduction: the languages of Australia This is a book about the Indigenous languages of Australia: the more than 500 languages, including Sign languages, Pidgin, Creole, and Mixed languages, which have been used on the Australian mainland, Tasmania, and the Torres Strait Islands from more than 50,000 years ago to the present.1 At the time of the 1788 European invasion, one language family—Pama-Nyungan—covered roughly 90% of the Australian mainland. Twenty-seven other language families, now known collectively as non-Pama-Nyungan, were found in the north, from the Kimberley region in the northwest to the Gulf of Carpentaria. There were also probably five Tasmanian language families, containing multiple languages. This linguistic diversity accrued over thousands of years. The contemporary linguistic landscape of Australia is different. Many of the languages which were in use at the time of European invasion and settlement are no longer spoken. What happened to the Aboriginal people of Australia in the years following 1788 was genocide. Even a purely academic approach to the linguistics of Aboriginal Australia cannot overlook this—it is embedded in the circumstances of documentation, in the types of examples recorded, and in who shapes the record. The Indigenous peoples of Australia are still here, and that is also celebrated and reflected in this volume. This book grew out of a wish to provide an up-to-date compendium of knowledge on the languages of Australia, particularly including material not covered in the handbooks from the past. To date, there have been several single-authored handbooks of Australian languages (Schmidt 1919a; Capell 1956; Blake 1987a; Dixon 1980; 2002; Wurm 1972), as well as several edited volumes comprising collections of papers about different aspects of language (Koch and Nordlinger 2014; Dixon 1976a; Walsh and 1 Traditional figures place the number at 250; see Walsh (1997b) and Bowern (Chapter 7, this volume) for discussion about the number of languages and why the higher figure is more accurate.

Yallop 1993). Because of the need to broaden focus on the ‘langscape’ (see Angelo, Chapter 66, this volume) as well as questions of linguistic structure, this handbook has not only chapters on typological topics in phonology, morphology, and syntax, but also a substantial number of chapters on semantics and pragmatics, as well as on sociolinguistic topics, on language use, and on linguistic ecology. We also present work on language change across the continent, as well as on the history of research and classification. Additionally, there are chapters with discussion of language areas, families, and selected individual languages. In this way, the handbook both represents the state of the art on Australian language work and reflects the concerns of linguists working on Australian languages.

1.2 Overview of the volume Given the number of chapters in the handbook, they are not all summarized here. Instead, I first summarize the sections, and then briefly highlight some of the common themes that authors raise.

1.2.1 Summary of parts The first part (‘Background’) provides information about the history of research, classification and language history, and documentation and philological methods. These chapters present information about the forces that have shaped contemporary documentation practices for Australian languages, as well as an overview of the landscape of languages to be discussed in subsequent sections. Part 2 concerns typology and grammatical structure. The chapters cover phonetics, phonology, morphology, syntax, semantics, pragmatics, and discourse. These chapters build on the long empirical tradition of Australian language documentation and grammar writing to provide a state of the art overview of these aspects of language. These chapters

Claire Bowern, Introduction. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Claire Bowern (2023). DOI: 10.1093/oso/9780198824978.003.0001

1

claire bowern contain a variety of case studies and typological approaches, depending on the availability of material. Part 3 is devoted to sociolinguistics and language variation. It includes an array of topics related to language use, including code-switching and multilingualism, childdirected speech (also known as ‘baby talk’) and auxiliary registers, as well as language contact and contact varieties (including Kriol). The five chapters in Part 4 involve language in the community—that is, applied linguistic topics around language policies, language acquisition, and language revival and reclamation. Finally, Part 5 provides additional case studies of many of the themes brought up in previous chapters by discussing individual languages, language areas, and language families. These areas were picked to illustrate particular points in the handbook, as well as to highlight a few of the languages or regions which lack good overviews in other handbooks.

1.2.2 Coverage of topics Several documentation themes run through the volume. Most authors mention at some point in their chapter that they did not have enough data, or that relevant facts had not been recorded for enough languages. Almost every author points out that some aspect of their topic is understudied— even the topics that have been well researched by Australianist standards. In compiling the chapters for this volume, my aim was to cover ‘Australian (Indigenous) languages’ as broadly as possible. This meant including not just the languages which were spoken prior to the European invasion, but also Indigenous languages in current use, such as Kriol and other contact varieties. A further aim was to not focus simply on grammatical topics—that is, the study of languages of Indigenous Australia involves a lot more than typological and historical study, important though that is. Language variation and sociolinguistics, for example, has been a chronically understudied area of Aboriginal linguistics. It is my hope, that by including in a single book work on language policy, language acquisition, language revival, language variation, code-switching, signed languages, auxiliary languages, and similar topics, along with work on phonology, morphology, syntax, and semantics, that the connections between grammatical description and linguistics with a social focus will be more apparent. For all that breadth, there were a number of topics that I would have liked to have represented in this handbook, but which could not be included. This was either due to lack of information, or in some cases, due to lack of space.

2

For example, there is no chapter on syntactic change because there has not been sufficient reconstruction already done to allow us to synthesize results. There are very few publications on comparative syntactic reconstruction, and the synchronic typology of syntax has focussed on a couple of areas—in particular, adjoined relative clauses, nominal classification systems, and complex predication. Not even nonconfigurationality, which has long been associated with Australian languages (Austin and Bresnan 1996; Hale 1983; Simpson 1991; amongst others), has received a thorough comparative treatment within Australia. Some topics are addressed in passing but would warrant treatment in more detail, such as clusivity (see Gaby and Shoulson, Chapter 24, this volume, in the broader context of pronominals). In other cases, we have tried to present a more nuanced view of ‘facts’ (or perhaps more accurately, ‘factoids’) about Australian languages—that is, aspects of the language which have been treated as received wisdom but are either factually incorrect, trivially true of human language in general, or require more discussion and explanation (see Bowern 2017b for further discussion of this topic). One example involves adjectives. ‘Received wisdom’ is that Australian languages typically lack a robust distinction between nouns and adjectives. However, as Kim (Chapter 25, this volume) shows, half the languages in her survey have some type of morphological or syntactic distinction between (referential) nominals and adjectives (see also Louagie 2017b). Therefore, we cannot simply maintain an opinion that Australian languages do not have morphological or syntactic distinction between nouns and adjectives (cf. Dixon 2002: 67), when so many clearly do; and where for others, a lack of distinction has been asserted without detailed investigation. In yet other cases, chapters in this handbook are the first in their respective topic to attempt to synthesize work across the country. For example, suppletion in Australian languages is severely understudied (see Bach and Round, Chapter 29, this volume). In further cases, chapters of this handbook provide some overview on certain aspects of language, but fuller treatments of individual languages await. Formal semantics and negation, for example, lack detailed treatments for most languages of the continent, and these chapters (Bednall, Chapter 32, and Phillips, Chapter 34, for example) are a first general survey that will form a foundation for others to build on. Unfortunately, several topics needed to be omitted, mostly due to lack of space. Associated motion (Koch 1984; 2021), for example, is a phenomenon that is robustly attested in several groups of Australian languages but relatively rare (though not unknown) elsewhere in the world (cf. Guillaume and Koch 2021). An ideal volume without page constraints would have included more work on lexicography,

introduction grammar writing, and lexical semantics beyond a few key areas, such as kinship (McConvell, Chapter 43), toponyms (Rosenberg, Simpson and Bowern, Chapter 44), and language names (Rosenberg and Bowern, Chapter 49). Also missing, though not by design, are Aboriginal voices around their language work (see Woods and Gaby 2021, for some discussion of this topic). While Miceli and Bowern (Chapter 5) discuss a few ways in which linguistics is part of multidisciplinary approaches to the past, we have not done justice to the many ways in which linguistics and anthropology have been combined over the history of work on Australian languages (cf. Stockigt 2020). Some other areas are ripe for further analysis and need a fuller study. We have no information about gesture or reference construction where gesture is part of the utterance (though cf. Green 2014). For example, Dahmen (2020) found, in a small-scale study of Jaru, that half the pronouns in the conversational corpus are accompanied by gestures, clearly indicating that gesture and speech are jointly signalling reference. Thus, even though this book is large, it is not, by any means, a fully ‘comprehensive’ study.

some analyses are based on primary work with individuals, while others are based on secondary analyses or archival materials, such as fieldnotes. Many of these materials are archived in three repositories: the library of the Australian Institute of Aboriginal and Torres Strait Islander Studies (AIATSIS), the Endangered Language Archive (ELAR) repository through SOAS, and the PARADISEC digital archive.

1.2.3 Authors’ approach

One consideration has been the extent to which a book like this should use the conventions that are most familiar to Australianists, vs. situating the languages in the broader typological literature. For example, for many years, Australianists spoke of three types of Australian languages (that is, in typological terms): prefixing, suffixing, and ‘affixtransferring’ (Capell 1972). The prefixing languages are the non-Pama-Nyungan languages in the north of the country; the suffixing languages are all Pama-Nyungan, with the exception of the Tangkic, Garrwan, and Eastern Mirndi families. The affix-transferring languages are mostly from the Ngumpin-Yapa and Marrngu subgroups of Pama-Nyungan; they have pronominal argument morphology that usually attaches to a second position clitic (termed a ‘catalyst’). We have referenced such material in order to increase the interpretability of earlier literature, while also keeping, as much as possible, within the functional-typological vocabulary most familiar to readers (thus, ‘ergative’ case rather than ‘operative’, for example, and ‘head/dependent marking’ rather than ‘affix-transferring’; see e.g. Nichols 1986).

As is expected from a volume of this scope, different chapters take different approaches to covering their topics. Many authors present a typological overview from a sample. The size of the sample varies, depending mostly on the availability of material or the sampling approach. For example, both Kim and Spronck (Chapters 26 and 30, respectively) have samples of around 40 languages; Kim’s is due to problems with material, while Spronck’s relates to sampling from major lineages across the continent. Other samples are samples of convenience, used to illustrate particular points but the whole continent was not systematically reviewed. Finally, some other chapters focus on particular, well-studied languages (Tabain’s EMA work in Chapter 9; Bradley and Gaby on gender in Chapter 53) to illustrate the phenomenon under discussion.

1.2.4 Supplementary materials and data archiving Many of the chapters have electronic supplementary materials. These supplements contain the data or source materials that were used in the compilation of the typological surveys. These items can be found at https:// zenodo.org/communities/ogal-supplement/. Although all data on which this volume is based came originally from Aboriginal and Torres Strait Islander people,

1.3 Conventions and context In the course of approximately 200 years of academic writing about Australian languages, conventions have arisen which may be unfamiliar to people working in other areas of the world. This section explains some of those conventions and notes the approach of the volume.

1.3.1 Terminological conventions

1.3.2 Referring to Australian languages There are many spelling conventions for referring to Australian languages. In some cases, there are different standards in the anthropological and linguistic literature (Dieri vs. Diyari, for example). In other cases, there have been several spellings according to author (cf. Anindilyakwa

3

claire bowern vs. Enindhilyakwa vs. Anindhiljaugwa). In further instances, the conventions represent minor differences in orthography, such as whether voiced or voiceless letters are used to represent the stop series (e.g. Bindubi vs. Pintupi, Nhanda vs. Nhanta) or how lamino-palatals are transcribed (e.g. Walmajarri vs. Walmatjari). Some examples are described in Bowern (Chapter 7, this volume). Over the course of descriptive and analytical work on Australian languages, a number of other language naming patterns have developed.2 Some names are traditional; others are the result of research practices by non-Indigenous linguists. For example, it has become customary to refer to subgroups of Pama-Nyungan after the word for ‘man’ or ‘person’ (viz. Karnic, Wati, Thura-Yura, Yolŋu, Yuin-Kuri, etc.). In some cases, this has led to problematic conventions. For example, ‘Wati’ is often used for the subgroup of Pama-Nyungan languages of the Western Desert and adjacent regions of Western Australia, South Australia, and the Northern Territory (see Babinski et al., Chapter 75, this volume). However, wati is not a neutral term meaning ‘man’ or ‘person’—in at least some of these varieties, it refers particularly or exclusively to an initiated man. Personal communication with women who speak these languages reveals that they are uncomfortable with the term, which both excludes them as speakers and implies that the language is solely the domain of initiated men. This naming convention apparently arose because of an assumption that many Australian languages have a homonymy between ‘man’ and ‘person’— which is not borne out. That is to say, while researchers have sought to use a consistent strategy for language naming that includes material from Indigenous languages, those conventions can end up as problematic. Another potentially problematic naming convention is to name a subgroup or family after a particular language (e.g. Gunwinyguan, Nyulnyulan, Warluwarric, Worrorran, and the like). These names sometimes appear to language users to prioritize one variety over another. An alternative convention is to use topographic features (cf. the ‘Flinders Island language’). Dixon’s (2002) classification makes extensive use of this practice, replacing names that were previously in common use with names based on such features (e.g. ‘South Kimberley Subgroup’ for ‘Bunuban’). While this avoids the problems of names based on single languages or words for ‘man’, they tie Aboriginal languages to Englishbased naming systems. In order to make it easier to refer to languages throughout the text, we have somewhat standardized the spelling of language names. We have tried to use the form that is, to our knowledge, the accepted form within the language community, or failing that, in the linguistics literature (e.g. 2 See also Rosenberg and Bowern (Chapter 49, this volume) on language names.

4

in the Austlang database, found at austlang.aiatsis.gov.au). Notes on variants in spellings are given in the index. This is the most seamless way to be consistent across the volume while acknowledging that those using the volume may try to find information about particular languages in many different ways. We have retained the familiar subgroup names and families while recognizing that this is problematic.

1.4 Overview of the languages of Australia 1.4.1 Language and descent groups Some important details of language tenure and land tenure may be unfamiliar to some readers of this book, and so are summarized briefly here. Naturally, details vary across the country, but the following schema is broadly applicable. Language is said to have been placed on the land by ancestral culture heroes, during the Dreamtime (a translation of a term known by various names throughout the country, including jukurrpa, bukarri(karra), altyerre, and milon, among many others). This ‘beforetime’ is the foundation of law and lore, morality, and many other aspects of social organization. However, it is not localizable in a particular time; as Stanner (1979: 24) wrote, it is ‘everywhen’, and an ‘unchallengeable, sacred authority’. Perkins’s (2016) frontispiece provides a very effective explanation of Dreamtime through language and images (Muecke and Roe 2021 is also evocative). Country is often said to be linguistically and culturally aware—aware of languages spoken on it, responsive to knowledge and proper address, a stance shown in Bawaka Country et al. (2019: 683), where Bawaka Country itself is given authorship on the article: This paper is authored by Bawaka Country. For Yolŋu people, Country means homeland. It means home and land, but it means more than that too. It means the seas, and the waters, the rocks and the soils, the animals and winds and all the beings, including people that come into existence there. It means the connections between these things, and their dreams, their emotions, their languages and their Rom (Law). It means the ways we emerge together have always emerged together and will always emerge together. This co-becoming manifests through songspirals, known more commonly as songlines or dreamings. Songspirals are rich and multi-layered articulations, passed down through the generations and sung by Aboriginal peoples in Australia to make and remake the lifegiving connections between people and place. (Bawaka Country et al., 2019)

introduction As Hobbles Danaiyarri3 succinctly explained: ‘Everything come up out of ground ~ language, people, emu, kangaroo, grass. That’s Law’. M. K. Turner (2010: 194) puts it this way: The Land needs words … It’s not only words that’s sacred but also it comes from our own Land, and comes from our Ancestors. It’s a gift from that Land for the people who join into that Land – fathers, and brothers, and sisters, and brothersin-law, and also our children. We come from the Land, and the language comes from the Land. And everything that grows from the Land, it really relates to our language as well … Akarre is a sacred tongue because it comes from the Land and it’s part of us, and because we use it to do things, to say things. Ane akaltye anthurre angkentye ikwerenheke. Ane angkentye itethe atnerte mpwepe-arenye apeke re. And that person knows his language, and he knows that his language is born out of the living flesh of that Land.

A person’s primary linguistic affiliations come from country (Sutton 1978; Rumsey 1993). That is, because language comes from country, one’s language ownership (or custodianship) comes along with the custodianship of country. Deborah Bird Rose (1992: 118) summarizes a discussion with Yarralin people, ‘Language, too, is a way of defining country, and thus of defining people who belong to that country. Speaking the language of one’s identity is not an issue here. People belong to the area defined by the Karangpurru language, for example, because one or both of their parents were Karangpurru, and this is so regardless of what languages they speak’.

1.4.2 Language and European settlement Works about Australian languages are accustomed to use the date of European settlement (or invasion) of Australia (1788) as a reference point, and they are written as though we are simply trying to recover as much as possible about the life of Indigenous Australians from several hundred years ago. In this volume, however, I have tried to strike a balance between discussing Australian languages across their histories (reconstruction and change over the course of the last 5,000 years or more) and the very rapidly changing linguistic landscape of the last 50–230 years. Linguists try to use descriptions of languages from points in time at which those languages were still in daily use. However, it is simply a reality that many languages of the Australian continent were described by linguists and the last speakers and signers of those languages, at points when the languages had already fallen out of use as everyday means of communication. In some cases, this was because communities shifted to other 3 A Mudburra man of Yarralin (Northern Territory), quoted to Rose (1996: 9); see also Rose (1992).

languages. More often, though, the shift was brought about because communities were dispersed or ‘disrupted’ (to use a common euphemism), children were forcibly removed from their families, and individuals were punished for speaking their ancestral languages. This has consequences, because how last speakers use their languages may be quite different from speakers at the time of the invasion; therefore, care must be taken in comparing earlier and later records, even if they appear to refer to the same speech varieties. It is traditional to talk about ‘pre-contact’ Australia, referring to the contact between Europeans and Aboriginal and Torres Strait Islanders. But of course, as Vaughan notes (Chapter 54, this volume), linguistic and cultural contact predates European arrival, and even after 1788 is not limited to contact with speakers of English. Other preEuropean encounters include (but are not limited to) contact with Macassans who fished along coastal Arnhem Land in northern Australia for trepang (sea cucumber) from the late seventeenth century, as well as contact with Malay, Japanese, and Chinese pearlers in the early twentieth century in the coastal Kimberley region of western Australia (see further Urry and Walsh 1981; Walker and Zorc, 1981; Evans 1992a). Wood (2018) has made arguments for contacts between speakers of Austronesian languages and Paman languages on eastern Cape York Peninsula.

1.4.3 Language classification In this volume, authors give language family classification information for non-Pama-Nyungan languages; for members of the Pama-Nyungan family, authors give the subgroup they belong to. A few chapters also make use of the primary divisions in Pama-Nyungan discussed by Bowern and Atkinson (2012). This book does not have a detailed discussion of the history of classification of Australian languages, though aspects are covered in brief in Stockigt (Chapter 2, this volume) and in detail for the 19th century in Stockigt (Chapter 6, this volume). Twentieth Century classifications have been covered in detail by Koch (2014a), as well as in the introduction to Bowern and Koch (2004). The front matter and Chapter 7 of the present volume gives some discussion of methods.

1.4.4 Contemporary language use These days, Indigenous languages in Australia are used in many different ways. Some languages are the primary languages of their communities; others are still present (but used less commonly), or used by a small number of people. Others are known, and still important, but not used as

5

claire bowern a primary or daily language of communication. Others are currently being awakened. Many handbooks of languages treat language as disembodied. This handbook attempts to show language as both elegant structures and as important parts of the lives of people. Along with reading this volume for linguistic information and understanding, I hope that readers of this volume will also mourn, witness, remember, respect, celebrate: mourn what has been lost, and the circumstances under which that happened; witness the consequences and remember what has happened; respect individuals and communities for the choices they make about their lives, land, and languages; and celebrate the work currently being done so these languages continue across the land and are handed on.

1.5 Conclusion The editor and authors of individual chapters wish to acknowledge that the contents of this volume have been compiled based on the knowledge of expert Indigenous speakers. Given the number of languages considered throughout, these experts cannot all be cited individually, and we had to list the compilers of knowledge as authors in the table of contents, following academic practice. It must nonetheless be recognized that the linguistic knowledge compiled here comes exclusively from Indigenous speakers and signers. The relative absence of Indigenous contributors as authors of these chapters makes it very clear that we have a long way to go to make linguistics more accessible

6

and desirable as a path of study for Australian Indigenous people (see also Gaby and Woods 2021 for discussion of linguistics in an Australian social justice context).

Acknowledgements Many people have made this book possible. Individual chapter authors have compiled their own acknowledgements, but as editor, I would particularly like to acknowledge the contributors themselves for their forbearance over the many years that this project has been in progress. Not least, the Covid-19 pandemic did not make the final stages of editing proceed smoothly or speedily. For editorial assistance, particular thanks go to Coralie Cram, Juhyae Kim, Jisu Sheen, Irene Yi and Parker Brody. Barry Alpher, Jane Simpson, and David Nash provided feedback on many chapters, which was much appreciated. Editorial and research assistance was funded in part by National Science Foundation grant BCS-1423711 ‘Pama-Nyungan and Australian Prehistory’ and by Yale University. Angela Terrill compiled the index and Brenda Thornley drew the maps for the front matter. I would also like to particularly acknowledge Aboriginal people from Bardi, Yan-nhaŋu, Yolŋu, Punthamara, Kullilli, Wangkatja, Tjupan, and Ngalia communities who have worked with me (in some cases over many years) and shared their languages and so much else. Furthermore, although Luise Hercus passed away in 2018, her influence is on many pages of this work.

PART I

Background

chapter 2

A history of the early description of Australian languages Clara Stockigt

2.1 Introduction The development of understanding about the diversity, internal relationships, and structure of the estimated 400 languages spoken in Australia at the time of British colonization in 1788 (Bowern, Chapter 7, this volume) has paralleled the exploration and encroachment of the coastal and vast interior regions of Terra Australis Incognita. Colonization was slow, and initial struggles were sometimes abandoned. That the mere thirteen Australian Aboriginal languages which continued to be acquired by children in 2014 (Marmion et al. 2014: xii; Bowern, Chapter 7, this volume) belong to areas of the country remote from large population centres and early European settlement, and typically with unproductive soils and inhospitable climates, tells of the detrimental impact that European occupation has had on Australia’s original inhabitants and on their languages. Aboriginal people who have survived the colonial onslaught long enough to have shared their languages with linguists in recent decades tend to originate from parts of Australia, or ‘countries’, which are remote, sparsely inhabited, and infrequently visited. Languages which have been lost before any record of them had been made, or for which only a fragmentary record was taken, tend to be those belonging to country close to European centres of population. While there are historically interesting exceptions to these tendencies—for example, Arrernte is heard today on the streets of Alice Springs, and Unggumi is no longer spoken despite belonging to country far from centres of European population—the pattern is nevertheless evident. The earliest written records of Australian languages predate colonization. The first was taken by William Dampier (1998 [1697]), who in 1688 landed close to Cygnet Bay on the northwest of the continent where he recorded the single Bardi word ‘gurry’—probably ngaarri ‘devil’ (Metcalfe 1979: 197) without translation, but which he described (Dampier 1998: 222) as being cried by people running ‘away as fast as they could’. The window which this single entry provides into Aboriginal experience of early European contact

is apparent in other early linguistic documents. In 1840, for example, Teichelmann and Schu¨rmann gave an example sentence in Kaurna, spoken close to Adelaide, which they translated as: ‘Don’t hang the black man, that the European be not charmed [ensorcelled]’. The next and more detailed records of an Australian language were collected in 1770 by members on board the Royal Navy research vessel the Endeavour commanded by Lieutenant James Cook (Banks 1770; Cook 1893 [1770: 322–3]), who recorded ‘words’ in Guugu-Yimidhirr, or a closely related variety, spoken on the northeast coast. Banks’s wordlist contains what is probably the best-documented (Roth 1901: 6; Breen 1970; Haviland 1974) pathway by which an Australian lexical item entered the English language. Members of the crew of the Endeavour were taught the Guugu-Yimidhirr word ‘kanguroo’ (gangurru) referring to the large black or grey marsupial, and the term subsequently entered the English language and came to be used across Australia where related species had their own Indigenous names. Following the colonization of New South Wales, Australia’s earliest established colony, records were made of the Yuin-Kuri languages spoken in coastal regions north and south of Port Jackson (Sydney). William Dawes’s documentation of the ‘Sydney language’ (see Wilkins and Nash 2008) is remarkable in having been written so soon after the colonization of New South Wales yet decades before any subsequent grammatical description. In 1834 Congregationalist Missionary Lancelot Threlkeld published a grammar of a language spoken near Lake Macquarie, north of Sydney. Threlkeld’s work may be considered the earliest extant Australian grammar, since Dawes did not describe case morphology on nouns or on pronouns, although he did conjugate verbs. Following Threlkeld’s grammar came the MS grammars of Wiradjuri, written soon after the ‘opening-up’ of agricultural lands west of the initially impenetrable Great Dividing Range. Wiradjuri materials written by the Basel Mission Institute-trained missionary James W. Gu¨nther (1838; 1840) were based on now lost

Clara Stockigt, A history of the early description of Australian languages. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Clara Stockigt (2023). DOI: 10.1093/oso/9780198824978.003.0002

9

clara stockigt analyses by the Basel-trained J. C. S. Handt and the Church Missionary Society missionary William Watson (Bridges 1978: 414–15).1 Threlkeld’s and Gu¨nther’s grammatical analyses were later published in an important edited volume of grammars produced in Australia by John Fraser (1892a). Fraser’s collation rescued much early grammatical material from oblivion, and printed other material for the first time, including Hugh Livingstone’s grammar of Minjangbal (1892). After the closure of Wellington Valley mission in 1842, the grammatical structure of languages spoken in New South Wales received little further attention until the Presbyterian Rev. William Ridley began describing Gamilaraay, spoken north of Wiradjuri, in 1852 (1855; 1856b; 1866; 1875). Note here that the terminology employed today by nonAboriginal and Aboriginal people to refer to Australian languages—for instance ‘Awabakal’, ‘Kaurna’, ‘Ngarrindjeri’, and ‘Arrernte’—is the product of post-colonial linguistic and anthropological investigation. These terms were not recorded in the early sources. The language Threlkeld described was named ‘Awabakal’ in 1892 by Fraser (1892a), the term being derived from the name of Lake Macquarie Awaba marked with the associative suffix –kal. The language is nowadays also referred to as ‘Hunter River and Lake Macquarie language’ (Lissarrague 2006; Wafer and Carey 2011). The mechanisms by which many language names, and all their variant spellings, have come to be the accepted descriptors of languages, and of the people speaking them, remain generally not well understood (see Sutton 1979: 89), and can be difficult to retrieve from the historical record (Stockigt, Chapter 6, this volume; Rosenberg and Bowern, Chapter 49, this volume). McGregor’s (2006b: 2–20) ‘history of the histories of Australian languages’ finds most Australian linguistic researchers agreeing that the 1930s and 1960s were watershed decades in the history of the description of Australian languages, which McGregor usefully considers as spanning three periods: a first pre-dating 1930; a second between 1930 and 1960; and a third post-1960. Each is defined by the types of linguistic materials collected, the context in which they were sourced, and the types of technical linguistic training brought to the analysis.

2.1.1 The first descriptive era: pre-1930 The record of Australian languages collected in the pre-academic descriptive era, prior to 1930, was recorded predominantly by missionaries (Stockigt 2015: 336–8). Records taken by explorers, pastoralists, men appointed 1 Rev. W. Ridley held a copy of Watson’s Wiradjuri grammar and vocabulary, which he describes as being in two volumes, and which he returned to the Colonial Secretary of New South Wales sometime before 1873 (Ridley 1873: 275–6).

10

to the office of Protector of Aborigines, and interested individuals are—in comparison with the missionaries’ contribution—likely to be shorter, and to contain no grammatical analysis. Notable exceptions include George Fletcher Moore’s detailed Descriptive Vocabulary (1842) of Noongar spoken around the Swan River Colony (Perth), the numerous grammars by R. H. Mathews (Koch 2008: 211–16) of languages spoken predominantly in Victoria and in New South Wales, and Daisy Bates’s record of many languages from Western Australia. Non-missionary scholars of Australian Aboriginal people and languages were often natural scientists, for example, Robert Brough Smyth, Walter Roth, and Charles Chewings, who were trained to observe and record empirical data, and whose work provided necessary contact with Aboriginal people. But the efforts of early missionaries who were motivated to acquire and describe Australian languages are atypical of the broader picture. Fewer than five percent of Australian languages were grammatically described before 1930. Most missionaries in Australia paid no attention to Aboriginal languages (Harris 1994: 805). Their disinterest might be seen to sit within what the missionary Threlkeld (1850: 10) described: [T]he almost sovereign contempt with which the Aboriginal language of New South Wales has been treated in this Colony, and the indifference shown toward the attempt to gain information on the subject, are not highly indicative of the love of science in this part of the globe, … for which it is difficult to account.

A lack of linguistic data emanating from Australia in the nineteenth century was noted in Germany (H. C. von der Gabelentz 1861: 489; F. Mu¨ller 1867: 241; 1882: 2; G. von der Gabelentz 1891: 403). As late as 1946 the Austrian linguist and ethnologist Wilhelm Schmidt (1946: 941) described ‘a desolate lack of scientifically recorded materials for most Australian languages’. More scathingly, in a review of Schmidt’s 1919 classification of Australian languages, the American anthropologist A. L. Kroeber observed (1921: 226) that the international contribution to Australian linguistics stood in contrast to the apathy of linguistic researchers in Australia: ‘Perhaps the realization that the first scholarly attempt to deal seriously with these tongues was made in German by an Austrian priest [Wilhelm Schmidt] will stir Australians into effort’. Missionaries who were linguistically active held individual views about the complementarity of linguistic study, anthropological study, and proselytization, and their views were sometimes at odds with views held by Church officials under whose authority they worked. The missionary Siebert, at the Bethesda Lutheran mission in South Australia, for instance, was forced to defend himself to the South Australian mission board against charges of having spent

a history of the early description of australian languages too much time pursuing academic collaborative research with the Australian anthropologist Alfred Howitt at the expense of Christian evangelization with the Diyari (Nobbs 2005). The role of linguistic study within philosophical approaches to mission practice are not easily categorized by particular Church denominations. In 1907, for instance, the French Trappist Catholic missionary at Beagle Bay in Western Australia, Joseph Bischofs, took part in a local ceremony ‘blackened from head to foot and adorned with ochre and feathers’ (Mary Durack, quoted in Akerman 2015: vii), while at the nearby Lombadina mission on the Dampier Peninsula German Pallottine Catholics translated scriptures but forbade all ceremonial activity.2 Some recent histories (Graetz 1988: 9; Kneebone 2005: 362; Kenny 2013: 87) have emphasized that the Lutherans’ missionary activity occurred within a philosophical tradition hailing from J. C. Herder, in which understanding a people’s language provided a window into their Volksgeist that was necessary for successful conversion. Such accounts have, however, obscured the fact that similar convictions were held within other evangelical denominations, as well as by Catholic missionaries, who dominated the description of the linguistically diverse non-Pama-Nyungan languages from the north of Australia. Missionaries of different denominations in Australia shared the post-Reformation conviction that Christian conversion occurs sola scriptura (through scripture alone) and were expected to learn Aboriginal languages in order to carry out two interrelated evangelical tasks: the translation of religious texts, and the preparation of materials for use in vernacular literacy programmes. The missionary Threlkeld’s linguistic achievements were made as part of an established tradition of London Mission Society Bible translation (Roberts 2008: 107). The missionary Watson, of the Church Mission Society, who established the Wellington Valley Mission in 1832, was instructed to ‘learn the language and reduce it to writing’ (Bridges 1978: 297). Congregationalist missionary George Taplin similarly wasted no time in advancing the grammatical descriptions of Ngarrindjeri, and Moravian missionaries in Australia were trained and also expected to learn the local language (Edwards 2007: 319). In 1836 Threlkeld produced An Australian spelling book in the language as spoken by the Aborigines…; in 1856 Rev. Ridley produced a Gamilaraay primer (1856a); and in 1885 Jesuit missionaries at Rapid Creek, close to Darwin in the Northern Territory, printed primers in Larrakia (Harris 1994: 461). 2 In 1934, in order that Phyllis Kaberry conduct Australian National Research Council funded fieldwork among Aboriginal women in the Kimberley, Adolphus P. Elkin, Professor of anthropology at the University of Sydney, had to negotiate with Western Australian authorities due to a perception that anthropological investigation led to a revival of ceremonies that thwarted the missionaries’ efforts (Gray 2002: 33–4).

2.1.1.1 Description of non-Pama-Nyungan languages Tasmanian languages and non-Pama-Nyungan Australian languages are defined by their negative membership within the Pama-Nyungan (henceforth PN) family and are spoken over much smaller areas of north-western mainland Australia. Non-Pama-Nyungan languages were scarcely described before the 1930s. In 1898 Sidney Ray attempted to address this ‘scanty knowledge of the tribes and languages of the north-west’ by tabulating (pp. 347–8) extant materials, and presenting vocabularies that had been sent by a ‘Mr E. B. Rigby’ to Ray’s collaborator, the Cambridge anthropologist Alfred C. Haddon. As was the case across the vast majority of the continent where PN languages were spoken, early grammatical analyses of non-PN languages are outnumbered by a plethora of wordlists of variable quality. They include an 1846 vocabulary of Iwaidja by Fr Angelo Confalonieri, and the lexical records of non-Pama-Nyungan languages by Daisy Bates, whose almost insurmountably chaotic but extremely valuable lexical and ethnographic legacy was recorded in the early 1900s (see Thieberger 2016a). J. W. O. Bennett’s vocabulary of Larrakia, initially published as a pamphlet (1869), was the only non-PN wordlist appearing in Woods’s 1879 (Part VII) compilation of material from South Australia, which until 1911 incorporated the Northern Territory. Of the 239 wordlists appearing in Curr’s three volumes of compiled lexical material (1886–1887), only six are of non-PN languages from the north of the country,3 with vocabularies of Tasmanian languages appearing in an appendix (appendix A, vol. 3). Although the French Trappist missionary Fr Alphonse Tachon had written a grammar of Nyulnyul by 1895, and the Jesuit missionary Adolf Kristen had written a grammar of Malak Malak by 1899, these works remained unpublished manuscripts, and the material was not assimilated into Australian linguistic thought. Ray (1897), for instance, was not aware of Tachon’s 1895 Nyulnyul grammar.

2.1.1.2 Linguistics and ethnology A strong tradition of integrated study of linguistics and ethnology existed in Australian before 1930. Many firstera grammarians made both linguistic and anthropological descriptions of the people they worked with. In the 1840s, Lutheran missionary-grammarians who published inaugural grammars of languages spoken in South Australia (Teichelmann and Schu¨rmann 1840; Meyer 1843; Schu¨rmann 3 The non-Pama-Nyungan languages included by Curr are Larrakia, Wulna, Iwaidja, Unalla, Yawuru, and a variety from Roper River (Curr 1886–1887, vol. I: 250–77).

11

clara stockigt 1844) each produced ethnographic publications describing the ‘manners and customs’ of the people speaking languages about which they had earlier published an account (Teichelmann 1841; Schu¨rmann 1846; Meyer 1846). Congregationalist missionary George Taplin and Rev. W. Ridley published linguistic and anthropological material about the Ngarrindjeri in South Australia and the Gamilaraay in New South Wales respectively. The detailed grammar of PittaPitta, spoken in south-west Queensland by Walter Roth (1897), who was practising medicine in the area (Stockigt 2020), appeared as the first chapter of a much larger and influential ethnographic work. Roth later pursued linguistic and ethnological studies in his role as Protector of Aborigines (Breen 2008). But the integration of the two disciplines in this era is most famously represented in writings of the Neuendettelsau-trained Lutheran missionaries Carl Strehlow, father of T. G. H. Strehlow, and to a lesser extent, Johann G. Reuther. Both produced multi-volumed opuses (Strehlow 1907–1920; Reuther 1903–1908), which detailed the language, mythology, and religious beliefs of the Arrernte and the Diyari respectively. Kenny (2013) argues that C. Strehlow’s study of Volk and Sprache (1907–1920) was born out of a particularly German philosophical tradition. But causal links between beliefs of a particularly German nature and descriptive practices in Australia remain tenuous and elusive. Strehlow’s relationship with German Humanist philological traditions is better described as ‘secondary and indirect’ (Gibson 2016: 636) or ‘osmotic’ (Sutton 2015: 257). The scholarly investigation of language made by these German missionaries may have sat within the German study of Ethnologie, with a focus on ‘the relationship between cultural groups in historical and geographical perspective … [and on the study of ] material culture, as well as mythology, ritual and language’ (Peterson and Kenny 2017: 4), but a dispassionate comparison of the ethnological practices of German Lutheran missionaries with other groups in Australia blurs a clear-cut national or religious philosophical distinction.

2.1.2 The second descriptive era: 1930–1960 By the 1930s, Aboriginal populations had been reduced by almost ninety percent, with many surviving communities residing at government depots and missions where they experienced limited autonomy. The second descriptive era of linguistic research in Australia (c.1930–1960) is described by Stephen Wurm (1972: 17) as having focussed ‘strong attention on structural and typological features’. Its onset is demarcated by three typological studies: the study of Australian languages made by the British linguist Sidney Ray

12

(1925), who had previously written grammars of Paman languages (1893; 1907), and the studies produced in 1937 by the Australian anthropological and linguistic researchers Arthur Capell and A. P. Elkin, both of whom steered Australian linguistics towards the modern era. Ranging freely over descriptions of Australian languages written before 1930, Ray, Elkin, and Capell collated and synthesized data recorded in the earlier, pre-academic descriptive era. Their works provide valuable insight into the understanding of Australian linguistic structures that was discoverable at the time. Improved understanding of non-PN languages was advanced in this middle descriptive era largely through the writings of Capell (1938; 1940; 1942), and missionarygrammarians in Western Australia including Presbyterian minister J. R. B. Love (1938 [1933]) and the Pallottine missionaries Hermann Nekes and Ernest Worms’s (2006 [1953]) investigations of Bardi and Nyulnyul collated on the Dampier Peninsula in the 1930s (see McGregor 2000a; 2006b). Capell (1937) advanced the typological distinction between PN and non-PN languages, terming the latter either ‘incorporating’, due to the fact that verbs tend to mark agreement for the number and case of the subject/agent and the object, or ‘prefixing’ (p. 43), because morphemes can attach as prefixes to the stem, which does not generally occur in the ‘suffixing’ PN languages. The beginnings of the 1930s saw the earliest institutionalized academic investigation into Australian languages. Capell conducted his research within the department of anthropology at The University of Sydney, where Elkin, the department’s third professor (1933–1956), had arranged for him to lecture in linguistics. In 1926 A.R. Radcliffe-Brown had been appointed as Australia’s first chair of anthropology at the University of Sydney, the position having been funded by the American Rockefeller Foundation. The establishment of a Rockefeller chair of anthropology in Sydney had been competed for by The University of Adelaide, where anthropological investigation was instead pursued by the Board for Anthropological Research (henceforth BAR) and funded by the Australian National Research Council. The BAR mounted multi-disciplinary expeditions into Central Australia between 1927 and the 1970s, when it was disbanded. In 1930–1931 three members of the BAR formed ‘a small language committee’ (Tindale 1935: 261). They were Norman Tindale—the entomologist and later curator of anthropology at the South Australian Museum—the geographer Charles Chewings, and John A. FitzHerbert, Hughes Professor of Classics at The University of Adelaide between 1928 and 1957. Around this time FitzHerbert was supervising the earliest grammars of Australian languages written for academic degrees in Australia. These were T. G. H. Strehlow’s MA thesis ‘An Aranda Grammar’ (1938), published as part of

a history of the early description of australian languages ‘Aranda Phonetics and Grammar’ (1944), which Elkin (1944: 1) introduced as ‘the first complete phonetic and grammatical study of an Australian language’, and J. R. B. Love’s (1938 [1933]) grammar of the non-PN language Worrorra. Both were published in Oceania, as was Presbyterian missionary R. M. Trudinger’s grammar of Pitjantjatjara (1943), spoken in South Australia. As noted by McGregor (2006b: 5) the founding of Oceania in 1930 provided ‘one of the very few outlets for articles on Australian languages’. By contrast, nineteenth and early twentieth-century linguistic works tended to be published by the Government Printers operating in capital cities. Grammatical material known to have been produced outside South Australia in the 1930s occurred under the direction of Radcliffe-Brown. In 1926, Radcliffe-Brown contacted Edward Sapir at The University of Chicago, whom he would have known had studied unwritten American languages, and suggested that Sapir come to Australia to study Australian Aboriginal languages. Sapir instead sent his student Gerhardt Laves, who subsequently made a survey of numerous Australian languages, and in-depth studies of Gumbaynggir, Karajarri, Bardi, Goreng, Matngele, and Ngan’gimerri, while travelling around Australia between 1929 and 1931 (Nash 1993; 2006). The earliest analyses of Australian languages informed by modern linguistic method were Laves’s manuscript Australian materials, and the manuscript grammar of WikMungkan, spoken in Queensland, made by Ursula McConnel. McConnel, whose anthropological studies were supervised by Radcliffe-Brown, had studied linguistics under Sapir and others at Yale. Thus, Radcliffe-Brown can be credited for attempting to address the dearth of erudite early description of Australian languages, which is likely to have initially motivated his 1926 contact with Sapir. But Laves’s manuscript notes made between 1929 and 1931 travelled with Laves back to America, where they remained largely unpublished, until being returned to Australia and deposited in the AIATSIS library in the 1980s (Nash 1993; 2006). While McConnel published ‘Wikmungkan phonetics’ in 1945, her manuscript grammar of Wik-Mungkan, written close to 1940 based on anthropological fieldwork conducted between 1927 and 1934 (Peter Sutton, pers. comm.) similarly had no immediate impact on the development of linguistic thought in Australia. This document was rescued from imminent destruction in 2006 (Sutton 2010b). The integration of linguistic and anthropological study practised by McConnel (1940; 1945; 1959) in this period is evident in the polymath abilities of Donald Thomson, who similarly worked on Wik-Mungkan (1935; 1936; 1946), and in Ernest Worms’s contribution to Nekes and Worms (2006 [1953]), described by McGregor (2017: 341–2)

as philosophically aligned with the Whorfian tradition. Subsequent to these studies, the integrated investigation of language and culture waned within Australian academic institutions, to the extent that the re-introduction of linguistic anthropology into Australia by Bruce Rigsby from America in the 1970s has been perceived as a ‘new’ approach in Australia (Anderson 2016: 320).

2.1.3 The third descriptive era: 1960– Thoroughgoing linguistic investigation commenced in Australia after 1960, when chairs of linguistics were established at Australian universities. With the appointment of Stephen Wurm as Senior Fellow in Linguistics at Sydney University in 1957, and of Go¨ran Hammarstro¨m as professor of linguistics at Monash University in 1965, the discipline was no longer largely the province of linguistically inclined clergy and members of the general public. That said, the technically excellent salvage work of Victorian languages made by Luise Hercus during the 1960s (1969; see Koch, Austin, and Simpson 2016) was initially not institutionally driven, but rather sustained by a perceived sense of urgency. The establishment in 1961 of the Australian Institute of Aboriginal Studies (AIAS), the precursor to the present-day AIATSIS (Australian Institute of Aboriginal and Torres Strait Studies) greatly improved funding and research opportunities for Australian linguistic research. This era of Australian linguistic research benefitted from the arrival of Ken Hale in 1959, which eventually placed Australian linguistic description in the sights of the Massachusetts Institute of Technology. But the era is branded most strongly by the grammars of R. M. W. Dixon (1972; 1977a) and the works of his students at The Australian National University, one of whom (Wilkins 1989: 59) describes the Dixonian lineage as being highly eclectic as far as their reliance upon, and use of, theory is concerned, and they have drawn freely upon a wide range of approaches to linguistic description … [They] have evolved in parallel with the evolution of theory and practice in linguistics over the past twenty years. While the grammars do little to build theories, they do test theories, and they frequently provide critiques on the ability of various theories to handle the particulars of the language being described.

Revealing of the significance of the body of work fostered by Dixon in the last decades of the twentieth century are the facts that the era prior to his involvement in the description of Australian languages is popularly referred to ‘bB’ (before Bob [Dixon]).

13

clara stockigt This third descriptive era saw the first generation of linguists with PhDs from Australian universities working on Australian languages, notably Donald Laycock, Peter Austin, Peter Sutton, and Terry Crowley. Before then, researchers were qualified either in England, for example, A. Capell (University of London, in 1938) or in America, K. Hale (Indiana University, in 1959), and G. O’Grady (Indiana University, in 1963). At the same time, and outside the realm of the Australian National University research, the complementary study of language and anthropology, or anthropological linguistics, re-emerged in Australia, with Bruce Rigsby’s arrival at the University of Queensland in 1972 and appointment in 1975 to the university’s new department of anthropology. Rigsby, who had studied Sahaptin ethnology and language in America, brought ‘a new and unique perspective that mixed an American ecological approach … with a strong foundation in the cultural anthropology of Kroeber … and the critical role of language and world view (Sapir 1921; Whorf 2012), all premised on a strong, field-based ethnographic approach’ (Anderson 2016: 320). Under Rigsby’s tutelage, accounts of the socio-geographic complex in Australian languages were developed by Rigsby’s students (cf. Sutton 1978; Merlan 1981a). Since then, anthropological linguistics has been integrated into the research precincts of Aboriginal Native Title (Sutton 2003). One might now conceive of a fourth era of Australian linguistic studies, commencing towards the end of the twentieth century, focussing on the description of language contact varieties (Meakins 2014a: 365–6; Angelo, Chapter 56 and Chapter 66, this volume), sociolinguistics (e.g. Mansfield, Chapter 51, this volume), language acquisition (Wigglesworth and Disbray, Chapter 62, this volume), historiography (McGregor 2008a; Stockigt 2017), and on the philological investigation of nineteenth-century records within language reclamation and revitalization movements gaining momentum among language owners whose forebears spoke their languages (Hobson et al. 2010; Gale, Chapter 64, this volume). Rather than providing an overview of the analytical and descriptive achievements made within each of these descriptive eras, which would inevitably overlook significant contributions, and would in any case add little to previously arranged accounts of existing materials (Elkin: 1937a; Wurm 1972; McGregor 2008a), the following discussion focusses on the developing understanding of—and ability to describe— Australian linguistic structures in the nineteenth and early twentieth centuries, and the rise of schools of descriptive practice.

14

2.2 Early analyses of Pama-Nyungan languages Figure 2.1 locates the Australian languages that were grammatically described before 1910 and marks them chronologically. The works of R. H. Mathews are not represented. Mathews’s large body of work, previously assessed by Koch (2008) and described there (p. 181) as a ‘bibliographer’s nightmare’, while broad in scope, is shallow in descriptive depth. The clustering of early grammars in some regions and the absence of grammatical records from others is telling of the treatment of Aboriginal inhabitants in each of six pre-federation (1901) Australian colonies. The butchering of Aboriginal people inhabiting what became the penal settlement of Van Diemens Land, Australia’s second oldest colony, resulted in a sparse lexical record and the absence of any grammatical analyses of Tasmanian languages (Capell 1970: 973; Crowley and Dixon 1981: 395; Bowern, Chapter 7, this volume). In contrast to the scant linguistic record of Tasmanian languages is the record taken in the colony of South Australia, which between 1863 and 1911 encompassed the Northern Territory, and where early grammars and vocabularies were initially produced in quick succession. The swiftness with which missions were established in South Australia, after its founding in 1836, is atypical of the wider Australian experience, and benefitted the description of languages in this colony.4 In 1838, the founding chairman of the South Australian Company, George Fife Angas (1789– 1879), assisted the passage to the colony of graduates from the Evangelisch-Lutherischen Missions-Gesellschaft zu Dresden (Evangelical-Lutheran Mission Society of Dresden) to work among the Aboriginal population (Lockwood 2014: 61–5). Lutheran missionaries consequently dominated the early grammatical description of South Australian languages, making comparatively detailed grammatical descriptions of seven languages before the turn of the twentieth century: Kaurna, Ramindjeri, Barngarla, Diyari, Wangkangurru, Yandruwandha, and Arrernte. In 1970 Elkin coined the term ‘BIITL [before it is too late] research’ to stress the necessity of prompt linguistic research in Australia. The sentiment is, however, nearly as 4 The 1834 South Australian colonization act was passed relatively late within British imperial history. Secretary of State to the Colonies, Lord Glenelg, was among a group of humanitarians who insisted in letters issued to the Colonization Commission in 1836 that the welfare and rights to land of Aboriginal people in South Australia be formally protected by the Colonial Office (Reynolds 1987: 94–102). Their dispossession proceeded anyway.

a history of the early description of australian languages 15 20 21

17

22

13

16 18 19 10

9 14 8

4

2 6

3 5 7

1

11 12

Figure 2.1 Map showing locations of languages described in the corpus, and lower-level PN subgroups. Source: after Bowern and Atkinson 2012: 820, used with permission of the authors

Language

Classification*

Spoken in the Colony of:

Early source material

South-eastern: Yuin-kuri

New South Wales

Threlkeld 1834; Hale 1846

2

Awabakal (HRLM language) Wiradjuri

South-eastern: Central NSW

New South Wales

3 4 5

Kaurna Nyungar Ngarrindjeri

Central: Thura-Yura Western: Nyungic South-eastern: Lower Murray

South Australia Western Australia South Australia

6 7 8

Barngarla Ngayawang Gamilaraay

Central: Thura-Yura South-eastern: Lower Murray South-eastern entral NSW

South Australia South Australia New South Wales

9 10

Turrubul Diyari

South-eastern: Durubalic Central: Karnic

Queensland South Australia

11

Wergaya

South-eastern: Kulin

Victoria

Watson no date (lost), Gu¨nther 1838, 1840; Hale 1846; Mathews 1904c Teichelmann and Schu¨rmann 1840 Symmons 1841 Meyer 1843; Taplin 1874, 1872[1870], 1878 Schu¨rmann 1844 Moorhouse 1846 Ridley 1855, 1856a, 1856b, 1866, 1875; Mathews 1903b Ridley 1866 Koch 1868; Schoknecht 1947; Flierl 1880; Reuther 1899; Planert 1908; Gatti 1930 Hagenauer 1878; Mathews 1902 (Djadjala)

1

Continued

15

clara stockigt Continued Language

Classification*

Spoken in the Colony of:

Early source material

12

Ganai

Victoria

Bulmer 1878

13

Western Arrernte

South-eastern: Eastern Victoria Central: Arandic

South Australia (Northern Territory)

14 15 16 17

Minjangbal WTS Pitta Pitta Guugu-Yimidhirr

South-eastern: Bandjalangic Northern: Paman Central: Karnic Northern: Paman

New South Wales Queensland Queensland Queensland

18 19 20 21 22

Wangkangurru Yandrruwandha Nggerrikwidhi Yadhaykenu Luritja

Central: Karnic Central: Karnic Northern: Paman Northern: Paman Western: Wati

South Australia South Australia Queensland Queensland Western Australia

Kempe 1891; C. Strehlow no date c.1907, 1908, 1910; Mathews 1907; Planert 1907; Riedel 1931 [c.1923]; T.G.H. Strehlow 1944[1938] Livingstone 1892 Ray and Haddon 1893; Ray 1907 Roth 1897 Schwarz and Poland 1900; Roth 1901; Ray 1907 Reuther 1901 Reuther 1901 Hey 1903; Ray 1907 Ray 1907 Mathews 1907; C. Strehlow 1910

Nyualnyual Malak Malak

Non-PN Non-PN

Kimberley Daly River region

Tachon 1895 Kristen 1899

1 2

* According to Bowern and Atkinson 2012 old as the study of Australian languages itself. Against a devastatingly high rate of post-colonial linguistic extinction (Bowern, Chapter 7, this volume), the description of Australian Aboriginal languages has, almost since its inception, been imbued with a resolve to record material from successive generations of ‘last fluent speakers’. The two earliest published grammatical accounts of Australian languages (Threlkeld 1834; Teichelmann and Schu¨rmann 1840) were written by missionaries who continued to refine their analyses at a time when they perceived themselves to have outlived the languages’ last fluent speakers (Threlkeld 1850; Teichelmann 1857; 1858a, b). Salvage studies have played a weighty role in Australian linguistic description. Many of the earliest described languages ceased to be spoken before any subsequent substantial record was taken. Consequently, early grammars have recently received considerable attention from within the language reclamation and revitalization movement that has gained momentum among Aboriginal descendants of speakers (see Hobson et al. 2010) (Gale, Chapter 64, this volume; Amery and Gale, Chapter 65, this volume). The nature of linguistic work carried out in Australia during the nineteenth century generally runs counter to the Zeitgeist of genealogical and typological linguistic classification. While similarities between words in Australia and those from around the world were observed

16

by researchers in Australia (Grey 1845; Taplin 1879a; Curr 1886–1887; Fraser 1892b) as part of what Capell (1970: 667) described as an ‘epidemic of origin hunting’, the early Australian grammars written in the country were predominantly synchronic, pedagogical, non-comparative, and nonclassificatory. That said, systematic attempts to collect linguistic materials for comparative purposes commenced in Australia in the 1870s. Within the earliest ‘survey era’ of Australian languages (O’Grady et al. 1966: 5; McGregor 2006b: 3) surveys were distributed among missionaries, police troopers, pastoralists, and others in contact with Aboriginal people resulting in important published collations of linguistic material, including Curr (1886–87), Brough Smyth (1878), Taplin (1879a), and Bates, the results of whose 1904 survey of Western Australian languages were not published until 1985 (Bates 1985). Curr’s four volume The Australian Race cast the broadest net of the four, and it is the most renowned of these survey works. Neither Curr nor Bates specifically elicited grammatical material from their informants, as did both Taplin and Brough Smyth. The linguistic material contained in Volume II of Brough Smyth’s (1878) two-volume work has been assessed by O’Grady, Voegelin, and Voegelin (1966: 5) as ‘quantitively impressive but qualitatively appalling’. Yet the volume contains the most significant record of many Victorian languages. The data Taplin sought in his

a history of the early description of australian languages Labial

Inter-dental

Apico-

Retroflex,

alveolar

Apico-domal

Palatal

Velar

Stop

p

th

t

rt

ty

k

Nasal

m

nh

n

rn

ny

ng

lh

l

rl

ly

r

y

Lateral

rr

Trill Glide

w

(r)

(h)

Figure 2.2 The inventory of consonant phonemes common to many PN languages. Phones separated by dashed lines tend to be orthographically undifferentiated in the early sources.

survey circulated in South Australia in 1874 (Taplin 1879a: 6) were largely anthropological, but the nature of the linguistic questions he posed indicates the developing understanding of Australian grammatical structure. (cf. Stockigt, Chapter 6, this volume).

2.2.1 Phonology and orthography Australian languages have fairly similar phonemic systems by cross-linguistic standards. Systems with three vowel phonemes are common (cf. Round, Chapter 10, this volume). Consonants typically show a limited number of manner contrasts but a more extensive set of place of articulation contrasts. Figure 2.2 shows a maximally contrastive inventory.5 Early orthographic treatments of Australian phonologies tended to give a broad transcription of vowel quality, while phonemic articulation contrasts of consonants tended to be under-represented. Phones separated by dashed lines in Figure 2.2 tended to be orthographically undifferentiated in the early sources. Generally, the filter of the European ear saw the orthographic collapse of coronal consonant phonemes, with the letters ‘t’, ‘n’, ‘l’ used generically to represent stops, nasals, and laterals at all coronal articulatory places (cf. also Thieberger, Chapter 3, this volume; Giacon and Koch, Chapter 8, this volume). All early records of Australian languages were compromised by the inability to aurally distinguish certain types of segments, the absence of phonemic analysis, and the absence of a standardized system of orthographic representation. The first language of the recorder influenced the spelling of Australian languages. The German-speaking missionaries chose ‘j’ rather than the ‘y’ 5 Diversions from these tendencies are too numerous to be dealt with comprehensively here, but of the early described languages it is notable that Arrernte and Diyari show additional distinctions, a series of pre-stopped and rounded consonants in Arrernte (Breen 2001), and a voicing distinction in Diyari restricted to apico-alveolar and retroflex stops in non-word-initial positions (Austin 1981a [2013: 13]). Guugu-Yimidhirr follows the pattern of many languages in the country’s eastern third in having no retroflex series.

to represent a palatal glide, for instance, and the Italianspeaking Benedictine Bishop Rosendo Salvado (1851) represented palatalized nasals and laterals as ‘gn’ and ‘gl’ respectively. The poor correspondence between sounds and letters in the writing system of English greatly impeded the unambiguous representation of Australian phonologies. Consequently, some early recorders (e.g. R. H. Mathews) purported to employ an Italian system of representing vowels, while others attempted to apply conventions established by the Royal Geographic Society (1831; 1885) for writing unwritten languages (see Moore 2013). Threlkeld (1834: vi) adopted the spelling system employed in the description of languages from Polynesia, where he had spent six years at London Mission Society missions prior to coming to Australia, because he sensed propriety in adopting the ‘same character to express the same sounds used in countries which are adjacent’. Although Teichelmann and Schu¨rmann (1840: v), Gu¨nther (1840: 338), and Ridley (1856b: 290) were subsequently able to reassure their readers that they followed Threlkeld’s ‘method of spelling words’, early researchers were aware of, and frustrated by, the inadequacy of writing systems. Palatal stops were more likely than other coronal consonants to be distinguished, and were represented as ‘ty’, ‘tj’, or ‘ch’, followed by palatal nasals and laterals. A notable exception here is the identification of word-final palatalized nasals and laterals by Benedictine missionaries at New Norcia in Western Australia, who having Spanish as their first language were probably predisposed to hear the sounds.6 While the velar nasal was often undifferentiated from other nasals, or not represented word-initially, some very early grammatical sources (Dawes 1790–1791b: 1ff.; 1790–1791a; Hale 1846) represented the phone using engma (ŋ). Rhotic phonemes, of which Australian languages typically have two or more (Round, Chapter 10, this volume), were sometimes, but inconsistently, distinguished. Impressionistically, the phonemes least likely to be distinguished in the corpus grammars appear to have been the retroflex series and 6

Thanks to Peter Sutton for pointing this out.

17

clara stockigt

Figure 2.3 Lepsius’s (1855: 64) two-dimensional representation of Australian phonology based on Teichelmann and Schu¨rmann’s (1840) account of Kaurna.

Figure 2.4 Fraser’s (1892a: 8) grid of Australian consonants.

interdental nasals and laterals. Orthographic treatment of Diyari phonology was gradually improved by successive generations of missionaries at the Bethesda mission between 1866 and 1915. After decades of practice, missionaries introduced the representation of retroflex laterals, at least in some words, as with the ergative interrogative warli, ‘who’ shown as warle (Reuther 1894) as opposed to wale (Koch 1868). Phonological science was slow to enter Australian description. The earliest presentations of the sounds of Australian languages in systematic diagrams—setting out consonant inventories in tables mapping place of articulation against manner of articulation, and vowels in triangular displays mapping height against backness—occur in descriptions of Australian languages published in Europe (Lepsius 1855: 64; 1863: 226; Mu¨ller 1867; 1882; Planert 1907; 1908; Gatti 1930: 1; Sommerfelt 1938: 42, 45) (Figure 2.3). These presentations of both consonants and vowels based on articulatory parameters made by German philologists appear not to have been read by grammarians in Australia, and if they were, not understood or assimilated into Australian practice. Aside from a confused attempt to show consonants in a grid made by Fraser (1892b: 8) (Figure 2.4), the earliest graphic representation of consonants published in Australia was Capell (1956: 8), and the earliest two-dimensional representation of vowel shape given in a grammar produced in

18

Australia appears in T. G. H. Strehlow’s grammar in Western Arrernte (1944: 4 [1938]). The first academic attempts to develop an orthographic system capable of consistently representing the phonological segments of Australian languages occurred on the cusp of the second descriptive era. The ‘language committee’ from The University of Adelaide, was described by Tindale (1935: 261) as having ‘formulated after much consideration, a working list of phonetic symbols applicable to the general study of central Australian languages’. This move was probably instigated by large lexical projects in which members of the BAR and the ‘language committee’ were engaged during the 1930s, which were buoyed by a long tradition of investigating Australian languages from Adelaide. Tindale was compiling a Pitjandjara vocabulary (1937b) which included 2,950 headwords in Western Desert varieties collected during the BAR expeditions (1935), and in 1931 Chewings had embarked on the ambitious undertaking to collate previously recorded Arrernte vocabularies into a single document (Chewings and FitzHerbert 1931–1932). The immediate problem of reconciling the different spellings for the same entry in multiple sources necessitated the development of an adequate writing system. The problematic orthographic treatment of Australian phonologies must have been at the forefront of FitzHerbert and Tindale’s minds during a meeting with Laves, who

a history of the early description of australian languages visited Adelaide in 1931 while conducting linguistic fieldwork during a trip around Australia (Nash 2006). The ‘language committee’ used bold, italic, and plain script to disambiguate phonemes, which they recognized as inconsistently represented in earlier sources. The system clearly differentiated two rhotics, the approximant was placed in italics, and the trill in standard print. The committee also deliberated about using a single symbol to represent voiced and unvoiced stops at identical articulatory places, a practice that the American ethnologist Horatio Hale (1846: 485) (see also Stockigt, Chapter 6, this volume) had considered nearly a century earlier, but had similarly not instigated. While the phonetic system of the Adelaide language committee evinces a peri-phonemic analysis, some phonemes remained orthographically undifferentiated. Palatalized laterals were not recognized although palatalized nasals and stops were differentiated from their alveolar counterparts as digraphs with a final ‘j’, and the system had no systematized method of representing retroflex sounds. Fitzherbert’s student T. G. H. Strehlow did not follow the system developed by his supervisor, but opted instead to create a broad phonetic orthography. Although Strehlow did not make a phonemic analysis of the languages, his diacritically embellished phonetic system distinguished all phones that were phonemically distinctive, and many more. An early attempt to reflect phonemic structure in the writing system of an Australian language is, however, given in an analysis of Arrernte made in the same year that T. G. H. Strehlow submitted his Master’s thesis. Alf Sommerfelt (1938: 42), informed by earlier published accounts of Arrernte, notably Planert (1907), postulated that the language ‘ha[d] at its disposal only three really distinct vowels: a, i, u’,7 and represented voiced and voiceless stops having the same articulatory place with a single symbol. Tracing lineages of understandings of Australian phonemic systems through the careful analysis and comparison of orthographies is a rich field of historical enquiry which remains to be done. John McConnell Black (1917; 1920) recorded South Australian languages using a current phonetic transcription, which represented retroflex and some interdental consonants (Simpson et al. 2008: 93). His influence on the system used by the Adelaide language committee remains undetermined. The little that is known about the exchange of understandings and practices between those studying Australian languages in Adelaide and those studying Australian languages in Sydney suggests that the movement of ideas was limited and unidirectional. 7

‘l’on constate que l’aranta ne dispose que de trois voyelles réellement distinctes: a, i, u’ (Sommerfelt 1938: 42).

U. McConnel’s 1945 publication titled ‘Wikmunkan Phonetics’, written within the realm of anthropology at the University of Sydney, acknowledges (p. 355) the assistance of FitzHerbert, professor of Classics at The University of Adelaide. She also thanks T. G. H. Strehlow. McConnel compares Wik-Munkan and Arrernte phonologies, referring the reader to Strehlow’s analysis throughout and noting where her orthographic choice differed from Strehlow’s.

2.2.2 Descriptive models All early grammarians employed a model of grammatical description that was termed the ‘word and paradigm’ (henceforth WP) model by Hockett in 1954 (Robins 1959). The model Hockett named had developed in Europe to best convey the fusional and synthetic typology of European languages. The model takes the word rather than the morpheme as the minimal unit of analysis.8 The application of the WP descriptive model to Australian structure resulted in an effective representation of an important range of Australian morphosyntactic data. But unlike the synthetic and fusional morphology of standard average European languages, Australian morphology is generally both synthetic and agglutinative (see further Shoulson, Chapter 17, this volume; Alpher, Chapter 40, this volume). Individual grammatical categories carried within the word tend to be inflected discretely. Thus, wholesale application of the WP model resulted in descriptions that were unnecessarily repetitive. T. G. H. Strehlow’s grammar of Arrernte (1944 [1938]) typifies how the indiscriminate application of the WP model might result in uneconomical descriptions. Presenting first full tables of declension for both indefinite nouns and definite nouns—i.e. those that are followed by a third person pronoun—after already having presented the case forms of third person pronouns in full tables of pronominal declension, he went on to provide full declensions for adjectives, which are syntactically undifferentiated from nouns (1944: 78–86 [1938]). But most early PN grammarians, especially those who wrote detailed grammars and had learnt the language relatively well, blended two descriptive models when describing PN languages: the WP model, inherited from the classical description of fusional European languages, and a commonsense word-internal model befitting the agglutinative typology of the languages at hand. Meyer (1843) and Schu¨rmann (1844), for instance, show a diminished reliance on the WP model to convey the function of case forms. They chose rather to list the multiple functions of nominal suffixes. In 8 Contemporary discussion of the word and paradigm model has developed different senses of the idea of the word and has introduced the notion of the lexeme (Koch 1990).

19

clara stockigt the introduction to his grammar of Minyangbal, Livingstone (1892: 3) justified his abandonment of the WP model on the basis of the agglutinative nature of Australian languages: It is well known that the Australian dialects are agglutinative, everything in the nature of inflection being obtained by suffixes. To this, Minyuġ is no exception; so that if I give an account of its suffixes, that is nearly equivalent to giving an exposition of its grammar. It will therefore, be convenient to take, first, such suffixes as are used with the noun and its equivalents, and, afterwards, those that may be regarded as verbal suffixes.

Sub-word units were commonly recognized and represented by the earliest PN grammarians, for example, Gu¨nther 1838: 45; Teichelmann and Schu¨rmann 1840: 5 (Figure 2.5). This occurred prior to August Schleicher’s analysis of wurzeln ‘roots’ and beziehungslauten ‘inflections’—literally ‘relational sounds’—given in an 1859 description of Lithuanian, which Blevins (2013: 383) describes as ‘almost entirely without precedent in the classical tradition’.

2.2.3 Schools of morphological description Nineteenth-century Australian grammarians were aware that the description of Australian languages might be compromised by ‘categorical particularism’: ‘one of the major insights of structuralist linguistics of the 20th century (especially the first half) that languages are best described in their own terms … rather than in terms of a set of preestablished categories that are assumed to be universal’ (Haspelmath 2010: 2). The realization that the study of language should be non-aprioristic is widely associated with Boas (1911: 81) (Haspelmath 2010: 4ff.), but is traceable to the writings of W. Humboldt (1827, quoted in MorpurgoDavies 1975: 105). In 1844, Schu¨rmann perceived a tension between the premises underlying received descriptive linguistic schemata and the new linguistic structures he encountered, and advised (1844: vi) that the description of Australian languages required authors to divest their minds as much as possible of preconceived ideas, particularly of those grammatical forms which they may have acquired by the study of ancient or modern languages.

In the earliest grammar of an Australian language, Threlkeld was similarly aware that linguistic principles deduced from the study of classical languages did not have universal application and that Australian languages could not adequately be described by the existing descriptive framework. He wrote (1834: x): Figure 2.5 Teichelmann and Schu¨rmann’s (1840: 5) presentation of case suffixes (Kaurna).

Some nineteenth-century grammarians in Australia not only recognized and represented sub-word units of meaning, but also innovated pre-theoretical practical descriptive responses that described the relative ordering of wordinternal constituents either in terms of process or arrangement, without of course using these terms. In his description of Pitta-Pitta (1897: 8), W. E. Roth gave predictive syntagmatic statements about word internal constituents, instigating a distinctive method of conveying the relative ordering of inflections for number and case on nominals that was more efficient than the traditional exposition of the forms in lengthy paradigms. The genesis of alternative descriptive models, which did not accord centrality ‘to the word as a fundamental unit in the grammar as a whole and as the basic syntactic unit’ (Robins 1959: 118) is generally associated (ibid., 111–12) with the description of American Indian languages in the early decades of the twentieth century by Boas (1911) and Sapir (1921, esp. Chapter 4). There are nineteenth-century precedents in the description of Australian languages.

20

The arrangement of the grammar now adopted, is formed on the natural principles of the language, and not constrained to accord with any known grammar of the dead or living languages. The peculiarities of its structure being such, as to totally prevent the adaptation of any one as a model.

But even with this awareness, the early grammarians’ descriptions were hampered by the absence of appropriate frameworks and terminology to describe the foreign structures. With reference to Threlkeld’s grammar of Awabakal (1834), H. Hale appreciated the difficulty in framing a ‘mass of information which is entirely new’ (1846: 482) without appropriately developed descriptive tools: It is not surprising that the novelty and strangeness of the principles on which the structure of the language was found to rest, should have rendered a clear arrangement, at first a matter of difficulty; and some degree of obscurity and intricacy in this respect have caused the work to be less appreciated than its merits deserved.

Understanding of the structures the early grammarians were trying to convey involves stripping back the veil of arcane terminology and inappropriate descriptive

a history of the early description of australian languages frameworks. The reimagining of the authors’ logic when describing previously undescribed structures requires a consideration of the looking glass through which the skewing of the description of Australian linguistic structures occurred.9 Elkin (1937a: 9) described the process of retrieving material from the early sources as ‘careful sieving’. The early grammarians’ knowledge of classical and modern European languages led them to anticipate the presence of grammatical features common to many European languages but absent in Australian languages. Hence early grammars include the description of categories inherent to the traditional descriptive framework but usually inapplicable when applied to Australian languages such as vocative case, articles, morphological marking of comparative and superlative degrees of adjectives, passive voice, and grammatical gender.10 On the other hand, early grammarians in Australia encountered an array of ‘foreign’ morphosyntactic features that could not easily be accommodated within the schema and descriptive models developed to describe the structure and typology of classical European languages. The description of these categories required grammarians to either extend themselves beyond what was descriptively familiar or to borrow techniques innovated by previous grammarians. Early grammarians were most likely to look for guidance from their predecessors’ descriptions when describing structures that the traditional grammatical framework was powerless to convey. These areas of the grammar, for which the early missionary-grammarians were theoretically and terminologically ill equipped, are most likely to evince lineages of descriptive practice. Notable instances include the early grammarians’ descriptions of the marking and function of the ergative case (see Stockigt 2015), systems of bound pronouns (Gaby and Shoulson, Chapter 24, this volume), the juxtaposition of constituents in inalienably possessed phrases (Ponsonnet, Chapter 22, this volume), the inclusive and exclusive pronominal distinction (Gaby and Shoulson, Chapter 24, this volume) and the large morphological case systems of PN languages. Comparison of the descriptions of these features reveals three distinct schools of practice that developed in the preacademic era of Australian linguistic description (Stockigt 2017). The first was instigated by Threlkeld (1834) and 9 The metaphor is borrowed from Nowak’s (1993) account of the early description of Inuktitut ergativity. 10 Minjangbal (Livingstone 1892) is the only PN language described in the pre-academic era that has grammatical gender. A relatively large proportion of the small group of about a dozen Pama-Nyungan languages that make a two-way gender distinction in third person pronouns (Dixon 2002: 461) were described in the pre-academic era. Early grammarians of Diyari (Homann 1892), Pitta-Pitta (Roth 1897), Minjangbal (Livingstone 1892), and Kalaw Lagaw Ya (Ray 1907) readily described the distinction (Stockigt 2014).

is largely confined to grammars of languages spoken in New South Wales (Gu¨nther 1838; Ridley 1856b; Livingstone 1892 [1876–1886]). The second was instigated by Lutheran missionaries describing languages spoken near Adelaide in the early 1840s, particularly Teichelmann and Schu¨rmann’s (1840) grammar of Kaurna. Different aspects of Teichelmann and Schu¨rmann’s descriptive template, which had not been inspired by Threlkeld (1834), influenced a large body of South Australian grammatical description. This school is the most strongly attested of the three identified schools. It is defined by a greater number of shared descriptive practices, which are found in a larger body of work, and it endured for a longer period (1840–1938). T. G. H. Strehlow’s grammar of Arrernte (1944 [1938]) is the culmination of this long history of Lutheran grammatical description (Stockigt 2017). The third school of early Australian grammatical description was instigated by W. E. Roth (1897). The features of Roth’s template, which define the Queensland School of description subsequently utilized in later grammars of Guugu-Yimidhirr (Schwarz and Poland 1900; Roth 1901) and Nggerrikwidhi (Hey 1903), relate to the description of nouns and pronouns in peripheral cases. Australian grammarians often worked in intellectual isolation from fellow grammarians posted across far-flung regions of the country. The only early grammar of a Western Australian PN language (Symmons 1841) appears not to have been read by any other early grammarians before republication in Fraser (1892a). Grammars of Diyari written by successive generations of Lutheran missionaries at Bethesda, of which Reuther (1899–1901) is the best known, remained obscure unpublished German MSS well into the twentieth century, and some (Koch 1868; Flierl 1880) remain so.11 Roth’s insightful and well-exemplified grammar of Pitta-Pitta (1897) makes no reference to earlier descriptions of Australian languages, including the Lutheran missionaries’ analyses of Karnic languages related to Pitta-Pitta (e.g. Homann 1892; Reuther 1894; 1899). The style of Roth’s grammatical description is unique, suggesting that like his ethnological investigation (Mulvaney 2008: 114), the grammar was uninfluenced by earlier analyses. That such a descriptively innovative work was written relatively late in the pre-academic era of grammatical description in Australia, with little or no recourse to previous analyses, reveals much about the development of linguistic ideas in the country. Like Roth, T. G. H. Strehlow’s (1944 [1938]) grammar of Arrernte presented morphosyntactic features as if they were 11 The grammars of Diyari by Koch (1868) and Flierl (1880) written in German Kurrentschift have in 2019 been transcribed and translated into English and deposited in AIATSIS as part of the Centre of Excellence in the Dynamics of Language, Language Documentation Grant, ‘Pilot digital archive of historical sources in Australian languages’.

21

clara stockigt ground-breaking discoveries, without referencing either the score of earlier Australian grammars in which the same structures had been adequately described, or the early typological overviews from 1846. Unlike Roth, T. G. H. Strehlow was aware of, and was indeed influenced by, earlier grammatical descriptions (Stockigt 2017). His reasons for choosing not to acknowledge his Lutheran predecessors, including his father, the missionary C. Strehlow, are partially treated in a biography (Hill 2002).

2.3 Concluding comment Perhaps due to the relative youth of the Australian colonies, and to their smaller population, Australia failed to produce nineteenth-century linguistic luminaries like those who worked in America and advanced the study of American languages within European thought: men such as John Pickering and Peter Stephen Du Ponceau, both of whom corresponded with William Humboldt (Nowak 2000), or William D. Whitney, whose presence in America kept nineteenthcentury American Indian linguistic studies ‘up to date with and [having] benefitted directly from contemporary linguistic thinking’ (Campbell 1997: 28). Nor did nineteenthcentury linguistics in Australia benefit from a political leader who promoted the study of ‘native’ languages in Australia in the way that Thomas Jefferson did as president of the American Philological society (1797–1815) in America. The dissemination of understandings across vast distances in Australia was hampered by the absence of coordinated institutionalized effort, and by the fact that missionaries speaking different languages worked in missions of different denominations operating in politically independent colonies. The ad hoc nature of the development of linguistic studies in Australia before the 1930s meant that pioneering descriptive responses to newly encountered structures by missionary-grammarians in Australia were not reliably integrated into a central body of emerging thought.

22

The schools of descriptive practice which are evident are regionally defined within the politically autonomous prefederation Australian colonies. Examination of the early description of Australian languages evinces pre-theoretical and common-sense descriptive practices that are important precursors to rudimentary advances that were cultivated elsewhere in the world but which failed to thrive in Australia. Roth’s descriptive response to the agglutinative structure of Australian languages (1897) occurs before Boas and Sapir were active in American linguistics (Boas 1909; 1911 Sapir 1911a, b; 1921), but his descriptive response to agglutination did not impact on the theoretical development of novel descriptive models appropriate to the description of agglutinative morphology. Innovative descriptive methods and ideas were not reliably integrated into a central body of emerging Australian linguistic thought. Further, the failure of scholars in Australia to assimilate the enduringly instructive two dimensional mapping of the place of articulation against manner of articulation of Australian consonants and vowels, that had been presented in European publications between 1854 and 1938 indicates that channels of communication from Europe to Australia were under-developed, and that the movement of ideas between the continents was largely unidirectional. As the opportunity for linguists to record Aboriginal languages diminishes, the early records of Australian languages are of increasing significance both within the revival context and for the reconstruction of Australia’s pre-contact linguistic landscape. Much work remains to be done, principally a comprehensive comparative study of the early representation of Australian phonology. Such historiographically rich fields of enquiry form the basis of a sound philological method of retrieving important data from these early sources, such as Elkin’s process of ‘careful sieving’. The recent scanner-driven revolution of historical investigation, launching manuscripts from archival oblivion onto the ‘www. platform’, has recently made thorough historical investigation reasonably possible.

chapter 3

Documentation of Australian languages Nicholas Thieberger

3.1 Introduction In this chapter I will focus on records of Australian languages—not the academic work of linguistic analysis (grammars and dictionaries are dealt with in other chapters), but documentation, including texts, transcripts, recordings, and other primary materials in the languages. Bruce Pascoe, discussing the journals of William Thomas (which include a priceless collection of information from the late 1830s), notes that they are one of the most important primary sources in Australian history; but he says: ‘So who published his journals? A university, a government department, his church, a private researcher? No, the Victorian Aboriginal Corporation for Languages published the papers in 2014’. He goes on to note further that Robinson’s diaries, a similarly important set of primary manuscripts, were only published in the 1980s: ‘While settler reminiscences, football club centenaries and books on outback toilets found plenty of researchers and publishers, two of the most important texts on Aboriginal culture waited over two hundred years. Sadly, this is a reflection of Australia’s interest in the Silent War’ (Pascoe 2014b: i–ii). I characterize this as Pascoe’s challenge to linguists, and I take up this challenge by exploring what can be done to increase the accessibilty of primary sources for Australian languages. There are many early records of Australian languages that would benefit from wider distribution, whether by publication as books, or—probably more usefully—as searchable webpages. Early sources, together with field-recordings made by linguists, constitute the main primary documents representing aspects of Australian languages. This chapter will both provide an overview of the primary documentation available for Australian languages and guide the reader in how to find records of these languages. While grammars provide analyses for particular aspects of the language, they have a narrow readership, and, moreover, are accessible in public libraries. It is the primary source material that will be the focus of this chapter and, as will be seen, there are few such records available for most Australian languages.

There are between a few hundred and 1,200 Australian languages1 depending on how they are counted.2 Because of the severity of the invasion that began on the east coast in the late 1700s and the subsequent population decline through disease and intentional murder by the invaders, the number of speakers of Australian languages declined significantly. The overwhelmingly racist nature of the subsequent settler society meant that little, and in some places nothing at all, was recorded of the Indigenous languages. These languages, together with other aspects of Indigenous life, were considered to be of little interest by the invaders (see for example Australian Human Rights Commission 1997).

3.2 Why documentation? In the 1990s, some linguists returned to recognition of documentation; Krauss’s plenary at the Linguistic Society of America conference (published as Krauss 1992) is often cited as a key moment in the move from the recognition of exclusively analytical linguistic work to also including primary records for citation, verification, and for return to the source community as part of the quid pro quo of fieldwork. However, an emphasis on primary records goes back at least to the US anthropologist Franz Boas in the early 1900s, whose trilogy of documents required for language description are the grammar, texts, and dictionary; Boas’s model was later exemplified in Australia by Heath’s Nunggubuyu volumes (Heath 1980d; 1982a; 1984). Similarly, Boas’s student, Edward 1 I will use the term ‘Australian languages’ throughout to include all Australian Indigenous languages. I capitalize ‘Indigenous’ when it refers to the original languages or people of Australia. 2 There are three main authoritative lists of language names and identifiers: the Austlang list at the Australian Institute of Aboriginal and Torres Strait Islander Studies has 1,215 language codes; Glottolog has 343; and Ethnologue and the International Organization for Standardization (ISO-639-3) have 359. The difference in numbers is due to each being constructed for slightly different purposes and with different criteria about what constitutes a language. See Bowern, Chapter 7 (this volume) for more detail.

Nicholas Thieberger, Documentation of Australian languages. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Nicholas Thieberger (2023). DOI: 10.1093/oso/9780198824978.003.0003

23

nicholas thieberger Sapir, discussing the nature of linguistic data created by fieldworkers, said he was not interested in the ‘smoothedover’ versions of native culture, but rather wanted to see the ‘… genuine, difficult, confusing, primary sources. These must be presented, whatever else is done. … There are too many glib monographs, most of which time will show to be highly subjective performances.’3 There are more Australian languages with some documentation (e.g. recordings and transcripts) than there are languages which have been the subject of academic research. Furthermore, while it could be expected that records of languages should result from academic research, this is not always the case. Prior to the late-1990s rise in interest in language documentation following Himmelmann’s (1998) article, it was normal linguistic fieldwork practice to ask questions, elicit paradigms and collect a few texts, but it was not considered important to create other useable records of the language (useable for researchers, or for the speakers and their descendants). Surprisingly, for example, there are languages with academic grammars written in the past thirty years for which there are no publicly available audio recordings.4 Of course there are a few exemplary researchers who have developed collections of records, but they are in the minority. In some cases, copies of recordings have been left in the community and thus can be distributed there, but they are often lost after a short time. There are two motivating principles for focussing on documentary language records. The first is the responsibility of the researcher to the people they have recorded: to provide useable records of the language, and to ensure they are available over time. This is particularly important in Australia where, as we will see below in the rest of the chapter, so little is recorded for even the best-known languages. As a result, the language records that have been created are all the more important, both for the descendants of the speakers who provided that information, and for the broader Australian society that needs to understand the diversity of languages spoken there over millennia. The second motivating principle is the researcher’s need to provide primary data in a form that can be used in their own analysis over time and then cited in any subsequent research. In Australia, colonial extraction of information from Aboriginal people was the norm for past research, with its focus on scientific outcomes and a lack of concern for primary documentation aimed at the speakers themselves. An example of the sentiment this attitude engendered among

3 Letter written by Edward Sapir to Fay Cooper-Cole, 25 April 1938, quoted in Darnell 1990: 139. 4 See the spreadsheet created to support the analysis in this chapter in Thieberger (2018a).

24

Aboriginal people can be found in the research guidelines published in 1982 by the Central Australian Aboriginal Congress: Aboriginal communities are saturated with research which is unrelated to their needs, which tends to be transitory, invasive, voyeuristic and dishonest and which provides no direct feedback, either with short term or long term benefits, to the relevant communities. (Central Australian Aboriginal Congress, 1982. Research Guidelines, Item 8)

Of course, there were also some humane and wellintentioned settlers who learned about the first people and who wrote records of their languages (for example Captain Collett Barker, whose diary is reproduced in Mulvaney and Green 1992). This first documentation is all the more valuable in places where the languages are no longer spoken today. As noted by McGregor (2008c: 6), early settlers had a need to learn about the country from Aboriginal people and so some effort was made to learn their languages, but that this interest ‘declined as the indigenous population declined and learnt English’. Blake (2016: 131) notes for south-east Australia that, ‘Almost half of the material we have on Indigenous languages was collected in the 1840s’, suggesting there was an initial interest that then waned. Part of that interest was manifested in the collection of responses to questionnaires sent out to elicit particular sets of vocabulary, occasionally including sample sentences. Examples of such surveys with wordlists to be filled out by local authorities include Curr’s (1886–1887) list of 125 words and Bates’s 1904 questionnaire of some 1,800 terms. Later, wordlists were published as prompts for fieldworkers (e.g. Sutton and Walsh 1979; Capell 1945) rather than being distributed to be filled out. From the perspective of linguistics, the past failure to be responsive in an appropriate way to the communities of speakers is being countered more recently by collaborative projects directed by speakers, and by the development of methods within linguistics that ensure that records are created in a form that allows them to be reused. In the past, analogue audio tapes could be copied, but this was a laborious task. An utterance on a tape could be cited, but it was not commonly done and was not simple. In both cases, access to recordings of utterances is easier if digital media is transcribed and then archived. Archiving with clear deposit conditions allows a user to know if they can get copies and does not rely on having to contact the original researcher with each request. The development of Indigenous language centres as local focal points for language resources similarly has the potential to increase language documentation both by making new records and by annotating existing records.

documentation of australian languages For a recent overview of the documentation of Australian languages see Koch, Austin, and Simpson (2016). McGregor (2008c: 1–26) summarizes earlier work and observes that there are three periods of research about Australian languages: the first up to 1930, the second from 1930 until 1960, and the third from 1960 up to the present. The first period saw the collection predominantly of wordlists, with some grammatical sketches and very few audio recordings, including wax cylinder recordings (probably the most famous of which are Fanny Cochrane Smith’s Tasmanian recordings).5 In the second period, there was more focus on structural features, resulting in grammatical descriptions, dictionaries, and some text collections. The third period includes the establishment of the Summer Institute of Linguistics, the Australian Institute of Aboriginal Studies, and university linguistics departments (Monash and ANU). The documentation produced by linguists has typically focussed on the kinds of grammatical or typological topics that were of interest at the time. Communities engaging in revitalizing or relearning their languages from these sources can be frustrated by the lack of everyday conversation, pleasantries, greetings, and leave-takings included in the documentation.

3.3 What records of Australian languages are there and how much is there for each language? The central register for information about Australian languages is Austlang,6 hosted by the Australian Institute of Aboriginal and Torres Strait Islander Studies (AIATSIS). For each language, Austlang provides information about the name, location, number of speakers, documentation, and classification. For our purposes, this documentation index, based on assessments done by AIATSIS, is a good first indication of the amount known per language. However, since it is not automatically generated from the AIATSIS catalogue, it needs an ongoing research effort to keep it up to date. Bearing that caveat in mind, a summary of Austlang’s documentation index can be seen in Figure 3.1. When analyzed at the end of 2017 (Thieberger 2018a), Austlang listed 1,215 Australian language varieties and provided a score for each language split into the following categories: wordlists/dictionaries (0–4 points); texts/stories (0–4 points); grammars (0–4 points).7 Figure 3.1 shows the 5 https://www.nfsa.gov.au/collection/curated/fanny-cochrane-smithstasmanian-aboriginal-songs 6 https://collection.aiatsis.gov.au/austlang/ 7 I thank AIATSIS for providing a table of the documentation scores, but note that Austlang includes an additional criterion, audio-visual (0-3 points), that were not provided to the author and so are not included here.

documentation scores for these varieties on a scale from 0 to 15 and the number of languages with those scores. To summarize the figures, 965 languages have a score of 5 points or fewer, 168 have between 6 and 10 points, and 108 have between 11 and 15 points. Only 12 languages have the maximum of 15 points. The Open Language Archives Community (OLAC) provides an aggregated set of records harvested from participating archives. OLAC is limited by relying on languages having an ISO-639-38 code (formerly known as an Ethnologue code), and, as noted earlier, many Australian languages listed in Austlang do not have these codes. Furthermore, Australian languages are not particularly well represented in OLAC archives because only two Australian archives, PARADISEC,9 and the Living Archive of Aboriginal Languages (LAAL),10 are members of OLAC.11 Nevertheless, it is instructive to see results provided by searching OLAC’s metadata service. The OLAC visualizer12 takes OLAC metadata and assigns values to the number of items per language. Of 238 Australian languages listed (as of June 2018), 15 have more than 100 items, 11 have between 51 and 100 items, and 144 have 10 or fewer items in an OLAC repository (as seen in Figure 3.2). The languages with the most resources listed in OLAC are presented in Table 3.1. The languages in OLAC with the most resources are listed in Table 3.1. Another source of information is Glottolog,13 provided by the Max Planck Institute in Leipzig, Germany. It has a slightly different focus to OLAC: it lists published sources that are added by human editors and thus is more comprehensive in this regard. A search for ‘Australian language dictionaries’ gives a total of 289. By contrast, the OLAC category ‘lexical resources’ has 138 items. A search of Glottolog for ‘Australia’ and ‘grammar’ or ‘grammar sketch’ gives 815 results, while the OLAC category ‘language descriptions’ has 234 items. It is not surprising that Glottolog and OLAC have different scores, given that OLAC lists archival resources (primary data files) while Glottolog is a curated list of known published or unpublished references per language.

8

https://iso639-3.sil.org PARADISEC (http://paradisec.org.au) currently (June 2022) has 9,300 items representing Indigenous Australian languages. 10 https://livingarchive.cdu.edu.au 11 The Aboriginal Studies Electronic Data Archive was part of OLAC but is no longer functioning. 12 This is a tool to harvest OLAC metadata and summarize it into a geographic representation, seen here: http://rebrand.ly/OLACviz. It is a coarse measure of the number of items per language in participating archives that is automatically generated every day. It makes no claims to assess the quality of the resources listed. It was built as part of my ARC funded project (FT140100214). It is discussed here: http://www.paradisec.org.au/ blog/2016/07/finding-what-is-not-there 13 https://glottolog.org 9

25

nicholas thieberger 568

600

Number of languages

500 400 300 200

148 102

100

69

0 0

1

2

3

39 4

30 5

29

40

6

7

30 8

37 9

31 10

20

19

22

11

12

13

12

12

14

15

Documentation score, range of 0 - 15

Figure 3.1 Austlang language documentation index. 250

OLAC items by number of languages 230

Number of languages

200

150

100 65 50

0

≤10

11 to 50

11

15

51 to 100

>100

Number of items

Figure 3.2 OLAC items by number of languages.

These broad overviews, while varying from each other, give us different ways of seeing the same basic point: that there are very few records of most Australian languages. They also point to the possibility of improving overviews, such as the one given in this chapter by including language codes in more repositories (discussed further in Section 3.4), with the primary benefit of allowing speakers of the languages and other researchers to find more information about each language. The next section deals with locating information for a particular language.

26

3.4 Where to look for records of Australian languages The first place to look for information in or about an Australian language is AIATSIS in Canberra. Their online catalogue14 lists most of the known records for all Australian languages. They also provide guides for searching 14

http://catalogue.aiatsis.gov.au

documentation of australian languages Table 3.1 Australian languages with the most resources in OLAC archives. Language

Number of resources Language

Number of resources

Burarra

194

Murrinhpatha

660

Djamindjung

444

Ngarinman

266

Eastern Arrernte 291

Pintupi-Luritja

345

Gunwinggu

192

Pitjantjatjara

118

Gupapuyngu

243

Tiwi

195

Gurinji

291

Warlpiri

1276

Iwaidja

365

Western Arrernte 113

Maung

205

the language collection.15 They can arrange access to the primary records if you are unable to travel to Canberra. A general guide to finding information in Australian languages and interpreting early sources about those languages, is the book Paper and Talk (Thieberger 1995/2005) produced following a workshop with the same name at AIATSIS in 1993. Despite its age, it is still a useful source, with sections on topics including: How to find your language name; Who recorded information about languages in the past?; Reading old sources; Interpreting old spelling; Making sense of the words in old wordlists; and What to do with old language material. Records created by linguists in the recent past should be relatively easy to locate as they are digital and should be in an accessible archive, or, if not, then the linguist should be able to provide accessible copies. Older records, including those produced by retired or deceased linguists can be more difficult to locate, particularly if they were never deposited in an archive or library. It can require some detective work to know that research has been done with a particular language, particularly if no publications came out of the work. While we could assume that a linguist’s fieldwork results in primary records, it was, as noted earlier, not normal practice to make records in the past. For example, Table 3.2 shows a list of grammars of Australian languages produced since 1970, together with Austlang scores lower than 10 (out of a possible 15). OLAC scores indicate that, for some languages, more exists in language archives than is accounted for by the Austlang score, for example Dhuwal and Dhangu, which have 13 and 21 points respectively. However, closer inspection reveals that the resources listed for these two languages

are early reading primers found in LAAL, and thus are not the primary records created during fieldwork. Consulting records held by researchers or by their executors relies on their willingness and availability to provide access. In the case of analogue records, access will involve having to travel to the place where the records are stored to look at them. For example, Arthur Capell’s collection of Australian material is held in 167 items at AIATSIS in Canberra. Peter Newton, the executor of estate, arranged to digitize 15,000 images of a further collection of papers representing over 300 non-Australian languages,16 otherwise only accessible at Newton’s house in Sydney. These papers, together with over 200 digitized audio recordings,17 in some 280 languages, are now freely available online, indexed in OLAC and findable via google. Access, mediated by appropriate conditions and acknowledgment of rights, is greatly facilitated by digitization if appropriate delivery mechanisms (websites, online catalogues, and so on) are provided by the collecting agency. Some institutions in Australia have been identifying what Indigenous language content they hold, for example the ‘Rediscovering Indigenous Languages’ project at the State Library of New South Wales has found materials in a number of languages that are now more accessible than previously.18 LAAL (mentioned above, Section 3.3) has been taking early readers produced in literacy production centres and schools in a number of Northern Territory languages and putting them online. In addition to the records held by collecting agencies, old language records can be unearthed from time to time,

15 The current URL for these guides is: https://collection.aiatsis.gov.au/ austlang/search (seen April 2021).

16 Many are non-Australian languages as can be seen here: http:// catalog.paradisec.org.au/collections/AC2 17 http://catalog.paradisec.org.au/collections/AC1 18 https://indigenous.sl.nsw.gov.au

27

nicholas thieberger Table 3.2 Grammars produced since 1970 of languages with an Austlang score of less than 10. Language

OLAC score

Austlang score

Language

OLAC score

Austlang score

Dhuwal

13

0

Bidyara

23

8

Nhanda

4

0

Gungabula

31

8

Dhangu

21

0

Warrgamay

5

8

Kuku-Uwanh

5

2

Wik-Ngathana

5

8

Yawarawarga

2

3

Kunjen

17

9

Wakawaka

4

3

Wandarang

9

9

15

5

Badimaya

4

9

7

7

Dieri

12

9

Nangikurrunggurr

12

7

Nyulnyul

6

9

Maranunggu

28

8

Madngele

24

9

Marithiel

38

8

Torres Strait Creole Kitja

for example, a tin trunk containing 3,000 manuscript pages by the anthropologist Ursula McConnel turned up in 2006 (Sutton 2010b). It contained records in a number of languages dating between 1918 and 1957. Another example is the massive collection of notes in multiple languages made between 1929 and 1931 by Gerhardt Laves19 , made public in 1983. Once manuscripts are found, the next question is how to make them discoverable and then accessible (with all necessary permissions in place). Searching for language names can be frustrating, as there are a number of possible spellings, even of the terms themselves. So, for example, searching the National Library of Australia’s catalogue for ‘Aranda’ finds many results, including people or suburb names, but only a few of which are related to the language name. Adding ‘dictionary’ to the search fails to find the major dictionary of that language, because it is listed as being in ‘Arrernte’. Many of the difficulties in discovering older sources that are already in a repository could be alleviated if existing repositories used language identifiers (like Austlang, Glottolog, or ISO-639-3) to allow targeted searching for just language names within their collections, and if these repositories then supplied their metadata to OLAC so that it could be searched via their aggregation service. The University of Hawai’i Library’s Pacific collection20 is an example of what can be done to add 19

http://www.anu.edu.au/linguistics/nash/aust/laves https://manoa.hawaii.edu/library/research/collections/hawaiianpacific/ 20

28

language codes to a catalogue and then make the collection available to OLAC’s system (Kleiber et al. 2018) and more broadly via search engines.

3.5 Accessing and presenting early sources on Australian languages As can be seen in Figure 3.3, while the bulk of published material about Australian languages has been produced since 1975, a significant amount was produced in the period before 1900. This often represents a major primary source for any given language, especially if nothing more was subsequently recorded. In the few cases where primary documents—fieldnotes, surveyors’ books, questionnaires, and so on—have been prepared for subsequent use by linguists, they were, quite reasonably given the focus of the research, typically edited for presentation with words extracted and inserted into a database (e.g. Noongar in Bindon and Chadwick 2011, Bunganditj in Blake 2003a). A problem with this approach is that the reader is not in a position to track from the published form of the words back to the manuscript to see original sources before decisions were made by the linguist about how to render them. As Dixon (1980: 15) notes for one much earlier example, ‘Unfortunately, Mathews tended to doctor and normalize his notes for publication, so that recourse must be had to the original field notebooks’. As Pascoe (quoted earlier) notes, original manuscripts were not prepared for access by major

documentation of australian languages 3500

3211

3000 2500 2000 1500 644

9 75

-1

99

4 19

50

-1

94 -1 25

97

9

4 19

-1 19

00

-1 75

168

92

9 89

87 -1 49

25

-1

84

4

55

9

4 18

18

00

-1

82

0 80 >1

47

4

18

2

0

623 273

18

500

19

1000

Figure 3.3 Publications on Australian languages over time. Carrington and Triffitt 1999.

linguistic projects or by collecting institutions, either as books, or from the 1980s, as digital files, when methods for presenting digital manuscript images together with texts became more common (e.g. Crane 1990). William Thomas’s diaries (Stephens 2014) are the work that Pascoe is referring to in the opening quote above and are similarly typed from a manuscript, which is referenced at a page-level in the published work, but the original manuscript itself is not published. As technologies permit linking between a page image of a manuscript and its textual version, it is not only possible to prepare manuscripts for production in books, as in these examples, but also to allow a further online publication of the same text with facsimile pages of the original source, linked at a page level. Due to the size of some manuscript collections, it is likely that the online version is the most practical way of delivering the material. Examples of this kind of work include Henderson’s (2008) detailed treatment of Laves’s Nyungar records (mentioned in the previous section), McGregor’s edition of Nekes and Worms (2006 [1953]), work on some of the Bates papers (Thieberger 2016a), and current work on the Fison and Howitt papers.21 The Dawes manuscript of an early vocabulary of the Sydney language has been put online with text and manuscript images.22 Textual rendering of images of typescripts (in contrast to handwritten manuscripts) is relatively simple to produce using optical character recognition tools. For example, a pdf and text of Brough Smyth’s vocabularies from south-eastern Australia are available via the Internet Archive.23 Nyingarn24

is an online platform for converting manuscripts and making the text and image available. It is being developed in 2021–2024 and will be housed at AIATSIS. Audio records require more work to digitize and make accessible, but it is critical that this work is done soon, before the tapes become unplayable or the playback machinery is no longer available. AIATSIS has put considerable effort into digitizing audio tapes and has the largest collection of Australian language recordings.25 Its focus has been on preservation, and it is planning to develop a digital platform for access. A recent example of creating a collection of records based on audio field-recordings is the work done by PARADISEC to digitize 265 hours of Ian Green’s recordings dating back to the 1980s.26 Funded by the ARC Centre of Excellence for the Dynamics of Language and organized by Rachel Nordlinger, the Daly languages recordings have been digitized and made available via a set of webpages27 and were taken back to the source communities in 2016 on USB sticks. This project shows how archival records can be presented with contextual information in the form of landing pages, as suggested by Woodbury (2014). The Daly Languages project was possible because the primary records were digitized and described by standard metadata in the PARADISEC catalogue. As each record in this catalogue is written out as an XML file to be stored together with the files it describes, each item then becomes a self-describing unit. Subcollections can be exported for delivery back to source communities, and each subcollection

21

http://howittandfison.org http://www.williamdawes.org, undated, but the book version is Nathan, Rayner, and Brown (2009). 23 https://archive.org/details/aboriginesofvict01smyt 24 https://nyingarn.net 22

25

https://aiatsis.gov.au/publication/117744 This article discusses this work in more detail: https://pursuit. unimelb.edu.au/articles/preserving-precious-indigenous-languages 27 http://www.dalylanguages.org 26

29

nicholas thieberger has its own catalogue harvested from these XML files. Primary records in this form are more comprehensible than a collection of files on a hard disk would be. The same routine is also used in setting up local Wi-Fi transmitters so that records can be delivered to mobile devices.28 We can foresee many novel ways in which archival sources will be found and reused in future.

3.6 Loss of documentation Records held in libraries and archives should be safe into the future, but there is a great deal of information produced in Australian languages that is at risk of being lost. This ranges from research recordings in filing cabinets, offices, or deceased estates, to the collections being made by cultural organizations. Small agencies, like language and cultural centres, are making recordings, transcribing them, and enriching archival records. Media services like ABC, NITV,29 and ICTV30 broadcast in Indigenous languages. Many languages are represented in social media pages, in YouTube and similar platforms (see for example the list of Indigenous Tweets31 ), but there is no repository storing all of these records of current language use. There is an effort to

28 http://www.paradisec.org.au/blog/2018/07/local-wifi-versions-ofparadisec 29 National Indigenous Television: www.sbs.com.au/nitv/ 30 Indigenous Community Television: https://ictv.com.au/ 31 http://indigenoustweets.com

30

determine the location of analogue media collections32 , but no national effort is yet in place to capture other productions in Australian languages.

3.7 Conclusions It is salutary to note that Australian languages do not feature in Hammarstro¨m’s (2010) catalogue of the least documented language families in the world, but, as was made clear in this chapter, while some Australian languages have been reasonably well documented, most have not. A task ahead of us today, if we are to respond to Pascoe’s challenge, is to find these primary sources and to convert them into formats that make them accessible for the communities they represent. In this way, these documents can be revived and, in turn, help to inspire the continued use of the languages they record.

Acknowledgements Thanks to Adam Bell at AIATSIS for providing information about Austlang and to Robert Forkel for the Glottolog information.

32 The lost and found survey (http://www.delaman.org/project-lostfound/) asks about such collections and has successfully digitized tapes as a result.

chapter 4

Australian languages and syntactic theory Rachel Nordlinger

4.1 Introduction Australian language research has made a significant impact on the development of linguistic theory over the last 50 or so years, much of which has been discussed in detail in various chapters in the edited volume by Koch and Nordlinger (2014). In this chapter, I provide a brief overview of the key areas in which Australian languages have featured in theoretical discussions of (morpho)syntax. My purpose here is not to revisit the details of the theoretical analyses offered for different phenomena, but rather to focus on the empirical issues relating to (some) Australian languages and point the reader to some of the relevant literature discussing the implications for different syntactic theories.1 The presence of Australian languages on the international theoretical stage was prompted primarily by Ken Hale’s work on Warlpiri (Pama-Nyungan, Central Australia) (e.g. Hale 1976a; 1981a; 1982a; 1983; 1984a; 1989b; 1992a) and R. M. W. Dixon’s work on Dyirbal and Yidiny (PamaNyungan, northern Queensland) (Dixon 1972; 1977a; 1979b; 1994). The essence of Hale’s work was to show that Warlpiri (and by extension other dependent-marking Australian languages) doesn’t have the same syntactic nature as more familiar languages in the areas of: phrase structure constituency (including the structure of the clause, and the existence of VPs), subordination, wh-movement, and extraposition. Dixon’s work on Dyirbal (and by contrast, Yidiny) showed that the familiar nominative–accusative syntactic alignment of English and many other languages was not universal, and that linguistic theory needed to be able to incorporate the phenomenon of syntactic ergativity as well. Each of these bodies of work led to great debate and a copious literature on the extent to which Australian languages can be analysed as underlyingly similar to other configurational and nominative–accusative languages in syntactic terms, or whether they (or, at least some of them) really constitute a fundamentally different language type. Such debate 1 This discussion draws in part on the discussion of these issues in Nordlinger (2014a), to which the reader is referred for further details.

thus relates to the very nature of human language, and thus has had enormous significance for the development of syntactic theorizing. In this chapter, I focus on four major areas in which data from Australian languages has made significant contributions to discussions of syntactic theory: phrase structure and nonconfigurationality (Section 4.2), ergativity (Section 4.3), case stacking (Section 4.4), and subordination (Section 4.5). Other areas in which Australian language data has featured in the theoretical syntactic literature include noun incorporation, polysynthesis, complex predicates, and templatic morphology; these are briefly surveyed in Section 4.6.

4.2 Phrase structure and nonconfigurationality One major area in which Australian languages have significantly impacted theoretical debate is the analysis of phrase structure and nonconfigurationality. This debate stemmed originally from Hale’s work on Warlpiri (e.g. Hale 1981a; 1983) in which he argued that Warlpiri is nonconfigurational, with a ‘highly permissive’ phrase structure (Hale 1983: 10), fundamentally different from more familiar configurational languages such as English. He focussed particularly on three primary characteristics of Warlpiri syntax, taken to be central to nonconfigurational syntactic structure: free word order, null anaphora, and discontinuous NP constituents (see Brody, Chapter 35, this volume). ‘Free word order’ is demonstrated by the fact that there are no fixed positions in the phrase structure for major clausal elements such as the subject, the object, and the verb, which can appear in any order with respect to each other and still produce a grammatical sentence. Thus, whereas a configurational language like English has fairly strict syntactic positions associated with key clausal elements, so that notions such as ‘subject’ and ‘object’ can be defined in terms of

Rachel Nordlinger, Australian languages and syntactic theory. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Rachel Nordlinger (2023). DOI: 10.1093/oso/9780198824978.003.0004

31

rachel nordlinger a phrase structure configuration, in a language like Warlpiri this is not the case (1). (1) Warlpiri (Hale 1983: 6) a. Ngarrka-ngku ka wawirri panti-rni. man-erg prs kangaroo spear-npst ‘The man is spearing the kangaroo.’ b. Wawirri ka panti-rni ngarrka-ngku. c. Panti-rni ka ngarrka-ngku wawirri. etc. ‘Null anaphora’ refers to the fact that NP arguments are not obligatory and are freely omitted, thus allowing for the possibility that core grammatical relations such as subject and object are not represented with overt material in the phrase structure (2). And the possibility of ‘discontinuous NP constituents’, where two (or more) related nominal elements occur distributed throughout the clause rather than forming a single NP constituent in the phrase structure (see also Blake 1983), highlights the phrase structure flexibility since we find that constituents of the same grammatical function need not even appear contiguously in the clause (3). (2) Panti-rni ka. spear-npst prs ‘He/she is spearing him/her/it.’ (Hale 1983: 7) (3) Wawirri kapi-rna panti-rni yalumpu. kangaroo aux-1sg.s spear-npst that ‘I will spear that kangaroo.’ (Hale 1983: 6) While Hale’s work focussed primarily on Warlpiri, related work showed that many of these syntactic features can be found in other Australian languages as well, including Kalkatungu (Blake 1983), Wubuy (Heath 1986), Wambaya (Nordlinger 1998b), Jiwarli (Austin 2001) and Bardi (Bowern 2005). While there can be no denying that the surface syntax of a language such as Warlpiri is different to that of a language like English, there has been considerable debate in the theoretical literature as to the nature and source of this difference and its theoretical implications. This variation is partly attributable to the different theoretical perspectives of the researchers, since the assumptions of different theoretical frameworks lead to different empirical facts being relevant to the debate. For example, Hale (1989b, and also Speas 1990) later decoupled the issue of free word order from nonconfigurationality, instead arguing that the definition of nonconfigurationality is more concerned with the mode of expression of arguments (e.g. the possibility of extensive null anaphora and bound pronominal agreement). However, for other researchers (e.g. Heath 1986; Laughren 1989; Simpson 1991; Austin and Bresnan 1996; Nordlinger 1998b), the fundamental issue is that a nonconfigurational

32

language does not use phrase structure for the identification of core grammatical relations such as subject and object. Consequently, nonconfigurational languages will show no evidence for a VP constituent which distinguishes the subject (external to the VP) from the object (internal to the VP), and therefore no structural asymmetries between subject and object. Such arguments have been made for Warlpiri (Laughren 1989; Simpson 1991; Austin and Bresnan 1996) and other Australian languages such as Wubuy (Heath 1986), Wambaya (Nordlinger 1998b), and Bardi (Bowern 2005). An alternative view considers Warlpiri and other nonconfigurational languages to have the same underlying phrase structure as configurational languages (including a VP constituent) and the surface differences to arise from other linguistic properties. Extremely influential in this respect was Jelinek’s (1984) pronominal argument hypothesis (see also Speas 1990, and the ensuing debate involving Simpson 1991; Austin and Bresnan 1996; Nordlinger 1998b; Austin 2001; Baker 2001; Legate 2001; 2003; Pensalfini 2004). Jelinek argued that the apparent nonconfigurational properties discussed by Hale (1983) arise not from different phrase structure possibilities, but from the fact that the argument positions in a language like Warlpiri are filled by the bound pronominal clitics, rather than by NPs, which are instead adjuncts. Given that NPs are adjuncts and not arguments, their lack of fixed phrase structure positions, freedom of internal ordering, and optionality follow straightforwardly and do not need to be attributed to fundamental differences in phrase structure (see Baker 2001 and Pensalfini 2004 for modifications of this general approach). Jelinek’s analysis is appealing in that it allows for nonconfigurational languages to be easily assimilated into a configurational framework, but it runs into a number of empirical problems, as discussed in detail by Austin and Bresnan (1996) and Nordlinger (1998b). The most serious of these include the fact that there are languages such as Jiwarli (Austin 2001) that show all of the nonconfigurational properties of Warlpiri without having any bound pronominal clitics at all; and languages like Bininj Gun-wok (Evans 2002) and Bardi (Bowern 2005) where there are strong empirical reasons not to treat the verbal agreement morphology as filling clausal argument positions (see also Section 4.6).

4.3 Ergativity The large majority of Australian languages are morphologically ergative, which means that they have an ergative case marker which encodes a transitive subject, distinct from an intransitive subject (see Dunn and Meakins, Chapter 20, this volume). Much of the theoretical discussion about ergativity

australian languages and syntactic theory has centred on the extent to which such morphological facts reflect underlying syntactic patterns, in particular whether morphologically ergative languages can be shown to have a unified subject relation in their syntax, grouping both transitive and intransitive subjects irrespective of the distinct morphological encoding. On the basis of syntactic tests for subjecthood such as raising, equi-NP deletion, reflexive binding, and control structures, the large majority of morphologically ergative languages can be shown to have nominative–accusative syntactic patterns in the same way as English and other familiar languages (Anderson 1976; Dixon 1979b; 1994). The Australian language Dyirbal, however, has been argued to work differently (Dixon 1972; 1979b; 1994). Dixon showed that in Dyirbal, syntactic processes such as the omission of coreferential nominals in coordinate clauses group intransitive subjects with transitive objects, rather than with transitive subjects. Thus, whereas a language with accusative syntax like English has an A/S2 pivot (4a, 5a), Dyirbal operates according to an S/P pivot (4b, 5b):3 (4) a. Mother(A) saw father(P) and __ (S) returned. (mother returns so S = A) b. ŋuma yabu-ŋgu bura-n father(abs) mother-erg see-nfut banaga-nyu. return-nfut ‘Mother(A) saw father(P) (and) __(S) returned.’ (father returns so S = P) (5) a. *Father(S) returned and mother(A) saw __(P). (P ≠ S) b. ŋuma banaga-nyu yabu-ŋgu bura-n. father(abs) return-nfut mother-erg see-nfut ‘Father(S) returned and mother(A) saw __(P).’ (P = S) In (4) we see that the omitted S argument in the second clause must be interpreted as coreferential with the transitive subject (A) argument of the first clause in English (4a), but with the transitive object (P) argument in Dyirbal (4b). In the examples in (5) we find that, while it is ungrammatical in English for an omitted P argument in the coordinated clause to be coreferential with the intransitive subject in the first clause (5a), this is perfectly fine in Dyirbal (5b). According to Dixon (1972; 1994), all major syntactic operations in Dyirbal, including relativization and complementation as well as coordination, group S and P together in this way. This has implications for analyses of grammatical relations, and in particular the universality of a notion 2 Following Dixon (1979b) initially, A, S, and P refer to transitive subject, intransitive subject, and transitive object, respectively. 3 Dyirbal examples are taken from Dixon (1994: 12–13).

such as ‘subject’ if it is understood as a grouping of A and S. Some have argued (e.g. Marantz 1984; Kroeger 2004) that the syntactic evidence points to an analysis whereby ‘subject’ in Dyirbal consists of a grouping of the S of intransitive clauses and the P of transitive clauses. This allows us to maintain cross-linguistic generalizations regarding syntactic processes such as relativization and complementation: we can assume that these processes consistently identify ‘subjects’, but that this will correspond to a category that includes the agent of a transitive verb in languages with nominative–accusative syntax, and the patient of a transitive verb in syntactically ergative languages like Dyirbal. Dixon (1979b) notes, however, that even in Dyirbal there are syntactic operations such as control relations with jussive complements, and the addressee of imperative commands, that call on a grouping of A and S (i.e. the traditional ‘subject’) rather than S and P. He therefore argues for a distinction between ‘pivot’ and ‘deep’ subject, where the former is relevant to syntactic processes such as relativization, and the latter is a semantic notion for a class of NPs that can control events. The ‘deep’ subject groups A and S in all languages, irrespective of whether their syntax is accusative (A/S pivot) or ergative (S/P pivot). This idea that syntactic operations apparently targeting ‘subject’ may in fact be split in syntactically ergative languages across two different constituents has been taken up in different ways in theoretical analyses of ergativity. Working within a movement-based Chomskyan framework, Bittner and Hale (1996a, b) see this as relating to the fact that syntactically ergative languages require S/P arguments to move via A’-movement into a structural position higher than the transitive subject to satisfy the Case Filter. Working within the framework of Lexical-Functional Grammar, Manning (1996) argues for an analysis of ergativity which distinguishes grammatical subject from argument structure subject, providing a unified account for syntactically ergative languages and the complex voice systems of Philippine languages. Legate (2008; 2012) draws on data from Warlpiri, Dyirbal, and other Pama-Nyungan languages to develop a theory of case in the Minimalist framework that can accommodate both nominative and ergative languages in a theoretically consistent way.

4.4 Case stacking Case marking in Australian languages is remarkable for its multifunctionality, a fact first brought to the attention of the field by Dench and Evans (1988) (cf. Simpson, Chapter 21, this volume). This multifunctionality, combined with often prolific case agreement properties, results in some Australian

33

rachel nordlinger languages having ‘case stacking’, whereby a single nominal can carry more than one case marker at a time, reflecting different levels of syntactic structure. Although instances of ‘double case’ or Suffixaufnahme can be found in other languages of the world (Plank 1995), in Australia we find the phenomenon in its most extreme and complex form. Dench and Evans (1988: 35) provide examples such as the following: (6) Panyjima Ngatha pilanyjayi-nha nyinku mirta-yu 1sg.nom frightened-pst 2sg.acc not-acc paka-rnu-ku ngalimpa-tharntu-karta-ku come-rel-acc 1du.incl-gen-all-acc yurlu-karta-ku. camp-all-acc ‘I was frightened you weren’t coming to our camp.’ (7) Kayardild Maku yalawu-jarra yakuri-naa woman catch-pst fish-abl dangka-karra-nguni-naa mijil-nguni-naa. man-gen-inst-abl net-inst-abl ‘The woman caught fish with the man’s net.’ In (6), we find the most heavily marked nominal ‘ngalimpa-’ inflected with three case markers. The first—GEN—marks its adnominal role as possessor (‘our camp’), the second— ALL—shows agreement with the head nominal of the allative NP (‘to our camp’) and the third—ACC—is an instance of c-complementizing case (Dench and Evans 1988), which appears on all members of the subordinate clause in agreement with the controlling accusative NP in the main clause (‘you (weren’t coming) to our camp’). In (7) we see dangka- also inflected with three cases: genitive to mark the possessor relation, instrumental in agreement with the head noun ‘net’, and the ablative case which works together with the verbal tense/mood marking to encode the tense/mood of the clause and is required on all non-subject constituents (see Evans 1995c for a detailed discussion of this ‘modal’ function of case in Kayardild). Double case marking on genitive NPs is not unusual among languages of the world, but it appears that case stacking to the degree we see in (6) and (7) is only found in languages of Australia and challenges standard assumptions that are made about case in many theoretical frameworks. For example, case assignment is generally considered to be a local phenomenon, assigned under government by a head to its dependent. On this view, there is no simple explanation for how a nominal as deeply embedded as ngalimpa- in the Panyjima example in (6) can carry accusative case which has been assigned by the verb ‘frightened’ to the nominal nyinku, three NPs higher up in the syntactic structure.

34

Most of the theoretical work on case stacking has been done within the framework of Lexical-Functional Grammar (Simpson 1991; Andrews 1996; Nordlinger 1998b; Sadler and Nordlinger 2004). Nordlinger (1998b) shows how case stacking, along with other case-related phenomena in Australian languages such as the use of case to encode tense/mood in Kayardild, follows naturally from a model of constructive case. In this model, case morphology carries information about the larger syntactic context in which it appears, and is therefore able to constrain levels of syntactic structure beyond that of the nominal to which it is attached. There are a handful of analyses in other theoretical frameworks. Malouf (2001) provides an analysis within the framework of Head-Driven Phrase Structure Grammar which accounts for case stacking through a principle of Case Concord, which propagates the case of the head onto all of its dependents (including some adjuncts), after any locally assigned case. Thus, case stacking arises as a result of higher locally-assigned cases cascading down the structure to more deeply embedded constituents. Richards (2013) proposes some revisions to assumptions about case assignment and its interaction with Spell-out in the Minimalist framework to provide an account of case stacking in the Tangkic language Lardil. The implications of case stacking for morphological theory are addressed in different ways by Sadler and Nordlinger (2006) and Round (2013).

4.5 Subordination The discussion of subordinate dependent clauses in Australian languages, both in descriptive grammars and in the theoretical literature, has been heavily influenced by Hale’s (1976a) paper on the ‘adjoined relative clause’—a nonembedded, multifunctional subordinate clause type found in Warlpiri and a ‘large number of Australian languages’ (p. 78). This paper has remained the authoritative work on subordination in Australia; virtually every Australian grammar written since contains some reference to this clause type, and it features prominently in any discussions of subordination in Australian languages (Merlan 1981b; Austin 1988a; Nordlinger 2006b; Van Egmond, Chapter 37, this volume) as well as typological and theoretical discussions that include reference to Australian languages (e.g. Comrie 1981b; Keenan 1985; Lehmann 1988; Diessel 2001; Andrews 2004; Cristofaro 2005, etc.). Hale (1976a) uses the label ‘adjoined relative clause’ to refer to a clause type that can have both adverbial and relative functions and which, in relative clause function, may appear in a linear position separated from the head nominal that

australian languages and syntactic theory it modifies. The syntactic properties of this clause type are reflected in this oft-cited quote: In a large number of Australian languages, the principal responsibility for productive recursion in syntax is shouldered by a structure which I will refer to as the adjoined relative clause. It is typically marked as subordinate in some way, but its surface position with respect to the main clause is marginal rather than embedded—hence the locution ‘adjoined’. Typically, but not invariably, it is separated from the main clause by a pause. (Hale 1976a: 78)

Hale provides examples such as the following from Warlpiri. (8) Ngajulu-rlu rna yankirri pantu-rnu, 1sg-erg aux emu spear-pst [kuja-lpa ngapa nga-rnu]. comp-aux water drink-pst ‘I speared the emu which was drinking water.’ ‘I speared the emu while it was drinking water.’ (Hale 1976a: 78) In this construction the subordinate clause is fully finite; the only marker of subordination is the initial complementizer kuja. This subordinate clause has (at least) two available interpretations: in the first, it is interpreted as a relative clause (Hale’s NP-relative interpretation); and in the second as a temporal adverbial clause (Hale’s T-relative interpretation). Furthermore, the subordinate clause is not ‘embedded’ within the main clause—by which Hale means it appears on the edges of the main clause, and is ‘never flanked by material belonging to the main clause’ (Hale 1976a: 86). This is true even when the clause has only a relative clause function, meaning that it need not form a constituent with the head noun that it modifies: (9) Ngajulu-rlu kapi-rna wawirri purra-mi, 1sg-erg aux kangaroo cook-npst [kuja-npa pantu-rnu nyuntulu-rlu]. comp-aux spear-pst you-erg ‘I will cook the kangaroo you speared.’ (Hale 1976a: 79) Hale’s claims have been frequently interpreted in the typological and theoretical literature as suggesting that Warlpiri (and other Australian languages) don’t have syntactic embedding—i.e. that Australian languages do not have ‘true’ subordinate clauses of the type that are familiar from many other languages (e.g. Lehmann 1988: 183–5; Diessel 2001: 439–40). Nordlinger (2006b) addresses these misperceptions and argues that it is possible for a construction to share all of the properties of Hale’s (1976a) ‘adjoined relative clause’ and yet be demonstrably syntactically subordinate. In fact, while the multifunctional property of the

‘adjoined relative clause’—namely, the availability of both relative and adverbial interpretations—appears to be common across Australian languages, there is ample evidence for the existence of subordination in many of these languages, as has been shown in numerous studies of individual languages (Dench 1988; 2006; Evans 2006c; Legate 2011; McConvell 2006a; McGregor 1988c; 1994a; Nordlinger and Saulwick 2002). The nature of subordination and the possibilities for embedding in Australian languages have also led to their discussion in the context of recent theoretical debates about the nature of the human language faculty and the centrality of recursion (e.g. Hauser, Chomsky, and Fitch 2002; Fitch, Hauser, and Chomsky 2005; Traxler, Boudewyn, and Loudermilk 2012). If Hale’s statement quoted above is taken to suggest a lack of syntactic embedding, then Australian languages are a potential counterexample to the claim that recursion is central to human language (Levinson 2013). However, much debate exists as to whether this is the proper interpretation of recursion as intended by Hauser et al. (2002) (Watumull et al. 2014); and whether this is the correct interpretation of Hale (1976a) (Nordlinger 2006b; Legate, Pesetsky, and Yang 2014). Irrespective, there can be no denying the impact that Hale’s (1976a) paper has had on the typological and theoretical literature on this topic.

4.6 Conclusion In the above discussion, we have seen that Australian languages have featured prominently in theoretical debates on the topics of phrase structure and nonconfigurationality, morphological and syntactic ergativity, case stacking, and subordination. While these are perhaps the areas in which discussion of Australian languages has been most prominent in the syntactic theoretical literature, there are many other grammatical domains in which data from Australian languages, and work by Australianist researchers, has been influential. The polysynthetic Gunwinyguan languages, especially Mayali (Bininj Gun-Wok), Rembarrnga, Ngandi, and Nunggubuyu (Wubuy), feature strongly in Baker (1996), in which he argues for a single macro-parameter from which all the properties shared across polysynthetic languages can be derived. Evans (2002) argues against the view advocated by Baker (1996), and building on Jelinek’s (1984) treatment of nonconfigurationality discussed in Section 4.2, which treats the verbal agreement markers in polysynthetic languages as the clausal (pronominal) arguments. The Jelinek/Baker approach allows polysynthetic languages, although quite

35

rachel nordlinger different on the surface, to be treated as essentially identical at another level of structure to languages like English, where these arguments are expressed with full NPs. On this view, the verbal element ba- in (10) fully instantiates the subject and object arguments, while the external nominals are adjuncts, as reflected in the English translation (Evans 2002: 25): (10)

Bininj ba-ngune-ng duruk. man 3/3lP-eat-pp dog ‘The man, he ate it, the dog.’ (Bininj Gun-Wok)

Evans (2002) argues that this analysis does not adequately capture the Bininj Gun-wok facts, since there are contexts in which verbal agreement morphology behaves differently to pronouns. Similar issues are raised by Baker (2002a) for Ngalakgan, which shows further complexity in allowing systematic ‘disagreement’ between the noun class of verbal pronominal agreement and that of external co-referential NPs. Discussions of Australian languages have also featured in typological discussions of polysynthesis (Fortescue, Mithun, and Evans 2017; Evans 2017b; Nordlinger 2017); noun incorporation (e.g. Mithun 1984a; Baker 1988; Evans 1996; Nordlinger and Sadler 2008; Baker et al. 2010); complex predicates (Evans 1997c; Wilson 1999; Nordlinger 2010a); and templatic morphology (Simpson and Withgott 1986; Nordlinger 2010b).

36

Australian languages show a number of interesting morphosyntactic properties that have informed our understanding of the structure of human language and its theoretical analysis. Over the past few decades as the descriptive work on Australian languages has increased in both depth and breadth, and information has become more accessible, the contributions these languages have made to syntactic theory have become more substantial, further reinforcing the important position they play in our understanding of language and how it works.

Acknowledgements My understanding of the complexities and riches of Australian languages has been fostered over many years by my many highly skilled language teachers from a variety of different Indigenous communities. I extend my heartfelt appreciation and admiration to all of them, and to their fascinating languages. My research has been financially supported by the Australian Research Council, most recently through the ARC Centre of Excellence for the Dynamics of Language (Project ID: CE140100041).

chapter 5

Australian languages and interdisciplinary approaches to the past Luisa Miceli and Claire Bowern

5.1 Introduction 5.1.1 Overview Many chapters of this book describe the variation across the languages of Australia, the social conditions under which languages are used, and how these change over time. In contrast, this chapter provides a discussion on the use of linguistic data to answer questions concerning the broader Australian past. It is not our aim to provide a survey of the historical linguistics of Australian languages; there are already substantive surveys of this type (Blake 1988; Dixon 2002; Evans 2005; McConvell and Bowern 2011). Rather, we examine the kind of contribution that linguistics can make in building a broader narrative of the deep history of the region, and to consider the implications of mismatches between lines of inquiry stemming from different disciplines—as well as within a single discipline. What do we do when the linguistic data tell one story, but the genetic or archaeological data suggest a different pattern of events? As a first step, we therefore examine some of the linguistics-internal issues that must be grappled with to ensure that genuine progress on an integrative understanding of the Australian past can be made. Australian comparative linguistic patterns are not easy to work with; we do not have a long history of detailed ‘established wisdom’; and there are some contradictions that need to be better understood before linguistic data and findings can be adequately incorporated in the reconstruction of the far past of Sahul. We then move on to interdisciplinary issues. Linguistics is one way—along with archaeology, genetics, and anthropology—of discovering pieces of the deep past, but each of these disciplines makes different assumptions on cultural and population processes, uses evidence in different ways, and provides insights with different levels of certainty and granularity for different time periods. When the different lines of inquiry disagree (as disagree they inevitably do), what do we make of it? And how do we go about reconciling these differences, if possible?

5.1.2 An aside: the Australian past as ‘deep history’ Our discussion of time scales, granularity, and assumptions calls to mind analogous debates in the field of history. Smail (2007; see also Shryock, Smail, and Earle 2011) has made an argument for ‘deep history’ beyond disciplinary boundaries. This work argues for a history based not just on the written record, but one which is truly integrative, taking data from a variety of sources and trying to cross-validate. It is also a history that pays attention to psychology and to Indigenous perspectives (McGrath and Jebb 2015). We might think of deep history as the type of work that gets us back into the far past by understanding people’s lives and behaviours; that work originates with scholars such as Elizabeth Barber (Barber 1994). Within the Australian Indigenous context, the study of the past has always been deep history, since we have never simply relied on the written record. The written record does not contain much information about most of the things we might want to study, since it tends to deal with the colonizers and their relationship to Indigenous people. Even where the written record is ethnographic and describes indigenous lifeways, that record is created overwhelmingly by community outsiders. This, coupled with the shallow time depth of Australian written records, means that Australianists have always looked to interdisciplinary and ‘deep history’ methods to understand the ‘history’ of the continent, a point clear from publications such as Jones (1971), or even, perhaps, the early ‘expeditions’, which included linguistic, ethnographical, archaeological, and biological experts (Haddon 1935). As Bowern (2012c) has pointed out, the linguistic record in many parts of Australia is possibly a more revealing record of the mid to late Holocene than the archaeological record, given the perishability and fragility of such a large part of the material cultural record. This makes the task of refining our ability to interpret the comparative patterns we find in the linguistic record all the more important. Those patterns

Luisa Miceli and Claire Bowern, Australian languages and interdisciplinary approaches to the past. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Luisa Miceli and Claire Bowern (2023). DOI: 10.1093/oso/9780198824978.003.0005

37

luisa miceli and claire bowern are a unique window into the past social networks of Aboriginal society, but we must take the time to better understand the processes that have given rise to them, rather than continuing to rely on traditional assumptions which may or may not be appropriate in this context—e.g. the assumption of a clearly identifiable L1 in intergenerational transmission within the speech community (Miceli 2019). Although our understanding of the Australian past must be approached from this interdisciplinary perspective, we should be fully aware of the danger as well as the opportunity of ‘building on each other’s myths’, using Renfrew’s (1987: 287) often cited expression. We must be aware of changing perspectives and ground truths within disciplines and be prepared to adapt and reconsider the bigger picture as our understanding progresses. With this in mind, we first bring the spotlight on some of the ongoing debates within linguistics before presenting an interdisciplinary case study.

5.2 Problems within linguistics 5.2.1 Missing information To what extent does scarcity of data affect the conclusions we can draw about the Australian past? All records of the past are imperfect, of course (cf. Sober 1991), and so missing information is not a problem unique to Australia. This section discusses some of the ways in which the linguistic record is imperfect, and the implications of that missing information for reconstructing the past. First: we do not have records of all the languages spoken in Australia at the time of European invasion. Some are known simply as names on a map: Nyarti, Wadikali, Pirlatapa (Austin 1990b). Others are attested from short wordlists, but not in any detail, such as Ngumbarl (Nyulnyulan), Bigambal, and (though to a lesser extent) the Tasmanian palawa varieties. Secondly, most (though not all) of the documentation of these languages has been made by people who were neither native nor fluent speakers of the languages. In some cases, they learned the language as part of making the documentation, or they wrote down the language(s) over a long period (e.g. Reuther at Killalpaninna Mission; cf. Stevens 1994). In other cases, the linguists making records of the language worked only over a short period. Even if the resulting documentation is extensive and they were productive in the time they had available, it’s not surprising that they missed aspects of the language. Gerhardt Laves, for example, spent only a few months on Sunday Island working with Bardi people, and even though he produced an extensive text collection of tens of thousands of words, there are signs in

38

the corpus of mistakes and imperfect understanding. After all, languages are complex systems, and it is not surprising that nuances are lost when the recorder is not a fluent speaker.1 Thirdly, there’s the question of what the makers of documentation have recorded. The focus has been on wordlists and narratives, on grammar, and then on various aspects of material culture (cf. Musgrave and Thieberger 2021, discussing the ‘Boasian triad’ of grammar, dictionary, and texts in an Australian context). We have little record of conversation before the last 15–20 years, for example. Others have made the point in the context of language reclamation and revitalization that linguists have tended not to record the types of information that are crucial for basic language reclamation activities, like greetings, interactions around asking and answering, and the like. For historical linguistics, this means that work involving small amounts of basic vocabulary can involve data sources from many more linguistic varieties than work involving conversational data or syntax and morphology, which can only include well-recorded varieties.

5.2.2 Sound change More problematic within linguistics is the issue of sound change in Australian languages (or rather lack thereof). Many linguists have commented (more or less critically) on the apparent phonological similarity of many languages across the country, both within the Pama-Nyungan family and in the other, Northern, families. Round (Chapters 10–13, this volume), along with earlier work by Busby (1980), Hamilton (1996b), and Gasser and Bowern (2014), goes some way towards quantifying the extent of the similarity and the degree to which these languages do show similarities. For example, Round’s chapters in this volume quantify the inventory similarities but also show the ways in which Australian languages differ, such as in the resolution of nasalstop cluster sequences. Moreover, despite the similarity in inventories, and some similarity in phonotactics, there is still a phylogenetic signal in the phonological differences overall, (Macklin-Cordes, Bowern, and Round 2021) which allows us to see something like regular change in the lexicon as a whole, even though individual correspondence sets are not identified in this method. That is, if sound changes were proceeding haphazardly across languages or if there was no sound change at all, there are nonetheless methods that permit certain types of phylogenetic determination. 1

See Stockigt (Chapter 2, this volume) for further discussion.

australian languages and interdisciplinary approaches to the past The similarities in synchronic phoneme inventories do not necessarily mean that there has been no sound change. Indeed, we have evidence for sound change in three different areas. First, there are temporally recent changes which have resulted in dialectal variation (see e.g. Austin 1981b for variation within the Kanyara and Matharta dialect groupings, or the phonological differences between dialects of Bardi discussed in Bowern 2012a). Secondly, there are sound changes which have resulted in morphophonological alternations (see Round, Chapters 12 and 13, this volume), either in subgroups or in individual languages. Finally, there are sound changes identified on the basis of comparative reconstruction within Pama-Nyungan and its subgroups (Hercus 1979: Black 1980; Austin 1981c; 1988c, d; 1990a; 1997a; Koch 1997b; Alpher 2004a, b) and within some non-Pama-Nyungan families (for example Harvey 2003c; Round and Evans n.d.). However, apart from Paman, Arandic, and isolated other examples, the majority of these sound changes do not seem to result in the creation of new segments. They radically alter the phonotactics of some languages (Mbabaram (Dixon 1991b), Nganyaywana/Anewan (Crowley 1976), Arandic (Koch 1997b), for example), but their role in creating new phonemic contrasts is limited. This situation poses an interesting challenge for theory in historical linguistics, since systematically divergent correspondence is the most crucial type of evidence in establishing genetic relationships and in establishing whether individual lexical items are cognates, under the traditional comparative method (Miceli 2019; Miceli and Round 2022). Within Pama-Nyungan, the subgroups which show the most sound change in terms of the creation of new phonological contrasts are Paman (Hale 1976d; Black 1980) (see also Verstraete, Chapter 72, this volume) and Arandic (Koch 1997b). Yet in contrast to the local-level sound change, we also see claims for cognate material that appears, apparently unchanged, across very deep and remote linguistic relations. For example, the monosyllabic verbs identified and discussed by Merlan (1979), Dixon (1980), McGregor (2002), and others, if truly cognate (in the sense of descended from a shared common ancestor as a result of unbroken, intergenerational transmission; cf. Harrison 2003), must descend from a shared common ancestor that predates PamaNyungan. However, this raises a contradiction. Why should these verbs be identical (or close to identical) and yet the rest of the vocabulary has changed beyond recognition of cognacy or has been replaced? Are these forms morphologically reinforced? Or is change ‘undone’ through analogical restoration? Or are they extremely resistant to lexical change? How does one part of a language seemingly resist regular sound change at the same time as other parts do not? Anyone who proposes these long-distance cognates

needs to have a good explanation that accounts for both the presence of change as well as the apparent conservatism in these forms. Important clues pertaining to the Australian past are likely to be preserved in these intriguing Australian sound patterns and it is imperative that we develop a better understanding of the diachronic processes that have led to them in order to more adequately interpret the broader history. For a more detailed discussion of issues related to sound change in Australian languages we refer the reader to Miceli and Round (2022) and Bowern (2022).

5.2.3 Language contact Another point of discussion in the literature relates to the role of language contact in the history of Australian languages. Although the situation was by no means uniform at the time of European colonization, multilingualism was common in Australia. It is therefore reasonable to assume that its prevalence extended into the past. Accepting this assumption means acknowledging the likelihood that language contact played a significant role in the history of these languages, but there are different views on both the extent of its significance and the nature of the language contact effects. Dixon (2002) interprets the patterns of distribution of widespread linguistic elements (whether similar lexemes, shared grammatical categories, or strategies) as reflecting diffusion. Other Australianists (e.g. Evans 2005) interpret the same features as inherited. In both cases, diffusion or inheritance are held as default explanations for widespread similarity. But both conceptualizations are flawed—similar features are not diffused until proven inherited nor inherited until proven diffused. The probability of both explanations must be considered in light of what is known about processes of language change. The reason why a default assumption of either diffusion or inheritance may have developed in the Australian context may be partly related to the paucity of evidence from regular sound change mentioned in the previous section. For shared lexemes and forms of grammatical morphemes, competing explanations are traditionally evaluated with reference to regular patterns of sound correspondence, but the Australian situation renders this difficult. Dench (2001) and Miceli and Dench (2017) present the problems of indeterminacy encountered when unravelling the history of languages in the Pilbara region of north-western Australia, and argue that in such cases deciding between inheritance or diffusion is mostly ‘a matter of taste’ rather than one explanation being demonstrably more likely than the other. Much attention has therefore been paid to the degree of borrowing exhibited by Australian languages in which

39

luisa miceli and claire bowern loanwords are more easily recognizable, such as languages of the Cape York region that have undergone some systematic sound changes (see Alpher and Nash 1999) or because the languages involved are unrelated (e.g. Gurindji and neighbouring non-PN languages (McConvell 2009a); Mudburra and Jingulu (Meakins and Pensalfini 2020)). Bowern et al. (2011) argue that although there are some outliers, high levels of borrowing are atypical in Australian languages and that they fall well within what Greenhill et al. (2009) call ‘realistic’ borrowing rates (0–20%), which they find do not affect the recoverability of known phylogenetic structure. Studies employing Bayesian phylogenetic analysis such as Bowern and Atkinson (2012) therefore argue that undetected borrowings, due to a lack of systematic sound changes, are unlikely to affect our ability to conduct phylogenetic analysis; or at least, if they do, that Australian languages are not likely to be more affected than large language families elsewhere in the world. But borrowing may not be the only factor to take into consideration when reconstructing the history of languages likely to have evolved in a multilingual context. Ellison and Miceli (2017) observe a bias against shared word forms (doppels) in bilingual lexical production when language distinctive synonyms are also available for selection. Although various factors, at both individual and societal level, can modulate the strength of the bias, simulations predict that under certain circumstances the bias can lead to the rapid replacement of shared word forms in multilingual communities over time. This bias, which they refer to as doppel avoidance or anti-doppel bias, poses a different problem for probabilistic tree building than does borrowing. Like borrowing, it affects the independence of cognate counts across languages, but the impact of the two phenomena on individual lexical items is different. A systematic doppel avoidance across the tree is likely to bias phylogenetic analyses in two ways. Firstly, by accelerating rates of change, it will estimate language splits as being older than they in fact are. Since rates of language change vary extensively over the tree, the effect would need to be systematic and widespread to have major consequences. Secondly, since doppel avoidance is related to phonological characteristics of semantically overlapping lexical items, it introduces a type of structured rate heterogeneity which is currently not modelled. Phylogenetic analyses such as Bouckaert et al. (2018) recognize and model variation in rates of change based on semantics. That is, it has long been clear that meanings change at different rates, with some types of vocabulary being stable and slowly changing, while in other categories the lexical turnover is much more rapid. Systematically modelling doppel avoidance would need to take into account the rarity of phonological and/or phonotactic shapes of lexical items.

40

While borrowing as a form of interdependence has been taken into account in phylogenetic models (see e.g. Greenhill et al. 2009), doppel avoidance is yet to be considered, and this is crucial given the prevalence of multilingualism in Australia (see Vaughan and Singer 2018; Vaughan, Chapter 54, this volume). Furthermore, if the transmission of features in one language in the repertoire is partly dependent on the features present in other systems also being acquired, then it is unclear whether the long-term outcome of these transmission biases can be understood in terms of the traditional dichotomy of vertical vs. horizontal transmission—in such circumstances the reality of intergenerational transmission is far from the dominant monolingual ideal associated with linguistic theory (see Miceli 2019 for further discussion).

5.2.4 Stasis Another issue to be considered is what we might summarize as the ‘dynamics’ vs. ‘stasis’ question in our reconstructions of the Australian past. Dixon (2002) presents a picture consistent with stasis, arguing that the current comparative pattern is the result of a long-term period of equilibrium, which has favoured the diffusion of linguistic features between neighbouring groups that have mostly been in situ since initial colonization, bar occasional small scale or regional punctuations. From this perspective, similarity results from contact within stasis and difference results from non-overlapping patterns of convergence. More recently, Evans (2020) has proposed a different kind of stasis. He argues that because—in his view—there is growing consensus that all Australian languages are related and there is archaeological evidence of cultural continuity since first human settlement, this suggests that Proto-Australian may have been the language of the initial settlers. This is, of course, an interpretation that greatly challenges the working assumptions of historical linguists, considering the fact that Australia has been settled for around 60,000 years. That there would be enough evidence of genetic relationship after so many years would imply a much slower rate of change than has been observed in languages with documented histories—possibly ten times slower—and this is of course problematic for the uniformitarian assumptions that guide our work. Others working on Proto-Australian, e.g. Harvey and Mailhammer (2017), assume a much later date that is more in line with assumed dates for other language families, and this paints a more dynamic picture of the Australian past. These are exactly the kinds of issues that will benefit most from interdisciplinary approaches, since discussions of stasis vs. dynamicity have also long preoccupied achaeologists

australian languages and interdisciplinary approaches to the past and anthropologists, and feature in DNA studies focussing on Indigenous Australian populations. For example, early work by Gould (1977) portrayed Australian desert culture as uniform and unchanging, while later archaeological studies (starting with works such as Smith 1989 and Veth 1989) as well as ethnographic studies (e.g. Tonkinson 1991) present a more dynamic picture. Recent genetic studies such as Tobler et al. (2017) have emphasized regional population continuity on the Australia continent. Although it is in addressing these common issues that interdisciplinary work can make the most important contributions, it is also where we must exercise the most caution. For example, dynamicity in the archaeological record is frequently explained in terms of demographic changes, often linked to documented climatic changes (see e.g. Fitzsimmons et al. 2013; Reeves et al. 2013), and to cultural transmission rather than population replacement or movement, but linguistic dynamicity such as language spread is difficult to account for in the absence of L1 speakers. We must therefore consider carefully the problem of reconciling the pictures of the past stemming from different disciplines.

5.3 Linguistics and genetics and where they disagree: a case study of three trees By way of illustration of the points discussed above, consider three recently published trees schematizing Pama-Nyungan speakers’ history: Bouckaert, Bowern, and Atkinson (2018) is based on language data (to be precise, lexical cognate evolution). The authors use a Bayesian framework of population movement coupled with lexical replacement to infer the dating of Proto-Pama-Nyungan, the most likely area of spread, and intermediate subgroups. They find a date range of 4–7 kya (that is, mid-Holocene) for the ancestor of the Pama-Nyungan languages, and a source of spread from the region of the base of the Gulf of Carpentaria (near modernday Mount Isa). Malaspinas et al. (2016), in contrast, is a tree based on biological genetic data. They find evidence for the spread of people across Australia starting 32 kya, from Cape York (though note that Malaspinas did not have samples from the same regions as Bouckaert et al. and included many fewer groups). They also have an east to west spread, and some details of the breakup of the tree appear to match the linguistic data, albeit at a much earlier age. Finally, Tobler et al. (2017) is another gene tree, based on mitochondrial DNA (Malaspinas et al. used whole genome data). Tobler et al. found evidence for continued ancient lineages and in situ diversification dating back perhaps 50,000 years.

The problems can be summarized as follows. Problem one is that the Malaspinas et al. gene tree and the Bouckaert et al. language tree appear to be topologically similar, but have very different time scales. If taken at face value, they reflect processes of migration that happened tens of thousands of years apart. The second problem is that the two gene trees are different; they have different time scales and different structures to the trees, implying different population histories. Tobler et al., for example, found evidence for coastal spreads, whereas Malaspinas et al. did not find evidence for this. Thirdly, putting it all together, the genetic data suggests population stability, but climate and language suggest variation and dynamic change across the relevant periods. Thus, in short, we are left with three versions of the past: the linguistic tree suggests recent dynamic changes and population movements, while the gene trees suggest long-term stasis; within the gene trees, however, one suggests longterm regionalism, while the other shows a continent-wide cline. There are several ways that we might try to reconcile these pasts. One is to try to match up the trees by assuming that the trees are skewed—overly old in the genetic case, overly young in the linguistic case, or possibly both (and so we meet in the middle). In order to do this, we should think about the things that would need to be true for the linguistic dates to be too young. The following sections explore these points in more detail.

5.3.1 Domain specificity and assumptions about change Before we examine the issues in more detail, it is worth thinking about the ways in which language and genes are similar, and how they differ. The similarities stem from both being evolutionary systems with features of transmission, variation and selection. That is, genetic evolution and linguistic evolution are both ‘evolution’. There are also differences between the two domains, besides the obvious ones that stem from the fact that languages aren’t corporeal species. Perhaps the most important difference is that language phylogenetic models are of traits—words, sounds, patterns, and the like, whereas genetic phylogenetic models model DNA correspondences, which (eventually) give rise to traits. That is, although linguistic phylogenetics models a system which has properties of an evolutionary system, it does not have the same relationship to the core features of those systems: variation, selection and transmission. Simply put, traits are not transmitted directly, whereas portions of DNA are transmitted and inherited. DNA produces traits, in a

41

luisa miceli and claire bowern complicated way, but it is not the trait itself; numerous genes can produce similar traits, and individuals can have identical strings of DNA and yet differ in the physiological traits they exhibit. A second point relates to how variation is measured in a population. In genetics, Effective Population Size is a measure of the number of variant systems in a population. But in linguistics there is no equivalent to Effective Population Size. Moreover, it is difficult to see how there is any way of capturing the full range of variation in a production system for language, because variation is, in part, context-dependent. It relates to speakers, linguistic context and situation. The Effective Population Size for linguistic variation, as currently conceptualized, is infinite. Another source of differences is how these domains relate to people. That is, we use models of genetics and linguistics to make inferences about what people did in the past. And people’s relationships to language (as hosts of language) and as genetic carriers are rather different. An individual can be a ‘host’ to more than one language; half the world’s population is multilingual, and so for half the world, drawing a one-to-one link between a linguistic population and a genetic population is not straightforward. Secondly, language is consciously manipulated. While most of the changes we study are probably either below the level of consciousness or not directly and consciously manipulated, some clearly are. One example is the doppel avoidance discussed above (Section 5.2.3)2 ; Thomason (2007; 2020) has discussed other cases of deliberate change. Also apropos, languages can be reformed and mixed at a population level (cf. the formation of pidgins, creoles and mixed languages discussed by Thomason 2020, Thomason and Kaufman 1988, etc.). Such language contact requires time and interaction to form. Genetic admixture is also found, of course, but in contrast, sporadic interactions can leave lasting genetic signals in a way that is unlikely for language. Thirdly, membership of an ethnicity or ethnolinguistic group comes from practice, not (only) from birth. Daily language use, for an individual, is determined in part by the linguistic repertoires of the community as a whole and by the individuals who interact with one another. There is no such community-level negotiation of genetic material as it relates to individuals, except over many generations (through marriage rules). Finally, another difference comes back to the modelling of traits vs. genetic material. Physiological traits can be gained and lost in individuals but 2 Doppel avoidance itself is not conscious, but rather results from a bilingual’s conscious decision to make one specific language their language of production. This then leads to monitoring in order to avoid intrusions, and doppels are at times accidentally rejected in the lexical selection process due to their ambiguous status (see Ellison and Miceli 2017 for a full discussion).

42

not inherited. Linguistic traits, however, can be gained by individuals and then transmitted. Let us now consider the analogues in genetics. Individuals have a single set of unique DNA, which they keep throughout their lives. They acquire it at conception from their biological parents and it doesn’t change. At the level of individuals, humans can manipulate their languages, but they can’t manipulate their genes. At the population level, however, genetics can be manipulated (epiphenomenally) through marriage proscriptions. This takes time, however, and its effects on the group cannot be monitored (unlike with language, where the effects of manipulation, for at least some features, are noticeable). Individuals can leave big genetic traces. For example, Genghis Khan’s lineage (his and his brothers) has been argued to have left such a trace in Central Asia that about eight percent of people alive today are his descendants (Derenko et al. 2007). No one individual can ever have that much effect on a language, which requires group buy-in to adopt and continue to use the system. Thus, in summary, there is no reason to expect that such trees would necessarily be homologous, even when they reflect the same population histories, except inasmuch as they both reflect things that communities do. But their instantiation in a community (or population) of people is very different, and so it would not be surprising if those differences end up reflected in different evolutionary trees.

5.3.2 Are the language dates too young? First, we could have underestimated the amount of change, if there are processes that make languages appear more similar to each other than they would otherwise have been. Language contact makes languages more similar to one another, and so extensive shared material which has been recently acquired would make dates younger than their true age. Australian languages are often assumed to have evolved in a multilingual context of extensive language contact. For example, Dixon (2002) proposed a whole theory based entirely around language contact to explain how Australian languages have similarities to one another. Other case studies, such as Dench (2001), Heath (1978a) and Hercus (1987), discuss linguistic diffusion in particular areas. Vaughan and Singer (2018), Vaughan (Chapter 54, this volume) and Singer (2018b) have discussed multilingualism, and language contact between individuals in multilingual communities. In contrast, however, Bowern et al. (2011) found that rates of identifiable borrowing in Australian languages

australian languages and interdisciplinary approaches to the past 70

%age of Languages

60 SAM

50

AUS

40

NAM

30

ALL WOLD

20 10 0

1

5

10

15

20

25 30 Loan %age

35

40

45

50

55

Figure 5.1 Loan figures in basic vocabulary for different languages across four regions (Australia, South America, North America and the worldwide sample in Haspelmath and Tadmor 2009a). Most languages have some borrowing in basic vocabulary, but few languages (in any region) have extensive borrowing.

were not that much higher than elsewhere in the world.3 That is, there were low-borrowing languages and highborrowing languages, just as there are in other regions of the world. There was no significant difference between the sample of Australian languages (which included both the Victoria River District and Arnhem Land—both highborrowing areas) and those in the world-wide language sample. Figure 5.1 above shows this result. Therefore, we conclude that Australian languages show varying effects of language contact, just as other regions in the world do. Language contact and lexical borrowing are features of the Australian langscape, just as they are for other parts of the world. Australian languages are not a special case in this regard. We can also use the results from linguistic phylogenetics to examine the extent to which borrowing might have occurred. The phylogenetic tree based on lexicon implies that there is relatively little borrowing—results don’t get systematically skewed in such cases, unless all the borrowing is in a single direction (and we know from Bowern et al. (2011) and Bowern (2014) that the directions of borrowing typically vary). We expect borrowing to lower the posterior probability of nodes, which does happen in parts of the tree. In summary, Australian languages clearly show the effects 3 Note that because of the relative uniformity of phoneme inventories, borrowing is often less identifiable on phonological grounds in Australian languages than in other parts of the world. However, other methods are available, based on morphology and word formulation, general patterns of lexical similarity and the like. Therefore, while it is possible that in individual cases, loans may go unidentified because of phonological similarities, the fact that we find similar rates of variation compared with languages elsewhere suggests that loan identification is not overly skewed by phonological similarity.

of language contact, but not to the degree that wipes out evidence of the intergenerational transmission that results in family tree structures. There are other ways in which a tree could reflect falsely young dates (that is, where the real age of the tree is older than the methods suggest). One is the use of incorrect calibration points. Bouckaert et al. (2018) used a calibration point on the Wati (Western Desert) clade to fix dates. Calibration points scale the tree linearly, so if the calibration point is 3 kya but should really be 6 kya, the root of the tree would end up half the age it should be. In the case of the Bouckaert et al. tree, the calibration point had a small effect on the overall tree inference. It was a loose calibration point based on the entry of humans into the Western Desert region. Note that the prior would have to be wrong by a great deal to bring the tree into line with the genetic tree: not twice as old, but rather eight times as old. Moreover, because the dates would scale the rest of the tree, the calibration point would have an effect on dialectal groups as well. Groups that show in our model as being a few hundred years apart (and are accepted by linguists as being shallow groupings) would end up thousands of years old. The same types of points apply to the clock models that are used to infer date ranges. Contemporary phylogenetics does not use a single rate of change, but rather a set of rates which vary. The models have a prior on rates of change, where the rates are drawn from a distribution. It is certainly possible that assumptions about possible rates of change are not correct. However, just as with the calibration points above, the assumptions would have to be off by a long way for that to make an appreciable difference in the estimated age of Pama-Nyungan. Again, a small change would not

43

luisa miceli and claire bowern produce a tree that is eight times as old as the one we see. Moreover, there is some evidence internal to the model that our rates are plausibly inferred. For example, the words which change slowly in Pama-Nyungan are the same items that change slowly in Austronesian and Indo-European. Another possible reason that dates might be off is in the identification of cognates. Phylogenetic models are models of lexical replacement, but lexical replacement is an amalgamation of several independent processes—borrowing, semantic change and sound change beyond recognition. That is, if a cognate word has changed so much that it is no longer recognizable as a cognate (and we can’t undo the sound changes to discover it), it will be marked as not cognate here. Conversely, if words are strikingly similar across wide spaces, we might assume that they are cognate but they might, in fact, be the results of lexical replacement. We leave this as a point for further investigation. Perhaps languages are older and look more similar because they have undergone high-fidelity transmission. That is an argument from Nunn (2018), that information of some types (including oral history and mythology) is likely to be preserved over many millennia because its successful transmission is important to the group, or that it is salient to community members and so attention is paid to its transfer. This is a good argument for information about poisonous plants and animals. However, it is not at all clear that the argument transfers to whole languages, for several reasons. One is that there’s no reason to think that full languages would be so salient that every detail would be transferred— language just doesn’t work like that. Language users are really good at constructing and co-constructing meaning. The types of information that relate to long range high fidelity transfer are literally about life and death. There’s also little repeated reinforcement for language, whereas there is extensive group-level reinforcement for some types of high fidelity cultural information. If a person dies from eating a poisonous plant, that’s a reminder to everyone in the group who learns of the experience that the knowledge is correct (and worth personally transmitting). But if someone’s F2 in the vowel /a/ is a little lower than average, people won’t even consciously notice. Thirdly, food knowledge can be invented and reinvented and therefore look stable, even though the trait has been reinvented. We have no way of recovering that sort of cultural loss and replacement over millennia (it would probably look indistinguishable from continuous transmission). But linguistic features aren‘t subject to continuous reinvention at the level of words (we don’t lose our word oola for water and then get the same one back 5,000 years later). All of these points are potential issues, but not at the scale that would be necessary for the dates to be really

44

young. Finally, we should note that the doppel avoidance mechanism discussed above (Section 5.2.3) may skew dates to be older, rather than younger. That is, if doppel avoidance plays a role in Australian language diversification, our linguistic dates are likely to be skewed to older than they should be, rather than younger. On the other hand, for languages where doppel avoidance has not been influential, and there are accidental similarities in forms (because the phoneme inventories are very similar), languages may be less closely related than they initially appear.

5.3.3 Are the genetic dates too old? If the linguistics dates are not too young (or at least defensible, or not easily scalable to 4 or 5 times the age without violating core knowledge about how language works), perhaps the trees are irreconcilable because the dates from genetics are too old. It is worth considering the types of assumptions in genetic modelling that might give falsely old dates. One is incorrect assumptions about generations; an average generation of 20 years rather than 25 years reduces the age of the family by 20%. That’s not enough to bring it into line with the linguistic dates, however. Moreover, we can’t bring a generation down to below about 15 (and even that is implausible). So, assumptions about the ages of generational replication alone won’t make the difference. Another possibility is considering the mixed ancestry population that spread. That is, consider the genetic consequences of a ‘pulsating heart’ model (Sutton 1990). In that model, there are repeated expansions and contractions of (related but intermixed) populations from centres of resource abundance (i.e. refuges, cf. Veth 1989) into more marginal areas in times when climatic conditions are favourable. Groups retreat to those areas in times of hardship. Such models are unlikely to fit with either genetic stasis or continent-level clines, but the implications are still to be investigated. Thirdly, genes do not track language history directly. They track migrations of individuals and groups. To see this, consider the following thought experiment. Imagine that 100 non-Indigenous Australians are randomly sampled for DNA and language use from five different locations. The language ‘tree’ produced from their varieties would have little resolution, since there are few systematic regional differences. The genetic tree, however, would likely show a much more complex set of lineages, with origins predominantly in Europe, Africa, and/or Asia. The linguistic lineage is homogeneous, but the genetic lineage is more complex, reflecting immigration from multiple origins. The mismatch is clear

australian languages and interdisciplinary approaches to the past with contemporary populations, but it is not difficult to construct scenarios where language and gene histories do not well align. Finally, consider the problem that analysis of the genetic samples did not find bottlenecks in the recent past. But we know there was a population bottleneck: the genocide of the Aboriginal population following European settlement. The Aboriginal population of Australia was reduced by 90% by 1920. This is an important and salutary check on the results: if the resolution of sampling does not find evidence of recorded genocide, perhaps other population changes in the more remote past are similarly masked from view.

5.4 Conclusions This chapter has highlighted the different challenges encountered in building an integrated history of the Australian past. We first discussed the linguistics-internal issues that have arisen in the course of building a picture of the linguistic past. An in-depth understanding of these issues is a necessary first step in moving forward, not just within

the Australian context but in rethinking theory in historical linguistics more generally. The comparative sound patterns in particular, discussed in Section 5.2, continue to place Australian languages in a central position in discussions dealing with working assumptions and in the development of novel approaches. We then presented a case study which brings to light some of the issues encountered in integrating interdisciplinary findings—focussing on linguistics and genetics. Languages and genes are both participants in evolutionary systems, and evolve according to similar principles (variation, selection, and transmission). They both trace population histories and can tell us something about the past. However, these population histories track different processes, and so we should not necessarily expect to see matches between the trees produced from language data and those produced from genetic data, even if they are languages and genes from the same continuous population. There are a number of reasons why we might see differences, and, as with the linguistics-internal issues, understanding these differences is crucial to an integrative science of the past, for Australia and for the rest of the world.

45

chapter 6

Nineteenth-century classifications of Australian languages Clara Stockigt

6.1 Introduction The earliest attempts to classify Australian languages and people were made in the nineteenth century. The non-classificatory and synchronic grammatical research produced in Australia mostly by missionaries (Stockigt 2015: 336–8) informed a body of historical and classificatory literature which was overwhelmingly produced outside the country. The history of Australian linguistic description is a fairly recent phenomenon. The earliest world-wide uses of the term ‘ergative’, for instance, to name both a peripheral case (Taplin 1872), and the ergative case (Schmidt 1902) that occurred in descriptions of Australian languages (Lindner 2014; Stockigt 2015) have until recently been overlooked in histories of the term ‘ergative’ (Seely 1977; Manaster-Ramer 1994). The role that Australian data played in debate about whether ergative constructions are passive or active (H. C. von der Gabelentz 1861; Mu¨ller 1882) has similarly not been reviewed (Seely 1977; Stockigt 2017). The delayed onset of historiographic investigation into Australia is partially due to Australia’s relatively late colonization. Much missionary linguistic effort in Australia postdates the era before 1850, upon which global missionary linguistic historical scholarship has focussed (Zwartjes et al. 2014: vii). The tentative beginnings of the study of Australian languages within secular academic institutions in Australia—at the University of Sydney in 1926, and soon after in Adelaide in the Classics department of the University of Adelaide and at the South Australian Museum—occurred at least half a century after the founding of parallel institutions in North America (Campbell 1997: 35–7, 57). Once Australian linguistic studies were underway, the assessment of material contained in the early records of Australian languages was understandably not prioritized as linguists concentrated on documenting endangered languages from the last generations of speakers. Thus, nineteenth-century classifications of Australian languages have received little attention. Koch’s (2004b)

historical overview of the classification of Australian languages, for instance, focusses on the twentieth century, since (p. 20) ‘it was not until around 1900 that professional linguists became involved in the historical study of the languages’. But many of the scholars whose views are examined in this chapter were professional ‘linguists’, by the standards of the time. Rev W. Ridley, for example, who published on Gamilaraay spoken in New South Wales between the 1850s and the 1870s, was trained at King’s College London, and held an MA from the University of Sydney. Ridley’s first publication on Gamilaraay (1855) appeared in the London-based Transactions of the Philological Society. Further, the nineteenth-century European philologists who read and reinterpreted the primary linguistic material produced in Australia, are among a group of scholars who worked to accommodate the foreign structures of ‘exotic’ languages within European philological thought, and developed linguistic typologies based on available material and current understanding. Indeed, the study of the nineteenth-century typologies and classifications of Australian languages reveals that a north/south division of mainland Australian languages into two groups was proposed as early as 1858, by the German philologist Wilhelm Bleek. The early documents addressing the relatedness of Australian languages illuminate pejorative views about Australian Aboriginal languages and people, but also reveal attempts to counter the popular assumption that ‘primitive’ people spoke simple languages, for example Schu¨rmann (1844: v), Ridley (1866: vi), and Taplin (1879b: 123). Much of the early classificatory literature about Australian languages is saturated with discussion about the putative relationships between Australian languages and those spoken elsewhere in the world, described by Capell (1970: 633) as ‘philological imagining’. This chapter focusses rather on the early typologies and internal classification of Australian languages.

Clara Stockigt, Nineteenth-century classifications of Australian languages. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Clara Stockigt (2023). DOI: 10.1093/oso/9780198824978.003.0006

46

nineteenth-century classifications of australian languages

6.2 The relatedness of Australian languages The discovery that Aboriginal people spoke distinct languages came as an initial surprise to the European members of an exploratory party, led by the first Governor of New South Wales Captain Arthur Phillip, who in 1791 ventured into regions adjoining those first settled in New South Wales. Finding that Aboriginal people further inland did not speak Iyura (Eora), the Dharuk variety from Port Jackson with which the colonists had gained some familiarity in the thirteen intervening years since British colonization (see Troy 1993: 43–4; Wilkins and Nash 2008), the colonists realized that they were to encounter greater linguistic diversity than initially anticipated. Their finding was confirmed as languages from further afield were documented. In 1802 the British navigator Matthew Flinders recorded a Nyungar vocabulary from Western Australia (2015: 258) noting its dissimilarity to Port Jackson (Eora) and Port Phillip (Boonwurrung). Australian linguistic diversity was soon perceived as an obstacle to missionary work which was not encountered in mission fields elsewhere (Zweck 2012: 43–5). By the mid-nineteenth century, however, some researchers were concluding that languages documented in different locations across the vast continent belonged to a single family. The relatively small body of data on which the assumption of shared heritage was founded did not sample the non-Pama-Nyungan languages from northern areas of mainland Australia or from Tasmania (Bowern, Chapter 7, this volume). In the very first grammar of an Australian language, Protestant missionary Threlkeld (1834) (cf. Stockigt, Chapter 2, this volume) suggested an affinity between Awabakal—the language he described spoken close to Newcastle in Australia’s earliest settled colony, New South Wales—and Nyungar, recorded in Scott Nind’s 1831 vocabulary and his ‘account of the natives’ collected at King George’s Sound, some 4,000km away in Australia’s southwest. Threlkeld (p. x) wrote that although ‘not one word … appears to be used or understood by the natives in this district … the language is formed on the same principles, and [is] perhaps radically the same tongue’. In 1841 Sir George Grey (1841, vol. II: 207–16) set out to show that Australian languages were ‘radically the same across the continent’, positioning his discussion to counter a perceived belief that ‘languages spoken in different portions of the continent of Australia are radically distinct’. Just as Threlkeld’s (1834: x) proposal that Australian languages were related was based on lexical evidence,1 Grey’s (vol. II: 1 Threlkeld (1834: x) noted the similarity between the Nyungar form, recorded by Nind as: Coo-whie, Coo-whie-ca-ca and the Awabakal exclamation, which Threlkeld recorded as ka-ai, Ka-ao, kai, ka. Threlkeld observed

212) evidence was primarily lexical, although he also offered (p. 208) the ‘general similarity of sound and structure of words in different portions of Australia’, without elaboration. Between 1837 and 1839 Grey explored in the north-west of Australia and in the south-west of Australia, and published ‘Vocabulary of the dialects spoken by the Aboriginal Races of South-Western Australia’ (1839). In 1841 (p. 211) he drew up a comparative vocabulary of twenty-eight items from Swan River (Nyungar), King George’s Sound (Nyungar), South Australia (Kaurna) and Sydney (Awabakal). The list, which did not include pronouns, provided body part terms common to many Pama-Nyungan languages: renderings of ‘hand’ mara, ‘tongue’ thalany, ‘foot’ thina and ‘eye’ mi:l. Conducting the earliest lexico-statistical analysis of Australian linguistic varieties, Grey (p. 211) calculated that ‘one-eighth of the words known as belonging to the Perth dialect, have been found also in that of Adelaide’. Grey played a decisive role in the developing understanding of Australian languages both directly and indirectly. As Governor of South Australia (1841–1845), he supported the publication of the early grammars of South Australian languages written by Lutheran missionaries (Meyer 1843; Schu¨rmann 1844) and by the first permanent Protector of Aborigines in South Australia, Mathew Moorhouse (1846). Significantly, the grammars of Schu¨rmann (1844 v: iii), Meyer (1843: iv) and Moorhouse (1846: v) are dedicated to Grey. Grey’s linguistic interest continued while Governor of New Zealand (1845–1853 and 1861–1868), and of Cape Colony, South Africa (1854–1861). His amassed philological library held in South Africa was to become important in the dissemination of Australian linguistic materials in Europe in the nineteenth century.

6.3 Early grammatical typologies supporting the notion of relatedness Grey’s (1841, vol. II: 207–16) very early claim that Australian languages were related was based primarily on lexical evidence. As more languages were described in greater detail, discussions about the relatedness of Australian languages tended to place emphasis on shared grammatical structure. In 1856 missionary Ridley (p. 293) observed that ‘[t]hough not one word in a thousand in Kamilaroi resembles that dialect [Lake Macquarie language, Awabakal] I already perceive important points of resemblance in grammar’. In 1844, Lutheran missionary C. W. Schu¨rmann (p. iv) wrote ‘[i]n forming however an opinion on the affinity of the languages or dialects, one has to look not only at the number that: ‘allowing for the difference in orthography [the two forms] would convey nearly, if not precisely the same sound, the meaning is halloo, halloo, approach, approach’.

47

clara stockigt of similar words, but still more to the grammatical structure and idiom’. In a well-informed overview of previous research into Australian languages, the British ethnologist and physician James Prichard (1847: 272) explained that: ‘… correspondences in vocabularies would not have afforded by themselves sufficient proof of a family relation between the Australian languages. But the evidence afforded by it has been confirmed by grammatical researches’. Prichard’s Australian material was presented in the fifth volume of Physical History of Mankind, which was added to the polymath’s life’s work the year before his death. The study of Australian linguistic typology was greatly advanced during the 1840s, although the synthesis of information was presented in a still relatively small number of publications about a limited range of languages. Two typological overviews were published in South Australia by men whose linguistic work was secondary to their professional dealings with Aboriginal people (Schu¨rmann 1846; Moorhouse 1846). Another two works were published outside Australia (Hale 1846; Prichard 1847) by philologists who sought out Australian linguistic material in order to infer the origins of the world’s languages and people. In 1846, Horatio Hale, an American ethnologist and later mentor of Franz Boas, published a comparative grammar of two Australian languages in the Reports of the United States Exploring Expedition. Hale visited New South Wales between November 1839 and March 1840, and was inducted into Australian linguistic structure by the missionary Threlkeld at Lake Macquarie mission, and the missionary W. Watson at Wellington Valley mission. Comparing the grammatical structure of ‘two dialects as widely separated as possible’ (p. 481), Hale advocated that the examination of grammatical structure was equally as important as lexical comparison in determining a relationship between languages. He concluded that Australian languages ‘though differing in many respects … preserve sufficient evidence of a common origin’. The two languages that Hale ‘selected’ for comparative grammatical purpose were Awabakal and Wiradjuri, which were, in reality, the only two languages to which Hale had reasonable access via the missionaries. He presented the ‘Wiradurei’ and ‘Kāmilarai’ (Awabakal) material in adjacent columns giving equivalent structures for each language side by side, a format which he described (p. 484) as favourable because ‘the points of resemblance and dissimilarity may be seized at once’.2 The same format was later used by Lutheran missionaries Flierl (c. 1880) and C. Strehlow (1910) in comparative grammars of Diyari and Wangkangurru, and of Arrernte and Luritja respectively. 2 Hale (1846) referred to the language that Threlkeld had not named, but had identified by location, as ‘Kāmilarai’. His reasons for doing so are not clear.

48

In 1846, Lutheran missionary Schu¨rmann (1846: 249–51), who had co-authored a grammar of Kaurna with C. G. Teichelmann (1841) and had independently written a grammar of another South Australian language, Barngarla (1844), presented a list of ‘peculiarities’ common to the structure of the languages he knew. A significant component of Schu¨rmann’s typology was grammatical. It was framed largely by features perceived to be ‘lacking’ in Australian languages: articles, auxiliary verbs, relative pronouns, prepositions, gender and distinct forms of the passive voice. Schu¨rmann additionally described features that the languages shared but which are not commonly found in standard average European (henceforth SAE) languages (Whorf 1941 [1939]; Haspelmath 2001): pronominal sensitivity to kinship, complex verb morphology marking categories other than SAE categories of tense, aspect and mood, and suffixation of nouns with case inflections marking functions that are not morphologically marked in SAE languages. Also in 1846, Mathew Moorhouse, the first permanent Protector of Aborigines in South Australia, published a grammar of Ngayawang (1846), spoken on the Murray River in South Australia. In the introduction, he (pp. v–vii) presented an overview of Australian linguistic structures (Figure 6.1) in order to show that the languages (p.vi) ‘had their origin from one common source’.

Figure 6.1 Moorhouse’s (1846: vi) typological summary of Australian languages.

Moorhouse’s discussion appears to refute Grey’s position (1841 vol. I: 365–6) that Australian Aboriginal languages were mutually intelligible. In Moorhouse’s view, ‘… if natives of any one language happen to meet those of another, they are obliged to converse in English, to make themselves understood’. Moorhouse also provided a ‘onetwentieth’ count of words ‘agreeing in root’ in unspecified languages, a figure much lower that Grey’s ‘one-eighth’ lexico-statistical count. Moorhouse considered whether different Australian linguistic varieties should be considered languages or dialects of a single language. While Grey (1841;

nineteenth-century classifications of australian languages 1845) had discussed ‘Aboriginal dialects’, Moorhouse (1846: v–vi) pointed out that [t]he term ‘dialect’ is scarcely applicable to the languages of New Holland. They differ in root more than English, French and German … yet there is evidence sufficient to satisfy any one they belong to one family.

It was pronominal evidence from around the country that largely alerted nineteenth-century scholars to the relatedness of Australian languages. Lutheran missionary J. C. S. Handt, for example, in a letter written in 1837 from Moreton Bay—close to present-day Brisbane—to missionary W. Gu¨nther, noted that the 2sg pronoun in the Moreton Bay language, Turrubul was the same as the form in Wiradjuri, spoken at the Wellington Valley mission in New South Wales (Newton 1987: 175). Grey (1841, vol. II: 214) published a comparative table of pronouns from Western Australia, New South Wales and South Australia showing the ‘remarkable degree of resemblance’. Moorhouse (1846: vi) offered the ‘striking similarity in the pronouns’ as evidence of shared heritage and provided a comparative paradigm in six languages. He also observed the relative dissimilarity of third person forms in comparison with other pronouns, which, along with the observed shared form of the dual suffix on nouns, led him (1846: vii) to make the fanciful claim

that Aboriginal people had ‘separated in pairs, and these words [1st and 2nd person pronouns], being in daily use were retained … as children were born … the terms for the third person had to be invented’. Prichard’s (1847: 273) comparative pronominal paradigm reproduced Grey’s pronouns in the three languages (1841, vol. II: 214) with the addition of pronouns from the ‘Parnkalla dialect’ and from Encounter Bay, based on Schu¨rmann (1844) and Meyer (1843) respectively. In 1866, Rev W. Ridley (p. 43) observed that ‘[t]he pronouns of the first and second person are nearly the same all over Australia’. Difference between pronominal forms was also used as an early diagnostic tool to determine language boundaries. In 1840, Moorhouse and Lutheran missionary C. G. Teichelmann collected pronominal paradigms from groups of Aboriginal people at locations south-east of Adelaide in order to determine the eastern boundary of the Adelaide language, Kaurna. Moorhouse and Teichelmann’s table of pronominal forms in four languages from south of Adelaide (Moorhouse 1840; 1846: vi), which was sent to Grey in 1845, informed a map published by Grey (1845: 365) in the Royal Geographical Society Journal (Figure 6.2) showing the location of the ‘five principal dialects’ spoken across the southern portion of the continent. Decades later, and also in South Australia, the Congregationalist missionary George Taplin, who had previously

Figure 6.2 Grey’s (1845: 365) map of Aboriginal dialects.

49

clara stockigt compiled a comparative vocabulary of twenty Australian languages (Taplin 1872) and had published a grammar of Ngarrindjeri (Taplin 1872 [1870]), circulated a survey seeking anthropological and linguistic information about Australian languages, the results of which he published in 1879 (Taplin 1879a) (cf. Stockigt, Chapter 2, this volume). The questions posed by Taplin show that in addition to the features perceived to be common to Australian languages by Hale, Schu¨rmann and Moorhouse, Taplin was aware that Australian languages were likely to exhibit systems of bound pronouns, and ergative morphology. The material also indicates that Taplin was fairly well informed about a range of case functions that were likely to be morphologically marked on nominals. As early as 1847, Prichard (p. 278) noted that many Australian languages have ‘two pronouns of the first person plural, one we including the person or persons addressed, and the other excluding them’. It is curious that Prichard was able to make this observation, since the only published grammars of Australian languages which had described an inclusive/exclusive pronominal distinction in 1847 were Threlkeld (1834: 23) and Hale (1846: 489) both of which presented the distinction as occurring on the dual first person pronouns, but not on the plural. Based in the assumption that linguistic diversity was more a matter of lexicon rather than of phonological or grammatical structure, nineteenth-century grammarians sometimes underestimated how grammatically different Australian languages might be, and incorrectly imported grammatical structures across language boundaries. There is a tendency among early grammarians to overestimate the structural homogeneity of Australian linguistic structures. Such instances include Moorhouse’s description of undifferentiated marking on non-singular nominals in Ngayawang (1846), which probably resulted from the extrapolation of data from Teichelmann and Schu¨rmann’s Kaurna grammar (1840), J. Flierl’s (1880) extrapolation of tripartite marking of 1pl pronouns from Diyari into Wangkangurru, and Meyer’s (1843) attempt to show that demonstrative pronouns could act to relativize clauses in Ramindjeri, as described by Teichelmann and Schu¨rmann (1840) in Kaurna (Stockigt 2017). The tendency might be seen to have endured. The similarity of Australian languages has similarly led ‘modern’ linguistic scholars to overemphasize the grammatical (Evans 2000a: 91), and phonological (Gasser and Bowern 2014) uniformity of Australian languages.

6.3.1 Dissemination of information about Australian languages Despite the typologies of Australian languages developed in 1846 by Schu¨rmann and by Moorhouse, and the intelligent

50

collation of grammatical data later published by Taplin (1879a), published in South Australia, nineteenth and early twentieth-century grammarians tended not to position their analyses within what was known about other Australian languages, or to advance the typological description of Australian languages. Due perhaps largely to their geographical and intellectual isolation, nineteenth-century Australian grammarians, such as Walter Roth (1897) (Stockigt 2020), often appear uninformed about any accumulated understanding of Australian linguistic structure. After Grey (1845), and aside from an attempt made by Fraser (1892a: xi–lxiv, 1–8) in the introduction to his compilation of grammatical material of Australian languages (1892a), the nineteenth-century classificatory and typological studies of Australian languages were produced outside the country by men who never visited Australia, for example Prichard (1847), H. C von der Gabelentz (1861) and Mu¨ller (1868; 1882). The pathways by which linguistic theory and practice moved between Australia and the rest of the world are largely untraced. The failure of scholars in Australia to assimilate the enduringly instructive two dimensional mapping of the place of articulation against manner of articulation of Australian consonants and vowels, that had been presented in European publications between 1854 and 1938 (Lepsius 1855: 64; 1863: 226; Mu¨ller 1867; 1882; Planert 1907; 1908; Gatti 1930: 1; Sommerfelt 1938: 42, 45) (cf. Stockigt, Chapter 2, this volume) indicates that channels of communication from Europe to Australia were under-developed, and that the movement of ideas was largely unidirectional. While grammarians in Australia were primarily evangelically motivated to record Australian languages, many were at the same time aware that their philological work was invaluable to ‘armchair’ scholars away from the field. As early as 1843, in a grammar of Ramindjeri spoken south of Adelaide in South Australia, Lutheran missionary Meyer (1843: vii) presented his material to an international audience. I submit these sheets … with the hope that … they will be interesting to the philosopher and philologist, as exhibiting the peculiar structure of a language spoken by a people very generally considered the lowest in the scale of civilisation.

Research alliances developed between those working in close contact with Aboriginal people and professional scholars in urban centres. The most renowned collaborations being the anthropological collaboration between Baldwin Spencer and Francis Gillen (1899; 1904; 1927) (P. Jones 2005) and the ethno-linguistic collaboration between Moritz von Leonhardi and Carl Strehlow (1907–1920) (Kenny 2013). Aboriginal people from the MacDonnell Ranges in Central Australia, who came to be named the ‘Arrernte’ (Arunta, Aranda) were the subject of both pairs of scholars.

nineteenth-century classifications of australian languages But before Gillen had met Spencer, or Strehlow had corresponded with Leonhardi, alliances had been forged between missionaries at the descriptive coalface and European scholars. These alliances provided important routes by which philological and anthropological understandings, and descriptive practices, were transferred between ‘the bush’ and centres of learning. In 1858, the Prussian philologist Wilhelm Bleek published a catalogue of the library of Sir George Grey. Himself an authority on Khoisan languages, Bleek held a doctorate in linguistics from the University of Bonn and had spent time at the University of Berlin where he studied under K. R. Lepsius. Reflecting Grey’s ongoing interest in Indigenous languages throughout his career, the library contained a comprehensive collection of linguistic material from around the world. Bleek presented and discussed numerous Australian linguistic sources in vol. II, Part I, with a short addenda in Part III. Grey’s ongoing patronage allowed Bleek to pursue philological studies broadly (Gilmour 2006: 170), and Bleek later authored ‘On the position of Australian languages’ (1872). Sometime before 1874, Bleek requested information from the South Australian Governor, Sir A. Musgrave, about the manners, customs and folklore of the natives of the colony. Musgrave (quoted in Bleek 1874: 6) recommended Taplin as ‘one of the best-informed men in the Colony on all subjects respecting the natives’. Taplin’s contact with Bleek occurred around the same time that the Oxford linguist Max Mu¨ller approached the missionary Ridley in New South Wales for information about Australian languages (Gardner and McConvell 2015: 109). Aside from the Lutheran missionaries at Bethesda in South Australia (Koch 1868; Flierl 1880), whose work remained unpublished, Ridley and Taplin were among the very few scholars researching Australian languages at the time. Like Taplin (1872; 1874; 1878; 1879a, b), Ridley (1855; 1856b; 1866; 1875) published both internationally and in Australia about the languages, manners and customs of Aboriginal people. Taplin and Ridley came to be considered experts on the Australian ‘natives’ by readers outside the country. Ridley’s Gamilaraay kinship terminology is, for instance, referred to by the Viennese philologist Friedrich Mu¨ller (1868: 8).3 These attempts by European scholars to gather Australian linguistic data from missionaries parallels the collection of anthropological data about Australian people in order to make typological classifications of social structure. In 1872, L. Fison circulated L. H. Morgan’s expanded philological lists of kinship terms in questionnaires circulated in the Australian press (Gardner and McConvell 3 Friedrich Mu¨ller (1834–1898) should not be confused with the German born Oxford Professor of comparative philology, Friedrich Max Mu¨ller (1823–1900), generally referred to as Max Mu¨ller, who in 1854 (p.158) classified languages from the ‘Great Southern Continent’ within the southern branch of the putative Turanian family.

2015: 105–8), resulting in Fison and Howitt’s Kāmilarai and Kurnai (1880), to which both Taplin and Ridley provided information. In the same way that Fison’s anthropological survey about Aboriginal kinship systems had been instigated from outside the country, by the American anthropologist L. H. Morgan, Taplin’s impetus to draw up and circulate his survey in 1874 (1879a) is likely to have been encouraged by Wilhelm Bleek.

6.3.2 Development of terminology referring to linguistic entities While the use of some language names has been relatively stable since the earliest nineteenth-century record, for example, Barngarla (Parnkalla), Diyari (Dieri) and Wiradjuri (Wirradhurri), we cannot assume that currently used terminology reflects classical Aboriginal practices of naming languages and societies (see Sutton 1979; Rosenberg, Chapter 49, this volume), nor that currently named linguistic or anthropological units performed any classical function that would have attracted a name. Driven by the scholarly need to clearly communicate developing understandings about the languages and social organization of different groups of Australian people, the terminology used by European scholars to refer to Australian languages and groups of people became standardized. Where a term used as an ethnonym differed from that used as a language name, the distinction tended to collapse in written records, and a single term was used to refer to both. Sutton’s (2010a) study of alterations in the use and meaning of terminology used by Aboriginal people to designate social and linguistic groups in the Western Desert, for instance, reveals a collapse in the range of terminology, and considerable shifts to the meanings of terms that have remained current. Examination of instances in which these processes are historically retrievable reveals the extent to which the current meaning of widely used terms can differ from that recorded close to the time of sovereignty. Take for instance the term ‘Arrernte’ (Aranda and other variant spellings) used to refer to related languages spoken in a large area of Central Australia, including Alice Springs (Wilkins 1989: 1–19). The term Arrernte was first recorded by Christopher Giles in 1875, who was stationmaster at Charlotte Waters telegraph station, close to the southern extremity of the country to which Arandic languages belong. In response to Taplin’s questionnaire (Taplin 1879a: 89–91) Giles gave ‘Arrinda’ as the name of the language spoken by the ‘Antakerrinya’ tribe, and he provided an Arandic wordlist (see Koch and Turpin 1997). The term ‘Arrernte’ next appears as ‘Ar-On-Tha’, on a map produced by Charles

51

clara stockigt Richards in 1892 denoting the same relatively small area close to Charlotte Waters (see Knapman 2011). It was Frances Gillen who first used the term ‘Arunta’, to name the people from the MacDonnell Ranges to the north—at Alice Springs and at the Lutheran Hermannsburg mission—who spoke closely related languages. He first used the term ‘Arunta’ in an 1894 notebook ‘Notes on some manners & customs of the Aborigines / McDonell Ranges / Tribe of the Arunta …’. Spencer and Gillen’s (1899; 1904; 1927) and C. Strehlow’s (1907–1920) publications brought into contest the supposed ‘primitiveness’ of Gillen’s ‘Arunta’ in international literature. In 1903 Gillen (1997: 434) wrote to Spencer: I am inclined to think that in all the CA [central Australian] tribes the tribal name comes from without[.] No one in Cent’l Aust knew the tribal name Arunta until I dug it up.

It is possible that Gillen ‘dug-up’ the term from Taplin (1879a), but it is also possible that he sourced the term independently when he was post-master at Charlotte Waters between 1875 and 1892. Either way, before the European appropriation of the term ‘Arrernte’, the term probably referred to just one of many sociolects that was comprised within what was to later become known as the ‘Arandic’ linguistic area. A shift in the designation of the term ‘Arrernte’ to refer to both the people and language belonging to country at the Hermannsburg mission, is evident in the correspondence between Lutheran missionary Carl Strehlow and his German editor Baron Moritz von Leonhardi. In his initial introductory letter (1901), Leonhardi seeks Strehlow’s opinion of the validity of Spencer and Gillen’s use of the term ‘Arunta’ asking: ‘[T]hey are called Arunta by those doing the reporting. Is that name correct?’. Strehlow’s reply is not extant, but the term, which had not previously been used by Lutheran missionaries, first appeared at Hermannsburg in the title of C. Strehlow’s Aranda service book published in 1904 Galtindinjamea-Pepa Aranda-Wolambarinjaka (Knowledge giving paper for the Arrernte speaking congregation). The earliest missionaries at Hermannsburg, for example Kempe (1891), did not name the people or language at the mission ‘Arrernte’. This has been wrongly misinterpreted as a failure, or at least an inexplicable oversight (Albrecht 2002: 4; Harms 2003: 130; J. Strehlow 2011: 271). The title of Kempe’s grammar: A grammar and vocabulary of the language spoken by the Aborigines of the Macdonnell Ranges, South Australia is typical of many early grammars (Threlkeld 1834; Teichelmann and Schu¨rmann 1840; Meyer 1843) in identifying the language by the location in which it was spoken. This may have better reflected the speakers’ method of identifying linguistic varieties.

52

It was the American linguist Ken Hale (1962) who coined the term ‘Arandic’ from ‘Aranda’ to refer to the higher-level family of related languages. In doing so he diverged from a common twentieth-century practice of using a term meaning ‘man/men’ or ‘person/people’ shared between the related languages to create a language family name, as evident in the Pama-Nyungan subgroups Karnic, Wati, Maric and Kuri. When, in the early 1960s, Hale coined the term ‘PamaNyungan’ (Wurm 1963: 136, editor’s note 2) he compounded words meaning ‘man’ in the languages spoken in the recognized extremities of the PN region—Pama ‘man’ in Cape York and Nyungar ‘man’ in the south-west of Western Australia. In doing so, Hale followed a ‘man-man’ compounding mechanism previously employed by Schmidt (1919a: 235) to name an Australian linguistic phylum that would not have been referred to in pre-colonial times. The PN subgroup that is today called ‘Thura-Yura’ (Simpson and Hercus 2004; Bowern and Atkinson 2012: 838) was earlier named ‘MeyuParnkalla-Tyura’ by Schmidt, using ‘meyu’ ‘man’ from the south-eastern Thura-Yura languages (Hercus 1992a: 1) and tyura from the north.4 In using the man-man mechanism Schmidt may have been influenced, at least in part, by Bleek, who, in 1858, had coined the term ‘Bantu’ as a descriptor of a group of related African languages, most of which shared the term ‘bantu’, meaning ‘men’, ‘people’, aba-ntu, PL-person (Silverstein 1968).

6.3.3 Northern and southern mainland languages The division of mainland Australian languages into two high level groups, one southern, one northern (Bowern and Atkinson 2012) was established in the earliest descriptive era by philologists who never visited Australia, initially by Wilhelm H.I. Bleek (1858), then by Friedrich Mu¨ller (1867; 1868) and later, and more compellingly, by the Austrian philologist W. Schmidt (1919b) (see Koch 2004b: 18–25), whose work is based on an impressively large range of primary sources. Schmidt’s mapped division was replicated in Italian by Giovanni Gatti (1934) (Figure 6.3), who had earlier (1930) published a grammar of Diyari within a broader discussion of Australian linguistic structure and classification. Although the present-day boundary between the PN and non-PN languages (Bowern, Chapter 1, this volume) aligns 4 Schmidt includes the term ‘Parnkalla’ in the compound ‘MeyuParnkalla-Tyura’ (1919a: 235) to denote the inclusion of a language that had had its name firmly established in the earliest record (Schu¨rmann 1844). Schmidt’s ‘Meyu-Parnkalla-Tyura’ subgroup was later named ‘Yura’ ‘person’ (O’Grady et al. 1966), and then, with the addition of Wirangu, ‘Thura-Yura’ (Simpson and Hercus 2004: 179) ‘in recognition of the th > y lenition which distinguishes the northern members of the subgroup.

nineteenth-century classifications of australian languages

Figure 6.3 Gatti’s (1934) map of Australian languages.

roughly with Schmidt’s delineation, Schmidt’s division of mainland Australian languages into ‘northern’ and ‘southern’ divisions differed from the present-day PN and non-PN division in including Arandic languages from Central Australia and Paman languages from northern Queensland into his ‘northern’ sub-group. The aberrant sound systems of these language families, which have lost initial consonants, and in some cases initial syllables (Alpher, Chapter 16, this volume), led Schmidt to exclude Arandic and Paman from the groupings to which he assigned their southern neighbours. Later twentieth-century reconstruction of soundshifts within Arandic and Paman languages (Capell 1956: 100; Hale 1964 respectively) has confirmed the relatedness of these language families to other mainland Australian languages. W. Bleek (1858: prelim. page) divided Australian languages into three categories, Southern Australian languages, Northern Australian languages and Tasmanian languages. Bleek’s 1858 classification, like Grey’s map (1845), was made before Europeans began to penetrate the centre of the continent, and before languages which came to be welldescribed in the later part of the nineteenth century, such as Diyari and Arrernte, had been encountered by Europeans. Paucity of data about Tasmanian languages prevented Bleek (1858: 30) from firmly classifying Tasmanian languages distinctly from mainland languages The material for a knowledge of Tasmanian Language or Languages … did not as yet enable him [i.e. Bleek, ‘the compiler’] to get an insight into their grammatical structure. It is therefore merely possible that the Tasmanian dialects belong to the Australian Family of Languages.

Yet the classificatory partition between languages from the north and the south of the continent is maintained, although it is based on a single manuscript describing the ‘Northern Languages’. Bleek (1858) does not refer to Confalonieri’s (1846) Iwaidjan vocabulary from the same region. The document informing his classification is listed as ‘A short vocabulary of the Port Essington language. – Drawn up by Mr. Wallach, Mate of H.M.S. Britomart’, and is said to contain 170 words and sentences (Wallach, no date).5 Bleek (1858: prelim pag.; 1869) had classified African languages into ‘suffix pronominal’ and ‘prefix pronominal’ classes. The Iwaidjan language recorded by Wallach is among the typologically diverse non-Pama-Nyungan languages from the north of the continent (see Evans 2000a), which Capell (1956) termed ‘prefixing’. While some nonPama-Nyungan languages utilize suffixes, and prefixes occur in limited domains in some Pama-Nyungan languages, Bleek’s division of Australian mainland languages into two classes was most probably motivated by a perceived prevalence of prefixes in ‘northern’ languages, which contrasted with what was known about the ‘southern’ languages, which had been characterized as suffixing (Schu¨rmann 1846; Moorhouse 1846; Prichard 1847: 278). Following Bleek, the notion that mainland Australian languages constituted two Sprachsta¨mme was put forward by F. Mu¨ller (1867: 241; 1868: xxiii), professor of Sanskrit and comparative philology at Vienna University, who like Bleek, 5 The Britomart arrived at Port Essington on the Cobourg Peninsula in 1838. The ‘Mr Wallach’ referred to by Bleek is very likely to be the ‘Lieu. Vallach [sic] of H.M.S Britomart’ referred to by Stokes (1846: 393), from whom Stokes said he ‘received much valuable information respecting the natives’.

53

clara stockigt proposed northern and southern subdivisions of mainland Australian languages. F. Mu¨ller, a member of the Imperial Academy of Science in Vienna, authored the linguistic (1867) and ethnographic (1868) reports of the Voyage of the Austrian Imperial ‘Novara’ Expedition. The frigate circumnavigated the world in 1857–1859 and docked in Sydney for a month in 1858. Mu¨ller did not, however, take part in the expedition or ever visit Australia. Mu¨ller sourced his material from Bleek, with whom he corresponded. With twenty-six pages of the report devoted to the description of Australian languages (1867: 241–66), this Viennese publication introduced European philologists to a large amount of information about Australian morphosyntactic structure.

6.3.4 Subgroups of ‘southern’ languages As scientific theories of evolution revolutionized midnineteenth-century thinking, the focus of philologists concentrated on genealogical classification of languages in order to determine the origins of people. Bleek and F. Mu¨ller’s ideas sit within a short-lived sub-school of comparative linguistics that was particularly well-developed in Germany, that theorized about the origin of language within biological evolutionary frameworks. While nineteenth-century classifications generally assigned language types an evaluative ranking according to a range of criteria (Morpurgo-Davies 1975: 671), Bleek and F. Mu¨ller took a transformationalist approach, believing that some linguistic types were the evolutionary precursors of others. Their ideas were formulated in conjunction with Ernst Haeckel (di Gregorio 2002: 239–61), the professor of comparative anatomy at Jena University, who was Bleek’s first cousin. The study of language approached that of social

Figure 6.4 Contents page from F. Mu¨ller (1882: ix).

54

and physical anthropology in providing important empirical data on which one could propose not only divisions between human groups but their evolutionary status. The classification of people was arranged within putative hierarchies with assumed correlations between the relative ranking of physical, linguistic, social and religious types. The methodology both confirmed, and was sustained by, preexisting beliefs that hunter-gatherers with sparse material culture spoke ‘simple’ languages and engaged in base social and religious practices. That evolution could be conceived of as being either progressive or regressive provided the model with convenient flexibility. Between 1876 and 1888 F. Mu¨ller published Grundriß der Sprachwissenschaft (Outline of Linguistics) in four volumes, which presented grammars of over one hundred languages from around the world. The Australian linguistic material is presented in vol. 2, Pt 1 (1882: 1–98) “Der schlichthaarigen Rassen” (the smooth-haired races). This ninety-four-page work on Australian languages was by far the largest single published collection of Australian grammatical material of its time, and remained so until Schmidt’s publications, which were also written in German (1919a, b). Mu¨ller’s four-volume publication classified and ranked races according to hair-type (1876: 24) and on this matter he referred to Haeckel (1876: 72–3). Here Mu¨ller re-publishes German translations of nearly all existing Australian grammars, while reanalysing some of the data presented in the original publications (Stockigt 2015: 353) (Figure 6.4). In this way Mu¨ller’s work differs from E. Eylmann’s (1908) German translations of Australian grammars, which did not provide alternative analyses. That the criteria which Mu¨ller used to place Australian languages among those of the world are now so arcane has almost certainly contributed to the fact that Mu¨ller’s

nineteenth-century classifications of australian languages Australian material has been overlooked in more recent histories of Australian linguistic classification (Elkin 1937a; Dixon 1980; 2002; Koch 2004b; McGregor 2008a), although the work was earlier referred to by Ray (1925).

6.3.5 Early rankings of Australian subgroups Bleek and Mu¨ller proposed an internal classification of Australian languages, which depicted some as more highly evolved than others. A differentiation in grammatical sophistication, from the east of the continent to the west, had first been perceived by Moore (1842: x), who in a description of Nyungar spoken in Western Australia stated: The grammatical construction appears to be inartificial and elementary, as might naturally be expected among so rude a people and wholly free from that startling complexity of form (especially as regards the verbs) which has been attributed to the Sydney languages in THRELKELD’S Grammar.

Moore’s perception, which was soon refuted by Schu¨rmann (1844: v), is likely to have resulted from the higher quality, and greater detail of Threlkeld’s material (1834) in comparison to what Moore was himself able to discover about Nyungar. Bleek (1872: 95), who had access to Moore’s Descriptive Vocabulary, similarly assigned a higher ranking to ‘Lake Macquarie Language’ described by Threlkeld (1834), on the grounds that Awabakal had gendered third-person pronouns, from which Bleek concluded that the language was less degenerated than other Australian languages. In the introduction to his ethnological volume Mu¨ller (1868: xxiii) proposed that the Su ¨dliche Abteilung (southern portion), being one of the two mainland language families, was itself divided into three groups. These were described as being westerly, middle and easterly. In 1882 Mu¨ller (pp. 2–3) expanded the hypothesis, postulating a three-step ranking of Australian languages. Subscribing to a dynamic interpretation of Schleicher’s (1859) enduringly influential three-way typological division of languages as isolating, agglutinative, or inflectional [read fusional] classes (Morpurgo-Davies 1975: 655), in which fusional languages were the most highly evolved, Mu¨ller classified Ramindjeri, of the middle subgroup, described by the missionary Meyer (1843) as more highly evolved than other Australian languages: Morphologically, the Australian languages fall into several categories. Some individual languages, for example, the languages from the West are no more advanced than are the languages without form from Indochina (Burmese and Thai). Others — for example, the language from Lake Macquarie — show agglutinative constructions reminiscent of the Ural Altaic languages, while others — for example, the language of

Encounter Bay — in the way they fuse the word-forming elements with the stem, show a tendency to raise themselves up to a higher level.6

Some early Australian grammarians (Schu¨rmann (1844: 4; Moorhouse 1846: 5; Taplin 1878: 8) considered that the agglutinative marking of case and number on nouns in Pama-Nyungan languages did not constitute a ‘declension’, a term from Latin declinere ‘to lean’, with origins in a conception of the marked case—from the Latin cadere ‘to fall’, casus ‘falling’—falling away from the nominative. Ramindjeri, however, supposedly had a ‘real’ declension, because the marking for case and number on nouns was presented as irregular (Meyer 1843; Horgen 2004: 96, 101), which resulted in Mu¨ller assigning the language to an evolutionary rank above Awabakal, ‘Lake Macquarie’, which he described as agglutinative. Awabakal was in turn seen as more highly evolved than the ‘morphologically formless’ Nyungar, ‘from the West’.

6.4 Concluding comment The pathways by which linguistic theory and practice moved between Australia and the rest of the world have remained largely untraced. The study of the nineteenthcentury typologies and classifications of Australian languages reveals the routes by which ideas about Australian languages were transferred between Australia and Europe. The exchange of ideas between missionaries posted among Aboriginal populations in remote Australian locations and European or urban Australian intelligentsia— for example, that between W. Ridley and M. Mu¨ller, between G. Taplin and W. Bleek, and between C. Strehlow and M. von Leonhardi—provided some conduit through which ideas were exchanged. The study of nineteenth-century ideas about the relatedness of Australian languages is unlikely to provide useful historical background to late twentieth- and twenty-first-century understandings about Australian phylogenetic structure. Significant theoretical and methodological disjunctions between current thought and practice, and nineteenth-century assumptions about the evolution of languages and of the people speaking them, renders much of the material examined in this chapter obsolete beyond a historiographical examination of the development of ideas about Australian languages and people. 6 ‘Morphologisch fallen die australischen Sprachen in mehrere Kategorien. Einzelne derselben (z. B. die Sprachen des Westens) stehen nicht ho¨her als die formlosen Sprachen Hinterindiens (das Barmanische, das Siamesische), andere (z. B. die Sprache vom Lake Macquarie) zeigen einen agglutinirenden Bau, der an die ural-altaischen Sprachen erinnert und wieder andere (z. B. die Sprache von Encounter Bay) zeigen die Tendenz; sich durch Verschmelzung der wortbildenden Elemente mit dem Stamme zu einer ho¨heren Stufe zu erheben’ (F. Mu¨ller 1882: 2–3).

55

chapter 7

How many languages are and were spoken in Australia? Claire Bowern

7.1 Introduction There are many definitions of ‘language’ and many ways to count languages (cf. Anderson 2012). Languages are social constructs created by communities, named by the communities themselves and by the communities that surround them; linguistic groups are dynamically constructed and can be re-constructed, reformed, and renamed as circumstances change. The current chapter takes one view of these questions. Here, I discuss the counting of Australian Indigenous languages in the past, along with estimates of numbers of languages currently spoken and signed. In the last fifty years, opinion has settled on 250 (Wurm 1972; Dixon 1980; Walsh 1997b) as the number of distinct Indigenous languages (as defined by linguistic measures) which were spoken in Australia prior to European invasion and settlement. This chapter calls that number into question and argues that the extent of linguistic diversity has been substantially underestimated, and that a more realistic number is at least 400 (with some additional uncertainty which will never be resolved). In order to examine this question, we need to investigate the genetic classification of Australian languages, the plotting of languages on maps, and the dynamics of language endangerment and loss over the past several hundred years. This chapter should therefore be read both as an investigation into the number of languages and as further information about the classification and maps given in the front matter of this handbook. In Chapter 6 of this volume, Stockigt presents a detailed investigation of the classification history of Australian languages in the 19th century. The current chapter does touch, at least to some extent, on classifications and their criteria in the 20th century. Koch (2004b; 2014a,b) has discussed the 20th -century classifications in some detail, ranging from Schmidt’s (1919a) classification to O’Grady et al.’s work in the 1960s (e.g. O’Grady, Voegelin, and Voegelin 1966; hereafter OVV66) to Dixon (2002). Rather than providing a detailed comparison of classifications or a detailed discussion of the methods of each author, I focus here specifically on issues

around the number of languages in Australia. In doing so, I provide some context for the typological studies that are the focus of Parts 2 and 3 of this volume while not duplicating work that is available elsewhere. Some additional relevant discussion can be found in the classification’s notes, given at the front of this volume. This chapter therefore explains the choices around what is represented in the classification and language map (and in doing so discusses some issues around language, culture, land, and communities). It then moves on to explaining and discussing the number of languages spoken in Australia at various times, from Walsh’s number of ‘250’ (in Section 7.2) to the current figure and classifications (in Section 7.5). Finally, it briefly presents issues around language endangerment leading to language loss (foreshadowing the chapters in Part 5 of this volume). As has been discussed elsewhere in this volume, the relationship between language, land, and family is complex in Australia. Languages are often ‘owned’, and ownership rights come through the land (rather than through being a speaker). Thus linguistic identity relates to land tenure, which relates to family and community structure (see further Sutton 1978: 14; 1997; 2020; Rumsey 1993). This relationship to language also provides a stability for multilingualism (see further Vaughan and Singer 2018; Vaughan, Chapter 54, this volume): while many members of a community may be multilingual, there is a discrete set of linguistic identities that come through land and family. That is, traditionally, a person’s linguistic repertoires do not define their linguistic identities and their relationships to land and language directly: one’s linguistic affiliations come from country first, and secondarily from community. This makes plotting languages on maps possible in the abstract, though of course, there are still difficulties in practice.1 1 One should also note that while this is a common model of land ~ language tenure, it is not the only model in evidence within Australia; in the Western Desert, for example, this model is inappropriate. It also does not seem to describe the way Bardi elders talked about their relationships to

Claire Bowern, How many languages are and were spoken in Australia?. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Claire Bowern (2023). DOI: 10.1093/oso/9780198824978.003.0007

56

how many languages are and were spoken in australia?

7.2 The number ‘250’ Walsh (1997b) is a clear summary of the issues around counting languages, and whether the number ‘250’ is an accurate estimate. He points out many ways in which the concept of the number of ‘languages’ is poorly defined: the term ‘language’ means different things in different contexts, and named linguistic varieties can represent several types of linguistic abstraction. He concludes that the number of languages has not been reliably estimated, in part because the pre-European Indigenous population estimates are so variable. He also notes (Walsh 1997b: 410) that current estimates of language ~ population size are not consistent with early estimates of language numbers, and that either ‘there were many more distinct linguistic varieties at first contact or there was a much larger population for each variety’. This point is also made forcefully by Sutton (2020) through an examination of languages and patrilineal estates. As he puts it (Sutton 2020: 358–9): Dixon has also estimated that on linguistic rather than geopolitical criteria there were probably in the range of 230 to 300 languages spoken in Australia at colonization, in the sense that a linguist distinguishes a language from a dialect (Dixon 2002: 7). At an assumed population of 1,000,000 this would yield average figures of 3,333 to 4,348 persons per grammarian’s language. These figures are surprisingly large.

Nonetheless, the number ‘250’ does not appear to have been a controversial figure for Australian languages, either before Walsh’s (1997b) publication or after. There has been a perhaps surprising uniformity around the number of languages (defined by linguistic criteria, such as mutual intelligibility). Mu¨hlha¨usler (1987: 1) gives 250 languages as ‘derived from standard lexicostatistical investigations’ (not otherwise referenced). Presumably, the origin is as discussed in O’Grady and Klokeid (1969: 300): following the O’Grady, Voegelin, and Voegelin (1966) classification and the O’Grady, Hale, and Wurm (1966) map, speech varieties were classed as separate languages where they shared fewer than 70% of items on a standard wordlist. This measure grouped 502 speech communities into 228 separate languages. They acknowledge that sampling is uneven in some parts of the country and the figure is not definitive. Likewise, Evans (2008: 342) gives the figure as ‘around 250 languages, or about 700 distinct linguistic varieties if one follows Aboriginal practice in how to measure language diversity’. Gaby (2008c: 211) phrases it as ‘more than 250 languages’. Likewise, Karidakis land and language: that ‘being Bardi’ comes both from knowing the language and having rights to country. Further discussion of differences and commonalities is given in Vaughan, Singer, and Garde (2022).

and Kelly (2018: 105) give the figure as ‘more than 250 languages with numerous dialects’, citing Dixon (1980) as the source, though Dixon (1980: 18) actually gives the figure as 200, rather than 250. Dixon (2002: i, 2), however, does give 250. Mailhammer (2021) cites Karidakis and Kelly (2018) for a figure of ‘about 250’. Smolicz and Secombe (2003: 5) say ‘some 250 languages, as well as some 600 dialects’ (this claim is unsourced). McKay (2011a: 297) says ‘approximately 250 Indigenous languages … many of these with multiple separate dialects’. Koch (2007b: 23) quotes Dixon (2002: 2) with the figure 250 but qualifies it along Walsh’s lines. McConvell and Thieberger (2001: 16) also discuss the historical number of Indigenous languages, coming to similar conclusions as Walsh (1997b), though emphasizing the issues around language names as a cause of variation in the total number of languages, as well as pointing out that linguists’ definitions of ‘language’ do not accord with Indigenous definitions in all cases. Two points emerge here: first, the number ‘250’ is fairly constant across discussions, whether directly referenced or not. Secondly, we also see that some authors give two numbers: a lower number and a higher one, with the higher number varying more extensively. This higher number is represented as ‘dialects’ (that is, subunits within a language) or as ‘distinct linguistic varieties’. This apparent recent uniformity is interesting, since Walsh (1997b: 395) says that ‘specialists in the study of Australian Aboriginal languages give a range of responses’ to the question of how many languages were spoken in Australia prior to European invasion. This quoting of two numbers, the 250 number and a higher number, seems to address Walsh’s (1997b: 396) point about the difficulty of defining ‘languages’ as distinct, based on mutual intelligibility, as opposed to a more general notion of ‘named variety’. Walsh and Yallop (1993: 1) also draw a distinction between distinct languages and internal dialectal varieties, but with the point that it is difficult to be precise because languages and dialects are themselves not very precise terms.2 The collective impression from these sources is that the number 250 is an approximation, but not a poor one. While there is discussion of whether a figure like 250 (drawing on linguistic definitions) or a number closer to 500 or 600 (drawing on Indigenous definitions) is most accurate, no one raises the issue of whether the number 250 is itself problematic, other than where arbitrary dialectal cut-offs may be drawn. One might wonder why this question is important. After all, all languages are important, and whether there were 100, 250, or 500 (or more), Australia’s linguistic diversity 2 The Austlang catalogue (austlang.aiatsis.gov.au) does not draw a distinction between language and dialect, instead listing all varieties, as discussed in Section 7.3.3 below.

57

claire bowern is clear—the number of languages still widely spoken is a very small fraction of those that were used two hundred or more years ago (cf. Schmidt 1990; Henderson and Nash 1997; Marmion et al. 2014). It should also be recognized that whatever single number is arrived at, it is only a single snapshot of a dynamic continent at one point in time. The linguistic situation would not have been static, given the population dynamics and climate change over 50,000 years (cf. Pascoe 2014a; Smith 2013; Hiscock 2007). Languages continue to split; small populations can merge and shift. Most views of Australian prehistory highlight climatic variation over long time scales as a source of population variation and movement, which of course has consequences for language use as well (Smith 2005; Veth 1989; Hiscock 2007; Bourke et al. 2007). There are several reasons why an accurate number and classification is important. One is in constructing samples for typological surveys, as we want to make sure we are capturing the appropriate variation. In reconstruction, it is crucial to know which changes should be taken into account in reconstructing prior stages at the subgroup or family level. And of course, there’s simply the aspect of ‘getting it right’—there’s no point in using a number if the number is known to be inaccurate. There is also the point of representation: if the right number is closer to 350 but estimates are usually around 250, languages and groups are being left out. And if we truly cannot reach an accurate estimate, then at the very least, we should dispense with treating the figure ‘250’ as uncontroversial. One might also wonder why a linguistic (that is, a linguist’s) definition of language is needed; after all, we could use an Indigenous definition of language that takes better account of the social roles of language and language naming. Named ‘languages’ are, after all, a social construct, so it makes sense to recognize the knowledge of those who made those social constructs in the first place. To take this position, however, implies that there is a single ‘Indigenous’ classification that all Indigenous groups agree on, which is not the case. Moreover, many Indigenous views of language (as described by Sutton, e.g. 2020) have different levels of linguistic groupings, some of which are named, some of which are not. Finally, some Indigenous names refer to multiple distinct groups. For example, the name Wangkumarra is both a particular linguistic variety and a more general term for a set of people, who probably spoke several different languages. Likewise, the term Yolngu Matha is a term of contrast rather than a single language; it can refer to Aboriginal languages of Arnhem Land in contrast to English, to Yolngu (Pama-Nyungan) languages in contrast to the non-PamaNyungan languages of the region (such as Burarra), or to a subset of those languages (those that use the word yolŋu for ‘person’). While such terms reflect a dynamic construction

58

of languages, they are difficult to use when the aim is to have stable identification for particular, comparable linguistic varieties.

7.3 Calculating the number of languages 7.3.1 Sources of names The impetus for discussing the number of languages was the anomalies that arose from my compilation of the language map and classification given in the front matter to this volume, as well as from totaling the language names used in the literature.3 Comparing prior classifications made it clear that there were inconsistencies between different parts of the country. While many classifications had similar language numbers in total, they varied extensively in the amount of detail given for different subgroups, such that the combined number of languages was greater than 250 (and possibly substantially so). The example of Wurm (1972) and Dixon (2002) is discussed below. Secondly, these classifications listed languages which did not always line up with the languages depicted on maps. The map produced for this volume is designed to show all Pama-Nyungan (and Australian) languages, but not dialects within a language. Compiling the map information gave a set of ‘languages’ based primarily on geography. Thus, the language maps provided another set of language names, though again these did not line up precisely with classifications based on language family and subgroup.4 The other way of constructing a language set was to go from names in the literature on Australian languages. These names were associated with publications primarily on language and cultural materials, such as oral histories. I first compiled a list of ‘variety’ names; currently, there are about 3,000 unique variety names in the Chirila database, representing a combination of spelling variants, alternative names, varieties named after individuals (such as are found in Zorc 1986), ethnonyms, and clan names used to refer to language. These items were then associated with a reference list of languages and matched for similarity. 3 My original intention was not to challenge estimates of number of languages from previous work. Only after this work had been done did it emerge that the numbers were rather different from previous estimates, which led me to try to understand why the estimates were so different, and in particular, whether my methods had led to inaccuracies in counting. 4 Note that these discussions are separate (at least somewhat) from the discussions around Walsh’s (1997b) lines of argument, which also include population numbers. For example, I do not take estimates about the 18th -century population estimates of Aboriginal Australia to be particularly informative for the number of languages, since all estimates of language numbers are within a plausible range produced from population estimates.

how many languages are and were spoken in australia?

7.3.2 Variants in naming Some types of language name variation are straightforward. Variants can be simply orthographic (e.g. voicing or hyphenation): Nhanta vs. Nhanda, Yingkarta vs. Yinggarda, and Kok Nar vs. Kok-nar are examples. Yolngu vs. Yuulngu, likewise, simply reflects different spelling conventions. In other cases, we have phonemic spellings vs. earlier spellings, which also appear in published literature. Chowie as an alternative for Jawi (cf. Bird 1910 vs. Bowern 2012a), or Dieri for Diyari are examples. Such variation is straightforward to address, though care must be taken where there are different languages with identical or nearly identical names. One example is Kungkarri vs. Gunggari: these are two distinct languages (the first Karnic, the second Maric) from Central Queensland (see Breen 1990b). Another is Kokatha vs. Kukatja, both Western Desert (Wati) varieties, but from different regions (cf. Platt 1972). Spelling variants are straightforward. However, there are many other sources of variation in naming: endonyms vs. exonyms, clan names used for language names (and vice versa), and language defined with respect to a particular place (either Indigenous, or a European name). This can mean that there are many distinct ways to refer to the same group of people and the language(s) they speak. However, these notes frequently also lead to contradictions. A good—but complicated—example of this is the name Mandandanji (or Mandandanyi). Tindale (1974) gives Mandandanji as one of several dialects of a language known as Kogai. He includes Gunggari and Barrungam as further dialects of the language. Breen (pers. comm. to Austlang database in 2006) suggests that Kogai, Gunggari, and Mandandanyi are probably equivalent terms; the first two being based on the Wangkumarra word for ‘east’, while Mandandanyi is a shibbolethnonym based on the word manda ‘go’ (see also Laughren 2013: 6). In the same personal communication, Breen suggests that Guwamu might also be a Kogai language. Several points are difficult to reconcile. First, Barrungam is actually a Waka-Kabi language (Kite and Wurm 2004); the others are Maric languages. Therefore, if Kogai is a term referring to language, either Barrungam must refer to more than one language group, or there has been recent and rapid language shift to a Waka-Kabi variety. Gunggari has also been regarded as a dialect of Gunya (Dixon 2002: 660 also includes Bidyara in that group), while Gunya and Marrgany (but not Gunggari) are closely related (cf. Breen 1981b). Breen (1973: 8) treats Bidyara and Gunya as distinct languages. Either we are dealing with a dialect continuum or with contradictory information—or more likely, both. Language locations also provide difficulties. Kogai is said to be the language of the Maranoa and Balonne Rivers;

Barrunggam is placed by Kite and Wurm (2004) on the Condamine River, and Guwamu is placed by Wafer and Lissarrague (2008: 323) north of the Culgoa River, that is, south of Kogai. Laughren’s (2013) map includes Gunggari and Guwamu, but not Barrungam or Mandandanyi. Holmer’s (1983) description of the Gunggari group places Gunggari in the area where Tindale places Mandandanyi (that is, Roma and Mitchell, and the Maranoa and Balonne Rivers). Therefore, the association with Kogai and Gunggari using geographical data is secure, as is the Mandandanyi and Gunggari link, but not the Barunggam and Kogai link. Though not all cases are as complex as the Kogai example, such issues around names repeat across the country. Thus, a linguist attempting to reconcile these statements cannot simply rely on ‘authorities’, because the authorities provide contradictory information. In such cases, we either need to choose on the basis of some principle, or work directly from original linguistic sources, such as wordlists.

7.3.3 Languages and dialects Next is to decide on how many ‘language’-level units should be included in a continent-level classification. As Walsh (1997b) and others have pointed out, languages can be named at many different levels of abstraction. Language is ultimately a social construct based on perceived similarity (and possibly ethnicity), so it is not surprising that mutual intelligibility is only one factor that comes into play around views of ‘sameness’ or ‘otherness’ in language. Mutual intelligibility is also not a good criterion for deciding on language distinctions in an area where speakers are routinely multilingual; that is, it has problematic application in practice, because varieties are likely familiar to speakers (see also Vaughan and Singer 2018). Mutual intelligibility is also problematic in dialect chains. In deciding on the ‘language’ vs. ‘dialect’ question, I looked at several different types of information. For example, how was the variety described in language material with respect to other varieties? Was it described as distinct? Did the people who were described in the source talk about themselves as one group, or different groups? In some areas, such as the Yolngu Matha speaking area of Arnhem Land, we have information from the way speakers of the languages themselves describe language differences and language classifications. Much of this information was difficult to work with, but it did give a sense of local and regional variation, which informed decisions about how many names to include. In some parts of the country, categorical decisions were simply impossible (e.g. in the case of the complex dialect chains around Bandjalangic, Western Desert, and Paakantyi). In such cases, for the purposes of counting languages, I have taken a lower and upper estimate.

59

claire bowern Finally, the language vs. dialect issue brings up another question about language names. Some language catalogues have names that refer to groups at different levels of granularity. Glottolog.org, for example, classifies strictly on the basis of mutual intelligibility, or whether the language is ‘assertably distinct from all other known languages’.5 Another major compilation of language names classification is Austlang (austlang.aiatsis.gov.au). It is keyed to the AIATSIS language codes (e.g. A53 Banyjima).6 The downloadable Austlang database as of November 2021 contains 1,204 names, 835 of which have map data (that is, latitude and longitude coordinates). These are, per Austlang, ‘variety’ names. That is, they are named speech varieties, not differentiated by mutual intelligibility, and not a single level of classification. The AIATSIS code system, for example, has three codes that relate to the Bunuba language: K5 for ‘Bunuba’; K67 for ‘Eastern Bunuba’; and K68 for ‘Western Bunuba’. One could treat Bunuba as having either one language or two, but not three, since K5 is a hypernym for K67 and K68. Therefore, care must be taken not to double-count varieties when using catalogue entries from Austlang.

7.3.4 Using phylogenetics and classifications to evaluate similarity In some areas, I also used the results of phylogenetic analysis to draw conclusions about the similarity of languages and wordlists. I assume that varieties that are the same ‘language’ should be identical (or nearly so) on basic vocabulary. There are, of course, well-known problems with this approach. Slaska (2005) describes the lack of stability in elicitation tests for basic vocabulary, because of near-synonymy, register differences, and term frequency differences (compare ‘vomit’ to ‘barf ’, ‘spew’, ‘chuck up’, etc., where the term that appears on the basic vocabulary list might not be the term that is most frequently used). It may be that otherwise similar languages will appear more distinct on such lists (hence O’Grady’s 70% similar level for ‘dialects of the same language’). However, rather than an absolute figure, we can use phylogenetic similarity to be consistent across different areas of the country. For example, if Djambarrpuyngu and Gupapuyngu are usually regarded as ‘the same’ but are actually as distinct as Kaytetye and Alyawarre (which everyone agrees are ‘different’), perhaps 5 https://glottolog.org/glottolog/glottologinformation, as of 25 April 2022. 6 The classification printed in this book is not keyed to the AIATSIS codes; however, the Chirila database on which it is based (which can be found on the zenodo.org supplementary materials for this volume) does include the digital file of the classification and includes AIATSIS, Glottolog, and ISO-639 codes.

60

we should reconsider our treatment of Djambarrpuyngu and Gupapuyngu as ‘the same’. Phylogenetics has been used to untangle conflicting classifications. For example, for the Bigambal language, several conflicting classifications are given in the literature. Dixon (2002) classes it as part of the Bandjalangic subgroup, while Oates (1975) and earlier O’Grady, Voegelin, and Voegelin (1966) have related it to Wiradjuri as part of the Central New South Wales group. Wafer and Lissarrague (2008: 332ff.) relate Bigambal to Yugambal. Part of the conflict may stem from confusion in earlier sources between Yugambal and Yugambeh; Yugambal has been classed as Yuin-Kuri (Oates 1975; Wurm 1972) or Nganyaywana (Crowley 1997: 284), but Sharpe (2005) has also identified Yugambal as a name for a variety of Bandjalang—though this appears to be based on an equation between Yugambal and Yukumbil in Tindale (1974). As Wafer and Lissarague show, Yukumbil is a variety of Bandjalang, but Yugambal is not. Close similarities of terms combined with short wordlists from early sources have thus led to a series of conflicting classifications that can only be reconciled with direct comparison of the original wordlists. In this case, the phylogenetic tree clearly shows that the sources used for Yugambal and Bigambal (Curr vocabularies 176a, 179a, b, 180) are related to one another and not particularly closely related to any of Central NSW, Yuin-Kuri, or Bandjalangic. This process is helpful where there are named varieties with linguistic records. It is possible to treat varieties as units in a classification and evaluate their similarity.7 Working this way implies that the unit of analysis is the ‘doculect’—the document of a particular variety at a particular time. This is possible where information on the languages is available. But this is not always the case—for example, where we don’t have information on the varieties, or the sources are very different but are named as the same language. With regard to the Mandandanyi example in Section 7.3.2 above, there is no language data to my knowledge which is unequivocally associated with the name Mandandanyi. There is material labelled Gunggari (Holmer 1983), Guwamu (Austin 1980), Bidyara (Breen 1973), and Gunya (Breen 1981a), but the association with Mandandanyi and Maric comes through the equation of language names and the overlapping location of Mandandanyi and Holmer’s Gunggari data, not through data labelled as Mandandanyi directly. However, by grounding the classification in the named and recorded varieties, we come closer to a classification based on linguistic data, and we avoid pronouncements around mutual intelligibility. 7 In doing this my aim was not to replicate the lexicostatistical classification described by O’Grady and Klokeid (1969), but rather to make use of some concrete evidence of similarity in informing decisions around classification.

how many languages are and were spoken in australia? It was noted in Section 7.2 above that O’Grady, Voegelin, and Voegelin (1966) used lexicostatistical cutoffs to give an indication of which varieties were related as dialects and which were distinct languages. However, it appears that other researchers have not followed the same cutoffs, leading to underestimates of the number of languages. To take one example, Barranbinya is said to be in a dialect relationship with Muruwari (Oates 1988: 198–9), even though the shared percentage of vocabulary was 44%, well under the OVV 70% threshold. It is examples like this which make it more likely that there has been under-counting of languages, rather than over-counting.

7.3.5 Languages with no sources The Mandandanyi example raises the question of what we should do in the case of names that have no recorded language material. In the Mandandanyi case, we can be fairly sure that Mandandanyi is probably not a separate linguistic entity from Gunggari. That is, while we do not know precisely whether the designation of Mandandanyi totally or partially overlaps with the designation of Gunggari, the evidence we do have (through names and locations) makes it appear that it is not a distinct entity. What should we do in cases where no such information is available? There are numerous recorded names where linguistic data is either absent or difficult to interpret. In some cases, a small amount of information is sufficient. Pirlatapa, for example, is known from only one short recording, but that material is sufficient to recognize that it is closely related to the Karnic languages immediately to the north (Austin 1990b). Other cases are more complicated, as the Western Queensland languages described by Breen (1990b), where all the languages in a region are poorly attested. Another example is Kiyuk, the language of Peron Island. Nothing is known of the language, in that there are no records of it. Hammarstro¨m (2020) treats the name as ‘spurious’ and cites Ford (1990) as saying that reports are second-hand ‘and do not necessarily indicate that the language was actually unintelligible to those of the mainland’ (Hammarstro¨m 2020, slide 50). Ford’s material is as follows: ‘Agnes Lippo’s step-father, Daly Young, was one of the last speakers of Kiyuk. Agnes remembers Kiyuk as mutually unintelligible with Bachamal [= Batjjamalh [CB]]. There is no-one left with even passive competence in Kiyuk’ (Ford 1990: 9). That is, a fluent speaker of Batjjamalh reports being unable to understand a family member when he spoke in Kiyuk. Rather than an unconfirmed report, this would seem to be a clear case where even though we do not have records of the language, Indigenous reports are sufficient for us to include the language name. In this classification, I included

language names such as Kiyuk, but it is noted that there are no data, and that genealogical classification is unassertable. For a list of languages aimed at classification, perhaps such languages should be omitted, since without linguistic data, they cannot be linguistically classified. But for the purposes of ascertaining the number of languages or which languages were spoken where, such languages should be included.

7.4 Problems with the language/variety model In addition to the conflicting source problems described in the previous section, three areas were particularly difficult to understand in terms of a language/variety model. These were the languages of the Western Desert (Pintupi, Ngaanyatjarra, Ngalia, and related varieties; see Babinski et al., Chapter 75, this volume), Yolngu languages, and the varieties called Paakantyi (Hercus 1982, n.d.-a). In those cases, the tests were difficult to apply. The Western Desert has more than one language; there are substantial grammatical and vocabulary differences across the region (e.g. verb systems, use of bound pronouns; see further Babinski et al., Chapter 75, this volume). But as Babinski et al. show, the innovations which could be used to define subgroups give conflicting evidence; it is clearly a complex dialect chain.8 My solution here was to name varieties which are well documented and different from each other, but that should not imply that they are the only ‘languages’ of the group, or that other varieties are ‘dialects’. The issues with Yolngu were somewhat different. In this case, there are multiple—but conflicting—levels of hierarchical classification that are partly ethnolinguistic and partly related to doculects.9 To take one example, consider the varieties called Dhuwal. Dhuwal is the word for ‘this’ in several Yolngu clan lects (or patrilects), including (but not limited to) Djambarrpuyŋu, Ḻiyagalawumirr, and Djapu. Dhuwal is contrasted with Dhuwala (also ‘this’) based on clan moiety: Dhuwal-speaking clans are Dhuwa moiety, while Dhuwala clans are Yirritja. Some language forms are specific to individual clans; others are used more broadly. Yolngu describe some vocabulary as bukmakku ‘everyone’s’; these are words which are common to many Yolngu clans and do not carry ownership rights. Other vocabulary, however, is particular to clans or clan groups; that is, it belongs to a particular clan (ba¨purru). However, while some clans are defined 8 The exchange in Miller (1971a) and Douglas (1971) provide similar points. Bowern (1998) provides discussion for the Karnic subgroup, where such changes are older (and subsequent changes make defining distinct languages easier), but where subgrouping innovations conflict. 9 Some information is from Schebeck (2001); other points were explained to me by Yan-nhaŋu (Yolngu) speakers over several years. See also Vaughan et al. (forthcoming).

61

claire bowern linguistically, other ba¨purru do not claim a distinct way of speaking (Keen 1994; 1995). Moreover, there is considerable heterogeneity in ways of speaking, both within clan lects and within varieties classed Dhuwal. Dhuwal is used as a lingua franca in the region and is widely spoken by people from different language groups and moieties across Eastern Arnhem Land. An additional complexity is added when we consider sources and how they are named. For example, Heath (1980c) called the variety he wrote about Dhuwal, which makes it look like the linguistic designation is broader than the Djapu or Djambarrpuyŋu grammars (Morphy 1983; Wilkinson 1991 respectively), because Dhuwal is a cover term for several varieties, including Djapu and Djambarrpuyŋu (both of which are represented in Heath’s documentation). Heath (1980c: 1) notes that the texts in the volume contain considerable dialectal variation. Heath worked with four speakers (two from Djapu clans, two from Djambarrpuyŋu). Thus, it is not actually representative of a broader speech variety.10 Second, both Wilkinson and Heath work on varieties called Djambarrpuyŋu, but their descriptions of the language differ. Djambarrpuyŋu is both a lingua franca (both under that name and under the name Dhuwal), so what we might be seeing in Wilkinson’s description is from a broader range of speakers, but younger and from further east (Galiwin’ku), while Heath’s were men from further west. This raises the point that languages vary internally, and that variety is seldom adequately captured in language documentation (Heath himself makes this point).11 This can lead to focussing on variation between language varieties without fully contextualizing it in the variation within languages. In the Paakantyi case, many names were given, but most varieties are not distinctly documented, making it difficult to determine how many names should be included in a classification.

7.5 Number of languages and sources of under- or over- counting Dixon’s (2002) classification includes 185 Pama-Nyungan languages (treating Western Desert as a single language but Yolngu as nine). There are also 61 non-Pama-Nyungan languages and the Papuan language Miriam Mer. Somewhat inconsistently, he includes Miriam (though a Papuan 10 In making this point, it is not a claim that Heath intended the scope of the grammar to be broader than it should have been; it is simply the point that language descriptions called by general language names cannot be assumed to cover broad speech varieties. 11 There are a number of reasons for this, including work with the last speakers, work with few speakers from necessity, and work which focusses on points which do not vary as much.

62

language) but no Tasmanian languages. This leads to a total of 246 languages. O’Grady, Voegelin, and Voegelin (1966) had 29 families and 230 languages, 160 of which were in Pama-Nyungan (counting Tangkic and Yanyuwa). Wurm counts 82 non-Pama-Nyungan languages. Like OVV66, he includes multiple Wororan languages, where Dixon has only three. His classification includes 174 Pama-Nyungan languages (counting Tangkic and Yanyuwa among the nonPama-Nyungan languages). He includes very few Maric languages, and, like Dixon, a single Western Desert language (though mentioning ‘Kardutjara and others’). Thus, Wurm’s total is 256. Austlang, as described above, has 1,204 names, 835 with a geographic location. Glottolog 4.4 includes 343 Australian languages (excluding Tasmanian), 248 of which are Pama-Nyungan. The current classification for this volume, in contrast, has 490 languages. It contains 106 languages from nonPama-Nyungan families and 384 Pama-Nyungan languages. (Also included in the classification—though not compared in these counts—are Tasmanian languages and new Indigenous languages.) While 490 is the number arrived at through the methods described above, I tended to err on the side of more, rather than fewer, language divisions. Removing those names where there is doubt (or where I included them only tentatively) reduces the count of languages to 440. Note, however, that this is still a much higher figure than previous counts, and is similar to Glottolog’s figure, which does not include languages where classification is impossible. This section describes the sources of difference in counts between the current classification and previous ones. One, as mentioned above, is simply inconsistency across earlier classifications. One should not have nine Wororan languages and eight Yolngu languages but only three members of the Nyulnyulan family, going by speaker perceptions, mutual intelligibility, or vocabulary distance. The biggest single difference between these classifications arises from the Maric subgroup of Pama-Nyungan, for which OVV66 and Wurm list only three languages—just given as Mari in Wurm—whereas I have 20 languages.12 However, although Dixon’s classification for Maric is also considerably expanded compared with the earlier classifications, this does not of itself account for the differences between the current and earlier classifications.13 Dixon includes more Pama-Nyungan languages (most of which are in the Maric group) but has fewer non-Pama-Nyungan 12 Note that Barrett (2005) works with 19 languages for his Maric classification. I have additional languages, following Breen (2009), because my classification includes languages with sparse data, which would not have featured in Barrett’s work, as his aim was identification of sound change. 13 Dixon (2002) includes only one language for Noongar and Western Desert/Wati.

how many languages are and were spoken in australia? languages. He includes substantially fewer named languages in Wororan and Nyulnyulan (his Fitzroy River Subgroup), and only one language for Umbugarla (missing from OVV66 and Wurm). A substantial source of difference between the earlier and later classifications is likely to be survey work which was completed after the earlier publications. That is, there are more languages in the current classification as linguists’ knowledge of the languages has increased. For example, for the Mayi subgroup of Pama-Nyungan, OVV66 and Wurm have three languages and I have six. My classification is based on Breen (1981b), which wasn’t published at the time of the earlier surveys. This classification also has the benefit of material published in the 50 years since Wurm’s classification, which has increased the literature on classification extensively, particularly for non-Pama-Nyungan languages. There has also been a sizable amount of work on 19th -century sources, which elucidates the relationships between language names in the South and East of the country. Wafer and Lissarrague (2008) is probably the most important handbook from this region. Areas of Pama-Nyungan where this likely accounts for big differences in language numbers include Karnic (especially the ‘Karnic fringe’, for which Breen 1990b is the major compilation of sources) and Kulin, though all language areas of Australia have had revised local subgrouping or other linguistic work in the meantime. A further source of language counting relates to practices around grouping poorly sourced languages together with languages that are better known. For example, Dharawal and Dhurga (Eades 1976), Bidyara-Gungabula (Breen 1973), and Punthamara-Kalali (Holmer 1988) are all examples where a better attested language is described, along with comparative notes (or further materials) of the other variety. However, although these pairs of languages are described in the same volume, it doesn’t automatically follow that the languages are necessarily the most similar to one another, or that the names should be accorded ‘dialectal’ status.

7.6 Changes in the number of languages over time Even if we get a figure for a particular point in time, that number of languages will change. In this section, I discuss some of the processes that have added or removed languages from the record in the pre-colonial and colonial period.

7.6.1 Pre-colonial period Pre-colonial language change has been the subject of much discussion, particularly as it relates to how Australian

language change compares with language change elsewhere in the world. This discussion has heavily revolved around the applicability of the family tree model and rates of language change. Dixon (2002) suggests that the usual methods do not apply, due either to exceptionally high amounts of language contact or different processes of language change. Since then, however, reviews of that work have comprehensively shown that Australia is like other parts of the world: variable in the amount of language contact, with parts of the country that show more treelike splits than others. (For discussion around these questions, see Sutton and Koch 2008; Evans 2005; Bowern et al. 2011; McConvell and Bowern 2011; Haynie et al. 2014; Bowern and Koch 2004; Alpher 2004a; and O’Grady and Hale 2004. Miceli and Bowern (Chapter 5, this volume) also provide some discussion of this topic. At this point, it’s probably sufficient to say that the causes of language change in Australia are the same as elsewhere. Children learning the language(s) of their community have the task of generalizing from the speech and sign they are exposed to, to create a mental grammar which replicates the linguistic structures that allow them to communicate with their community. Individuals are exposed to different sets of structures and so come to different conclusions about the underlying grammar of those languages; furthermore, language variants can index social features, and language users have repertoires of these variants. In addition, new variants can spread across communities, and over time, can replace other variants. When those variants accrue in social groups, new languages emerge. Although all these areas of change apply across the world, this very general framework still provides scope for very different outcomes of language change. For example, different patterns of multilingualism and language contact, different relationships between language and identity, and patterns of migration against language diversification in situ, will mean that tree models are able to capture the dynamics of change to different degrees. In the record of language we have available, we do not have clear evidence for language mergers, mixed languages, or creolization (though see Dench 1998a for Pidgin Ngarluma and contact varieties from Northern Australia with Macassar cf. Evans 1992a; MacKnight 1976). Vaughan and Singer (2018) have talked about different patterns of multilingualism, language contact, and code switching, for both Indigenous and settler/colonial varieties. We have evidence for contact-induced convergence (Heath 1978a; Hercus 1987; Dench 2001), and Ellison and Miceli (2017) have provided evidence for contact-induced divergence. There is some evidence for what we might call ‘lineage pruning’ in the history of varieties. This is where dialectal varieties are merged into a larger regional ‘standard’ group. This goes along with complex dynamics of group membership, affiliation, and the signalling of difference through language. This can be illustrated with Jawi,

63

claire bowern a Nyulnyulan variety from the islands off the coast of the Dampier Peninsula. Jawi, as recorded in the earliest sources, has features that make it different from Bardi, the language that borders it most closely. For example, the Jawi in the Laves materials from c. 1930 attests a sound change where intervocalic ŋ is deleted (the allative is -an for vowel-final stems, cognate with but distinct from Bardi -ŋan). There are also a few different verbal agreement forms for third person plurals. However, the Jawi of speakers recorded in the 1960s and later does not have these features; they use forms with -ŋan in all contexts. In sum, we do have evidence for some language creation and language loss in the pre-colonial period, though no ways of quantifying the number of varieties that might have existed. Miceli and Bowern (Chapter 5, this volume) and Sutton (2020) have further discussion of this point.

7.6.2 The colonial period The colonial period has led to language loss and language gain—an extensive reduction in the number of languages, but also the creation and reclamation of new languages. There are many paths to language endangerment. Some of the most common are discussed in global terms in Campbell and Belew (2018), focussing on the ways in which different language practices make it more or less easy for languages to be transmitted. Causes of language endangerment fall into three types: those that affect the number of people who know and use the language—for example, through genocide; those that affect intergenerational transmission; and those that facilitate rapid language shift. Some of this work is also portrayed in terms of language choice—that is, community members choose which language to speak in which circumstances. However, this framing of language use as language ‘choice’ can be interpreted as portraying a neutral ‘choice’ in the context of language events which are far from neutral. As I argued in Bowern (2017a), for example, older children in boarding schools technically had a ‘choice’ in what language(s) they spoke, in the sense that at any given time, a multilingual individual can use any language in their repertoire. On the other hand, to choose their first languages (i.e. the Indigenous languages) was to risk physical and psychological punishment, a consequence which was well documented. That is not a choice for a seven-year-old in practice, and it should not be framed as such. As mentioned above, we can describe language endangerment factors in several different ways, including those that disrupt intergenerational transmission: that is, the languages became endangered because of policies, practices, and changes which made it harder for children to acquire the languages of their parents and grandparents. These

64

included, most obviously, the stolen generations, but there were also other types of policies which led to community and linguistic disruption. Such practices and policies are well documented throughout the literature on Australian languages and Aboriginal 20th -century society (Australian Human Rights Commission 1997, for example). Ford (1990) describes how Wadjiginy (Bachamal-speaking) people were removed from their traditional lands to a community where they were a linguistic minority. Donaldson (1980) describes a similar situation for Wangaaybuwan (Ngiyambaa). The move away from traditional lands was highly significant for numerous groups, including Yir-Yoront (Alpher 1991: 2–3) and Bardi (Bowern 2008c; 2012a). Beckett (1978) documents the forcible deportation of Aboriginal people from the Tibooburra area in 1938. These are not isolated stories. While these might be some of the most consequential issues for the 20th century, let us not forget that one of the biggest issues for language transmission in the 19th century was genocide. It may seem idle to discuss language in such a context, but I discuss genocide here as a way for readers of this volume to think of the lasting impact that these policies had. Just to start, the Aboriginal population of Australia had been reduced by 90% by 1920, in the course of just over 100 years. That is a massive social tragedy and caused lasting social upheaval. While most of the 20th century has seen extensive language loss in Australia, new languages have also arisen based on contact varieties, including varieties of English (see Chapters 56, 57, 58, and 66, this volume), and from language reclamation (e.g. palawa kani; see Chapter 78, this volume). The front matter to this volume, in addition to including classification of traditional languages, also includes a list of new languages; this list is probably incomplete, however, since it only includes languages documented in publications. For more information on this topic, the recently released National Indigenous Languages Report (Department of Infrastructure, Transport, Regional Development, and Communications (DoITRDC), Australian Institute for Aboriginal and Torres Strait Islander Studies (AIATSIS), and Australian National University (ANU) 2020) provides detailed information on the current usage patterns of Aboriginal and Torres Strait Islander languages. They note, for example, that there are 31 reclamation programmes currently underway.

7.7 Conclusion In conclusion, the figure of ‘250’ languages should be retired. If there is need to refer to a number of languages counted on linguistic grounds, either 440 or 490 is an appropriate number to more accurately reflect the linguistic record.

chapter 8

Philological methods for Australian languages John Giacon and Harold Koch

8.1 What is philology? We use the term ‘philology’ to refer to the description of a language from historical sources, generally text and audiotapes and very rarely film.1 The term ‘philology’ has traditionally been used to describe the academic discipline of describing ancient cultures through the analysis of their (primarily literary) written documents. A more specialized usage refers to the description of the language of the documents.2 For discussion of the terms ‘philology’ vs. ‘linguistics’ see Koerner (1997). Of more relevance to Australianists is Goddard (1973), which emphasizes and illustrates the application of the philological approach to early records of Amerindian languages. As Amery (1998: 112) points out, in the study of Australian languages philology generally refers to the study of these materials mainly, in the words of Goddard (1973: 727), ‘in order to gain information about the languages in which these records are cast’. Although philological methods may be applied to just one document, typically the main purpose of Australian philology is to arrive at a description of the language that is as complete as possible. The results have been called ‘consolidated accounts’ of individual languages, using all the available old sources plus any available later ones, including taped material. Examples are Amery (1998; 2000), Blake (2003b), Lissarrague (2010), Morelli (2015) and Giacon (2017). Philology, in our sense, has the same aim as any language description, but contrasts with descriptions based on currently spoken languages in its scope and sources. When a language is no longer used, this is the only description available. When a language is still actively spoken, philology can be used to compare previous and current versions of the language. As with other language descriptions, philological descriptions can develop over time, with later work building on 1 See Thieberger (Chapter 3, this volume) for more on documentary sources. 2 Furthermore, the term ‘comparative philology’ is used, especially by the British, for the sub-discipline of historical linguistics.

earlier work and taking advantage of increased linguistic knowledge, both generally and of languages closely related to the language being studied. The first step in this philology is phonemic reconstruction of word forms. In many instances, the nature of the sources means that this is the major task of philology. It is also the task that is most different from other methods of language description, those which have access to speakers to determine word forms. Other parts of philologically-based language description will have much more in common with language descriptions derived from current speakers. While word forms are the first step, the aim of philological language description is to describe the language as fully as possible. This description will be in terms of the current orthography of the language and current conventions for the descriptions of other aspects of Australian languages. In many, if not most, instances, the motivation for philological language description is language revival—the desire, especially by descendants of speakers, to again use the ancestral language.3 This purpose gives philology a different emphasis from purely academic language research (which may be motivated in part by considerations of linguistic typology or language change, for example). Amery (1998: 113) points out that, for a time, historical work on Australian languages was ignored, while description was based exclusively on data recorded directly from modern speakers.4 This was especially true in the period from the 1960s through the 1980s, when the urgent task was seen as documenting what remained of spoken languages before it was too late. Some examples of such descriptive work which did not attempt to integrate older written sources are Hercus (1969), Eades (1976), Donaldson (1980) and Austin (1981a). For many languages, especially in longer colonized areas, early written and (sometimes) sound records are the only substantial sources for the language. 3 For language revitalization, see Amery and Gale (Chapter 65, this volume). 4 Cf. Stockigt (Chapter 2, this volume) and Thieberger (Chapter 3, this volume).

John Giacon and Harold Koch, Philological methods for Australian languages. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © John Giacon and Harold Koch (2023). DOI: 10.1093/oso/9780198824978.003.0008

65

john giacon and harold koch Even in areas where there are current speakers, their knowledge of the language is often far from complete; hence the evidence of earlier records may enable a more complete description. The information held in historical documents is important for an understanding of a fuller picture of Australian languages, including: the assignment of languages to Indigenous local groups, the description of individual languages, the recovery of heritage languages by contemporary Indigenous groups, studies in the typology and areal distribution of linguistic features and the genetic classification of Australian language. Since the main purpose of much recent philological work is to assist in the process of language revival, it is important to have as complete a description of the language as possible. For example, Besold (2012: 73), says of her work on languages of the NSW south coast: The primary aim of this study has therefore been to provide detailed language descriptions that will allow the languages to be taught to a higher proficiency level than had previously been possible.

Katrina Power, a Kaurna woman, talking of neologisms/language engineering in revived Kaurna, says: ‘I still want to know that the guts of these originated from here (from the traditional Kaurna)’ (Amery 1998: 233). To fulfil the aim of complete description, philological methods should be applied to phonological, lexical, morphological and syntactic information, and may also cover other areas such as narrative and discourse patterns. As Amery (1998: 112) notes, the scope of the description is dependent on the information in the sources, but it also depends on the knowledge and skill of the describer and the time they have for the project. Recovering the relevant facts from historic documents requires different methods from documenting languages by means of fieldwork with fluent native speakers—methods that we characterize as ‘philological’. Here we discuss some of the issues involved in using such old materials and offer guidance in their analysis (see also Thieberger 1995; Amery 2000; Besold 2012).

8.2 Sources 8.2.1 Types of sources Historical documents, unlike sources for most living languages, are finite, and limited in both quantity and the type of language they record. Actual sources vary enormously (see Besold 2012: xxiii–xxvi; Thieberger, Chapter 3, this volume) as to whether they are published or not, how

66

accessible they are, degree of comprehensiveness, and quality. However in general they are quite short, as is clearly seen by that fact that Teichelmann and Schu¨rmann’s (1840) grammar of Kaurna, at 24 pages, is one of the longer sources. The range of genres and sociolinguistic usages recorded is very limited. Janet Mathews’ Yuwaalaraay tapes have many elicitations involving traditional culture (hunting, camping etc.) and grammatical paradigms (e.g. pronouns), as well as translations of traditional stories and pronunciation of words in old wordlists. She does have some connected text, as does Threlkeld (1834), whose Awabakal has lengthy biblical translations. Older texts, however, are generally lacking in children’s language, conversational material, talk between men and women, not to mention scatological terms. Nor are these found in Gu¨nther (1892) and Hale (1846), the main sources for Wiradjuri, or Threlkeld’s (1834) Awabakal. Some early writers emphasize and others avoid certain topics such as explicitly sexual material. The simplest documents are vocabularies, often referred to as wordlists; few early vocabularies were comprehensive enough to merit being called dictionaries. Some wordlists (e.g. Teichelmann and Schu¨rmann’s 1840 72-page ‘Vocabulary’) contain numerous example sentences, which add to semantic knowledge and from which grammatical information can be obtained. Two major collections of wordlists are by E. M. Curr (1886–1887)—up to 120 words from a standard questionnaire in nearly 500 communalects from most parts of Australia—and R. Brough Smyth (1878) from languages of Victoria. Wordlists necessarily reveal only a fraction of the original information about an Indigenous word. And at times it is difficult or impossible to know even the primary meaning of the word. For instance, Ridley (1875: 22) has three words glossed ‘parrot’. These presumably refer to individual species of parrot, but it is impossible to know which species.5 Early written documents include grammatical descriptions; this is especially true of early German missionaries working in southern and central Australia (Stockigt 2017). The major author of early sketch grammars from around 1900 is the surveyor R. H. Mathews, who published 57 grammatical descriptions, mostly of languages in New South Wales and Victoria (see Koch 2008). While these provide useful information—sometimes the only available grammatical data on the language—they were typically cast in the framework of Traditional Grammar, which is no longer widely understood.6 They tend to focus on inflectional morphology, and cover a rather narrow range of grammatical topics 5 It is very unlikely that any of these was a general term, since these are fairly rare in Yuwaalaraay-Gamilaraay (YG). No general terms have been recorded for kangaroo, lizard, animal, plant, or fruit. The only YG general terms referring to birds are dhigaraa ‘bird’ and garrangay ‘duck’. 6 For a concise overview see Bernard (1975).

philological methods for australian languages (see further Section 8.3.5). Some authors (e.g. Threlkeld 1834) have included a list of sentences which illustrate the grammar—largely leaving it up to readers to do their own linguistic analysis. Historical documents include texts, i.e. traditional or everyday stories provided with a free English translation. Another kind of connected text is translations especially of parts of the Bible. (See Section 8.3.6 on interpreting texts in historic documents.) Some preliminary questions should be considered when approaching historical documents for a given language. Which documents are relevant to the description of the language in question? For any particular document, we should ask concerning the author: Who recorded the data? In what year? In which place and in what circumstances? For what purpose? What was their background relevant to this task? Concerning their source we should know: Who supplied the data? What did they call the language? What named social group did they attribute the language to? Concerning the document: Is the material recorded raw language data or analysis? On the basis of such metadata and the internal analysis of the data, we then need to relate the given language sample to languages that are recognized in modern linguistic study.7

8.2.2 Limitations of records and sources of errors Careful study of early documents may bring to light errors in the data. Mention should be made of mechanical errors of the type signalled in Section 8.3.3 below. Other errors are introduced when materials are copied or interpreted. In the source documents themselves misinterpretations may be due to the recorder, the consultant, or the interaction between the two. (More examples of errors are given below in Section 8.3.) Some early recorders of Australian languages were welleducated missionaries, whose prior studies included languages such as Hebrew, Greek and Latin—common enough at the time and particularly relevant to Biblical studies (Amery 1998: 173). And some had previous Australian grammars (e.g. Threlkeld 1834) which they could use as a template (Amery 1998: 145; Stockigt 2017: passim).8 Other recorders, including many of Curr’s (1886–1887) correspondents, had more limited education. Others were educated, but not in linguistics; prominent among these 7 Koch (2011b) discusses these issues with respect to the linguistic materials of G. A. Robinson. 8 In Australia, however, there was a much weaker tradition of Indigenous language documentation than in North America, for example (for which see Hoijer 1973).

was Janet Mathews, a musician, who made over 100 taperecordings of NSW languages in the 1970s, using as elicitation material wordlists, stories and features of the interview situation. Even linguistically educated recorders, however, had substantial difficulties accurately describing Australian languages. The first step in language analysis is generally to fit the language into a system one is familiar with, and so, for those with a background in the ‘classical’ languages, Australian languages presented problems (see especially Stockigt 2017 and Section 8.3.5 below). The time linguists had to spend with consultants also limited their knowledge of the language and their analysis. Teichelmann and Schu¨rmann (1840: v) admit that their published grammar was based on only eighteen months’ experience. The surveyor R. H. Mathews could not have spent a lot of time acquiring the data on each of the 57 languages he described. Even more recently, Williams’s (1980) grammar of Yuwaalaraay, as an Honours sub-thesis, was based on less than a year of work. Another source of error, which is not limited to early descriptions, has been the tendency for Australianists to unjustifiably assume that the language of their description conformed to patterns established for other Australian languages (see Bowern 2017b). In some records very little information is provided about consultants—in some instances only their name, sometimes not even that. However, tape-recorded speakers can actually be heard, and so much more is known about them, and many differences between them are clear. One assumes that some of those consulted for written documents also had similar qualities. Consultants vary in their approach to the task. For Yuwaalaraay, Fred Reece, in around 30 tapes, says something like ‘There is a way of saying that’ over 300 times, happy enough to acknowledge the limits of his knowledge. Arthur Dodd, on the other hand, very rarely uses similar words but rather tries to come up with a paraphrase or approximate translation, however much he struggles with it. Consultants vary in their knowledge of the language. Their degree of fluency may not be readily noticed on a first hearing, and it may be only after learning more about the language, perhaps by hearing other speakers, that the hearer is in a better position to judge a speaker’s fluency. While it is clear that some languages were recorded when they were still strong, and so presumably from fluent speakers (e.g. Gu¨nther’s (1892) Wiradjuri, Threlkeld’s (1834) ‘Awabakal’), language knowledge can be lost very quickly, and so it cannot always be assumed that speakers were fluent. The interaction between the recorder and language speaker can sometimes be the source of errors of interpretation. Speakers may or may not correct the interpretations of recorders or their own mistakes. Arthur Dodd, for instance,

67

john giacon and harold koch occasionally slips into Wayilwan in his Yuwaalaraay elicitation (and vice versa). At times, but not always, he explicitly corrects himself. Other mistakes are less obvious. The Yuwaalaraay verb root element -ma-li, for instance in the verb dhama-li ‘feel’, can indicate ‘action done with hand’. The element -dha-li, second in dhadha-li ‘taste’, indicates ‘do with mouth’. When Arthur Dodd is asked to translate ‘the dog stole the meat’, he initially uses manuma-li, but then changes to banagaali-y ‘run away with’, which suggests that he realized that manuma-li was the appropriate word for reference to a person, but not for a dog’s action, which is ‘do with mouth’.9 It can be assumed that earlier YG speakers would have used manudha-li (Giacon 2017: 344). ‘Gratuitous concurrence’ by the consultant—explicit or implicit agreement without really assenting to the recorder’s question10 —can be the source of misrepresentation in the data, as in this exchange between Corinne Williams (CW) and Arthur Dodd (AD), where AD corrects CW’s first pronunciation but agrees the second time she errs. (1) CW: elbow? AD: ngunuuga CW: ngunuga AD: ngunuuga CW: ngunuga AD: mm Sometimes there is a difference between the meaning of the elicitation sentence and the translation given. Wurm (1955: 13) has a sentence normalized as minyagu-nda yanawaanh’ translated as ‘I have come to talk’, whereas the Yuwaalaraay actually means ‘What are you coming for?’

8.3 Practical steps in philological analysis Besold (2012, vol. 1: 75) [Part A] points out that Although there has been an increasing corpus of literature on the early collectors and their language collections, very little has been published on methodology and approach of working from this material in language recovery.

She then describes some of the processes she used. Below we consider these and some other practical approaches that facilitate the interpretive work of philology.11 9

The verbs manuma-li and manudha-li are both found in Wangaaybuwan. See Eades (2013: 221 and passim) for this term. 11 Our relevant experience derives mainly from Giacon’s grammatical work on Yuwaalaraay and Gamilaraay (YG) (see Giacon 2017), based on copious written and taped sources, and Koch’s work on interpreting the 10

68

8.3.1 Background knowledge The importance of building on existing knowledge cannot be over-emphasized. The philologist needs to have as broad as possible a knowledge of the current understanding of Australian languages, and in particular of widely found features of languages related to the one they are studying. As well as taking advantage of the accumulated knowledge, it allows their work to be easily read by those familiar with the field. Some common features of Australian languages are: relatively similar phoneme inventories; agglutinative morphology; case systems which include ergative and local cases; comitative (or proprietive) and other suffixes which may or may not be analysed as case; dual pronouns and many more. There are many less common or less fully described features such as middle/intransitivized verbs, inclusory constructions and alternative case suffixes on proper nouns.

8.3.2 General processes There are a number of steps common to most if not all philological work. Many of them are also common in producing any grammatical description. The first step is to collect as many of the sources as possible. There may be questions of access, and some sources may not be well known. The next step is to interpret the sources and make the information in them readily searchable. If the language does not have an established orthography one will need to be developed. At times, established orthographies can be adapted. When substantial research into YG restarted in the 1990s, the major recent Gamilaraay documents were Austin (1993a, b) and the web dictionary (Austin and Nathan 1996). These used voiceless stops. The main Yuwaalaraay source, Williams (1980), used voiced stops and a number of diacritics. Those then working on these languages decided to use digraphs rather than diacritics, since they are easier to type. They also decided to use voiced stops because these were more common on tapes and in order to have consistency across Yuwaalaraay and Gamilaraay. This made it easier for resource materials (dictionary and grammar) to cover both languages. There is very little information about Gamilaraay grammar, and what there is indicates it is very similar to Yuwaalaraay. That is, they decided to treat the grammar of both languages as identical except for the very few recorded differences. They also decided to ignore the dialects within Gamilaraay even though a number of dialect names have been recorded. Even where wordlists of the Yuin languages (cf. Koch 2011b; 2016) and kinship terms for the Austkin project (http://www.austkin.net/).

philological methods for australian languages dialects are known, standardization makes language revival much more feasible (cf. Gale, Chapter 64, this volume). Transcribing documents and tapes into one orthography enables them to be much more easily searched, but also necessitated a number of decisions. The first was whether to use a phonemic or phonetic transcription. A phonetic transcription is very time consuming and not easily searchable, so phonemic is generally preferable, whether of tapes or existing texts. For instance Wurm (1955) has for ‘man’ maṙi, máṙì, ˈmai, ˈmaṙi, mari, ˈmáṙì, maṙṙi, maɽi. By far the most common is maṙi, and maɽi represents a change in Wurm’s orthography. In standard orthography these are all mari. Giacon’s transcription of Wurm has all of his forms written as mari, thus making it easy to find and compare all instances. Williams had dhiyamali ‘pick up’, and this word is clearly heard on the tapes. This makes it possible to interpret what could otherwise be a very confusing set of forms in Wurm, given in Table 8.1 with Wurm’s gloss and the current orthographic representation and analysis. In these instances, with a common word, recorded by a linguist, with one main form (for mari) and tape information (for dhiyamali), it was relatively easy to decide what the standard form would be. In many instances, however, this was not the case. If there is already an established wordlist/dictionary, this can form the basis of transcription, although at times the philologist may make changes. For instance, most of the Yuwaalaraay words in the current Gamilaraay Yuwaalaraay Table 8.1 Wurm’s variant verb forms. Wurm’s forms

Wurm’s gloss

Interpretation

ðiεmari, ðijεmalli, ðiεmalli

collect

dhiyama-li Future

djamali, ðijamalli

will help

dhiyama-li Future

djami, jami, ði:mai

picks up, brought

dhiyama-y Past

ði:mala

Go up!; Take it out!

dhiyama-la Imperative

Ðiamala

take (an axe)

dhiyama-la Imperative

ðiεmala

pick it up; call (the doctor)

dhiyama-la Imperative

dja mala:βui/dja mala:βai,

pick.up-allPast

dhiyama-l. aaba-y

Yuwaalayaay Dictionary (Ash, Giacon and Lissarrague 2003) are as found in Williams (1980). However some changes were made, such as: Williams’s yinaay to Ash et al.’s yanaay ‘go’— after listening to the tapes and considering the cognate forms in Gamilaraay and other related languages. Transcribing phonetic variants makes searches for syntactic and morphological information more difficult (regardless of their value for other purposes); for instance, the Wurm transcriptions above, or the Yuwaalaraay demonstrative nhama. In a slightly detailed phonetic transcription it would be recorded as nhama, nham’, nyama, nyam, nama and nam’. With programs such as ELAN (https://tla.mpi.nl/ tools/tla-tools/elan/) it is possible to have the full phonemic transcription in one tier, a narrower transcription in another, and notes (for instance, speculation about the conditioning factors) in a third. The notes tier can also be used for other comments, or a further tier may be set up. Where there are only a few sources it may be simpler to search them individually. Where there are multiple sources a different method is needed. Some programs (e.g. Text Wrangler) can search a set of documents simultaneously. Sources can also be combined. Giacon created a single document ‘alltapes’ with all the tape transcriptions, but also documents that had only the tapes of a single speaker, so that differences between their versions of the language were more easily found. (He also transcribed all the major written sources and filed them separately.)

8.3.3 Issues with wordlists The documentation of vocabulary from wordlists might appear to be straightforward. In practice, mistakes in the meaning of words can be found, especially with recorders who did not have much contact with the language. The following misidentifications are mentioned in Koch’s (2011b: 157) study of G. A. Robinson’s language materials from New South Wales. In some wordlists, the words for ‘I’ and ‘you’ (normalized as ngayamba and yindiki) have been reversed, and once the word for ‘my’, ngayamba-dyanu was given as ‘wife’. Dyinang was given, with slightly different spellings, for both ‘foot’ and ‘toes’; the same form dhawang underlies three glosses, ‘stomach’, ‘belly full’, and ‘eat too much’; and kunang was translated as both the noun ‘excrement’ and the verb ‘evacuate’. One can almost imagine the kind of interaction which led to a term otherwise recorded as ‘rainbow’ being glossed as ‘milky way’. On the other hand, when dharra is given for both ‘thigh’ and ‘creek’, this appears to be genuine polysemy (or extension of a body part term), since it is attested in other sources for related languages. It is worth mentioning here more mechanical causes of misinformation. Handwriting of manuscripts may be

69

john giacon and harold koch misread, either by the linguist or by an earlier copyist or typesetter. A slippage has obviously occurred in the Ngarigu wordlist by Bulmer in Curr (1886–1887), which came to light only by comparison of forms (and not just meanings) across wordlists from the same and closely related languages. Unfortunately 15 of the items (from ‘canoe’ to ‘wind’) are matched with what should be the gloss of the next word on the list. Thus ‘canoe’ should rather be glossed ‘sun’, the next word in the list, which other sources establish as /mamady/. (Koch 2016: 146)

Wordlists can sometimes be useful for supplying a few details about morphology. From Robinson’s wordlists for languages of the Yuin subgroup it is possible to discern some suffixes that mark possessor (‘my’ or ‘his’) in variant forms of bodypart or kin nouns; e.g. from the Thawa language— his Twofold Bay (TB) and Cape Howe (CH) vocabularies—he gives for ‘brother’ TB marn.di.je and CH marn.jeen.gong, which may represent /mandyi-dyi/ ‘my brother’ and /mandyingung/ ‘his brother’ respectively (Koch 2011b: 158). Verbs given in wordlists typically include some inflections, which may mark purposive (‘to VERB’), imperative, present, or past tense. A careful comparison with information given in sketch grammars or sentences in texts may in some cases allow inference about their functions.12

8.3.4 Phonology and the interpretation of spelling The first problem encountered with old documents is interpreting the spelling. Many early recorders spelled Indigenous words using the conventions of English (or German or French if that was their native language). Thus their munna may reflect /mana/, where the vowel quality is indirectly signalled by the following doubled consonant according to English spelling rules, whereas moona or muna may be their representation of /muna/. If the recorder was a speaker of German, French, or a northern variety of British English, this problem with orthographic u will not apply. Where a long vowel is required, early recorders might spell /ma:na/ as mahna or possibly marna. Use of r before a consonant or word-finally may signal vowel length rather than retroflexion, as in modern Australian orthographies. Single vs. double r between vowels is typically used to differentiate vowels as in English furry vs. fury, and is unlikely to reflect modern Australianist orthographies, which use rr for a tap/trill and r for an approximant. 12 R. H. Mathews has provided wordlists for many languages, including ones for which he wrote sketch grammars, but gives no indication of what inflected form the verbs represent.

70

Phonemic differences in Indigenous languages are often not indicated, especially that between retroflex, apicoalveolar and dental consonants (e.g. /mana/, /marna/, /manha/), the different rhotic consonants (e.g. /mara/ vs. /marra/) and long vs. short vowels (e.g. /maana/ vs. /mana/. An initial velar nasal [ŋ] is often either not recorded at all or is represented by another nasal, usually n but sometimes m before a rounded vowel. (Variation between m and n can therefore be a useful clue for velar articulation.) In contrast to this under-differentiation of phonemes, over-differentiation typically results from the recording of phonetic but non-phonemic differences between voiced and voiceless stops (p vs. b, etc.) or high and mid vowels (e vs. i, o vs. u). A mid vowel which is an allophone of /a/, e.g. [e] after a palatal consonant or [o] after /w/, is typically represented simply as a mid vowel. These over-differentiations result from the fact that the recorders have notated the phonetic form of the words, having had no access to their phonemic representation. The same word is often written in different ways by different early recorders, and sometimes even by the same recorder. Any of the following factors (summarized from Koch 2009: 129) may be responsible for this variability in spelling. • variable pronunciation of the phonemes of the Indigenous language (owing to free variation, different phonological conditioning, careful vs. casual articulation, etc.) • unfamiliar phonemes of the Indigenous language, which are variably perceived by European recorders (e.g. dental, retroflex or palatal consonants, trilled rhotic) • unfamiliar sequences of sounds of the Indigenous language, which are variably interpreted by European recorders (e.g. word-initial [ŋ], sequences like /nk/, /uy/) • variable degrees of attention paid to phonetic details by European recorders (especially the nature of vowels in unstressed syllables) • differential use of the English spelling system to represent the same sound (e.g. ch, tj, j, g, dg for a palatal stop) • different dialects or languages of the European recorders (e.g. whether English speakers pronounce post-vocalic r or use the vowel of put in butter) The interpretation of spelling of words in old written sources has been characterized as ‘reconstitution’ (sometimes as ‘reconstruction’). But what are we reconstituting? Given that the recorders only aimed to reproduce the phonetics of the word, we should in the first instance infer the phonetics of each word. But linguists are aware that the uttered phonetic form may vary between different utterances

philological methods for australian languages Table 8.2 Phonological reconstitution process. PRODUCTION

/S/ >

[S] >

[R] >

RECONSTITUTION < */S/ < *[S] < *[R] < (because of free variation between voiced and voiceless stops, mid and high vowels, or as a result of casual vs. careful pronunciation). Hence we are really after the phonemic representation in the mind of the speaker. Then we want to express this using a consistent orthography, presumably one that follows the modern orthographic conventions for the Australian languages. The input to this reconstitution process is not straightforward, however, but is mediated by the recorder’s orthographic representation of what they thought they heard (and they may have missed details of the pronunciation). The linguist doing the reconstitution must first infer from the recorder’s spelling of the word what the recorder considered to be the phonetic form of the word. The reconstitution process thus consists of a series of inferences that undo the processes which produced the written record of the spoken form. Table 8.2 diagrams the processes of production (by the speaker S and the recorder R) and reconstitution (by the linguist L). The production begins with the speaker’s phonemic representation /S/, uttered as the speaker’s phonetic representation [S], which in turn is perceived by the recorder as a (possibly different) phonetic representation [R] and recorded orthographically by the recorder as . This recorder’s orthographic representation is the input to the linguist’s reconstitution process, which begins with inferring the recorder’s phonetic representation *[R] (asterisk indicates hypothetical forms), then the assumed speaker’s phonetic representation *[S] and finally the assumed speaker’s phonemic representation */S/. This is then represented orthographically by the linguist as . The typical description of phonological reconstitution as simply deriving from masks the complexity of the procedure. A hypothetical example would be if a linguist were to take a word spelled with final ang and believe there is reason to assume the speaker intended /aɲ/, which they produced as [æɲ] but which the recorder heard as [æŋ] and spelled as ; the linguist reconstitutes /aɲ/ and spells it as according to the convention that palatal nasal is spelled ny. The requirements for doing phonological reconstitution are knowledge of 1) the orthographic conventions of the recorder’s language 2) the phonetics and phonology of the recorder’s language

3) the likely phonetics and phonology of the Indigenous language 4) other orthographic representations of the same word in the same language 5) the phonemic representation of cognate words in related languages

Point (3) highlights the role of typology. There are great similarities in the phonology of many Australian languages, which provides the linguist with a rough idea of what the phonology of the words under consideration is likely to be.13 Point (4) emphasizes the great usefulness of comparing different spellings of what is judged to be the same word, in the hope that one spelling will have captured phonetic details that have been missed in another rendition. See Austin and Crowley (1995) for applications of this technique. Another reason for comparing different spellings follows from the ambiguity of English spelling. More than one phonetic representation may be inferred from a given spelling. Comparing different renditions helps us to decide between alternatives. Point (5) is relevant since many words recur in several languages, especially in languages genetically or geographically close to the one under consideration. Representations of the phonologically identical word in other languages typically provide more orthographical variants that can be compared. It is especially useful if the word being reconstituted has been heard and accurately recorded by a modern linguist. Here are a few examples. For the Ngarigu language of the Monaro we have for ‘I/me’ (sometimes wrongly given as ‘you’) the early spellings ime.bar, imeba, ngimba, ngiamba, niamba, ngaimba. The end of the word seems to be [ba], with ar once indicating the low vowel quality. Word-initially the records give ng, n and no consonant—a typical reaction to initial [ŋ]. The dot in the middle of ime.bar tells us that this recorder heard two syllables, the first being [aim]. The last version supports this perception of a diphthong [ai], so in the third version the ngim is presumably to be interpreted as [ŋaim]. But three syllables are implied by ngiamba and niamba, with the third syllable having a vowel [a]. The forms can be reconciled if the ia is interpreted as [aja]. Then we conclude that the middle syllable was not heard in the disyllabic representation. Knowledge of Australian phonotactics, however, leads us to avoid positing diphthongs like [ai] and to rather expect vowel glide vowel sequences like [aja]. So the phonemic reconstitution is /ŋajampa/, which would be

13 Gasser and Bowern (2014) however find that these similarities have sometimes been overstated.

71

john giacon and harold koch Table 8.3 Comparison of forms of ‘I’ in Ngarigu. Gloss

Source

i

me

b

ar

you, me

Robinson

i

me

b

a

you

Curr

ng

i

m

b

a

I

du Vé

ng

i

a

m

b

a

I

Bulmer

n

i

a

m

b

a

I

Howitt

ng

a

i

m

b

a

thy

Howitt

ŋ

a

j

a

m

p

a

Reconstructed Phonemic

ng

a

y

a

m

p

a

Reconstructed Orthographic

spelled in an orthography that uses for /ŋ/, for /j/ and for the labial stop, which does not distinguish voicing but is typically voiced after a nasal consonant. The compared forms are displayed formally in Table 8.3, arranged in the style of Austin and Crowley (1995); source details are from Koch (2011a: 139). The ambiguity of English spellings and the role of using cognates can be illustrated from the word for ‘yamstick’, which is widespread in languages of New South Wales. A spelling kunni could be interpreted as representing several pronunciations: [kɐni], [kɐnai], [kʊni], [kʊnai], depending on whether u is a rounded vowel or not and whether i is a plain vowel or a diphthong. The double n might further be interpreted as representing a dental [ṋ]. The final [ai] could be interpreted phonemically as /aj/, /aji/, or even /a:ji/, with a long vowel. All of these interpretations have in fact been made by linguists. Granted further that either k or g could be used to represent the velar stop, and spelling /ṋ/ as , /j/ as and /a:/ as , all of the following have been used to spell this ‘yamstick’ word in various NSW languages: guni, ganay, kanay, ganaay, ganayi and ganhayi. All of them probably represent the same phonemic form /kanaj/. Such a plethora of different reconstitutions could have been avoided if more attention had been paid to cognates in other languages. This example highlights the inadequacy of basing language reconstitution narrowly on the recorded spelling variants of a single language. The interpretation of glides provides particular problems. While English has diphthongs such as /au/ and /ai/, Australian languages typically lack diphthongs as unit phonemes, as well as sequences of two vowels in adjacent syllables. However, they frequently use /j/ and /w/ as consonants, of which both occur inter-vocalically and /j/ occurs word-finally in some languages, but not in consonant

72

clusters such as /kw/ or /pj/.14 Further difficulties are provided by the fact that palatal consonants abound, with audible on-glides, in sequences such as /aɲ/, which is likely to be perceived as [ain]. We suggest that the following principles could be applied in interpreting the phonology of apparent VV orthographic sequences. 1) It is useful in the first instance to notate all glides as if they were vowels, and then interpret the vowel sequences in the manner that best reflects the assumed Indigenous phonology. 2) Treat the high front glide [j] of apparent diphthongs as a palatal consonant, either consonantal /j/ or part of a palatal consonant /c/, /ɲ/ or /λ/. Thus spellings like bing-ngai and binghi yield a phonetic reconstitution [biŋai], interpreted phonemically as /piŋaj/ and spelled . A spelling womboine yields phonetic [wamboin], interpreted as phonemic /wampuɲ/ and spelled . 3) Interpret any sequence of three (or more) phonetic vocoids as a phonemic sequence Vowel Glide Vowel (VGV). Thus orthographic byoo-wan15 and buon are assumed to represent phonetic [biuan], which is the realization of /piwan/. (Phonemic /iwa/ is liable to be realized as—or at least sound to English ears like— [iu.a].) 4) Use the shortest possible phonemic sequence compatible with the presumed phonotactics of the language. Thus a representation koo-ee-wun, reflecting phonetic [kuiuan], is most compactly interpreted as /kujwan/, orthographically . 5) Remember that a word which is ambiguous from the spellings in one language may become clear when compared to equivalents in adjacent languages. Thus guir which appears in Curr’s wordlist from Yass for ‘fish’ may suggest an interpretation as /kuwirr/. Bearing in mind, however, the possibility that final r does not represent a rhotic in the spelling of some English speakers, the intended phonetics may be not [guir] but [gwia]. When the vowels and glides are all represented as vocoids, a pronunciation [guia] results, which is plausibly interpreted phonemically as /kuja/. Now in view of the fact that in many eastern languages /kuja/ is the word for fish, a kind of fish, or eel, the source spelling is most reasonably taken as an idiosyncratic spelling of the familiar Australian fish word . 14 These generalizations are subject to the proviso that not all Australian languages necessarily have the same phonotactics. 15 Hyphens or full stops are used in some early sources to separate what are perceived as separate syllables.

philological methods for australian languages 6) In presenting the results of phonological reconstitution, best practice requires the presentation of all the source spellings on which the reconstitution is based. This gives other linguists a basis for any alternative interpretation they may consider. There are a number of reasons why it is worth putting considerable effort into getting the right phonological interpretation. First is respect for the traditional speakers of the language. Second, our scholarship should be as authentic as possible. Third, it provides assistance to other linguists engaged in a similar task for different but related languages. Fourth, it makes the data more usable for comparative purposes. For instance, if differently reconstituted instances of cognate words are entered into comparative databases, the identity of cognates such as /kuya/ is obscured if the word is represented variably as , and .

8.3.5 Interpreting grammar For many languages, the only available source for understanding the grammar is a description written before the advent of modern methods of linguistic analysis. Early descriptions typically use the framework of Traditional Grammar, which was based on methods of description used by grammarians of ancient Greece, Rome and medieval Europe.16 This framework has provided the basis for most pedagogical grammars written by European scholars up until the twentieth century. Unfortunately, the principles of such grammars are now not understood by many people, even students of linguistics. Anyone trying to interpret older descriptions of Indigenous languages would do well to read a work such as Bernard (1975). Three salient characteristics of descriptions according to the Traditional Grammar model can be mentioned. In the first place, the order tends to be according to part of speech (word class), and within each word class, according to inflectional categories. Thus they may begin with ‘Articles’, even though there are no articles in the language, then continue with the inflectional categories of nouns—mentioning Gender, even if there is no Gender, then Number and Case—and eventually discussing ‘Prepositions’, even if there are none in the language. The second characteristic is that word classes, and also inflectional categories, tend to be ‘notional’ rather than ‘structural’; i.e. they are based on the translational equivalents of English words and categories rather than being justified by contrasts within the system of the language. A third feature of traditional grammatical descriptions is that they are word-based: while paradigms of inflected words may be given, there is 16 See Koch (2008) for further elaboration of this problem, with reference to the grammatical sketches by R. H. Mathews.

no attempt to isolate the formatives that mark particular grammatical values. Without an understanding of these basic characteristics of the model used by early descriptions, the modern linguist may be puzzled by the facts and wording presented in older grammars. Here are some examples of possible confusion. For a language without nominal genders, indicated by the agreement of modifiers, we may encounter the claim that there are masculine and feminine genders, as demonstrated by separate words for ‘man’, ‘woman’, ‘male possum’ and ‘female possum’. Regarding case (cf. Dunn and Meakins, Chapter 20, this volume; Simpson, Chapter 21, this volume), it is common to read that there are two ‘nominative’ cases, one of which occurs only as the subject of transitive verbs; in modern linguistics, of course, we treat this as a separately named ergative case. The locative case is typically omitted, since the European languages known to the describers do not have a distinctive locative case. On the other hand, the term ‘ablative’ might be used to describe a locative or instrumental case, or even the agent, since the Latin ‘ablative’ had these functions. A claim may be made that prepositional meanings are expressed by verbs, citing verbs with meanings of ‘go up’ vs. ‘go down’. Finally a person-number paradigm may be given for verbs, with ‘I’, ‘you’, ‘he/she’ accompanied by a verb form which shows no distinction according to subject person—because European languages known to the author typically do inflect for their subject. Early analysts did not recognize other features now standard in the description of Australian languages. They do not list verb classes, which are common in many languages, even though their verb paradigms show differences in inflection and classical languages have verb conjugations (see Koch, Chapter 27, this volume). Australian languages are generally agglutinative in their morphological structure. Description of morphemes is central to most modern grammars, but is rarely if ever found in older sources. Ridley’s Gurre Kamilaroi (1856a) has ŋummildone, kakuldone (now interpreted as ngami-lda-nhi ‘see-cts-past’; gaga-lda-nhi ‘call-cts-past’). The morphemes -lda-nhi are very common, but Ridley did not recognize them. He did not have the models to base this sort of analysis on.

8.3.6 Interpreting texts While other types of philology often study narrative texts or other kinds of literature, in Australian philology much of the material being studied is explicitly linguistic, i.e. wordlists, grammatical analysis and paradigms (Amery 1998: 87). Another source of information is newly composed material, often biblical translations, whose language is neither natural

73

john giacon and harold koch nor fluent, and likely to be largely influenced by the (nonspeaker) author; for instance Ridley’s (1856a): short translations of Gospel passages. These are of limited value since it seems very clear that they consist of very simplified and almost certainly inaccurate Gamilaraay—possibly written by Ridley or Greenway on the basis of very limited knowledge. The absence of ergative suffixes is just one indication of this. Where original, fluent materials are available, they often provide a much richer source of language knowledge than wordlists, attempts at grammatical analysis, or texts composed by those with limited knowledge of the language. Especially for languages which do not currently have speakers, these materials, whether written or sound recordings (Goddard 1973: 727) can be a major source for information of the language. An accurate interpretation of the texts or tapes is the first step in any linguistic study. Examples of fluent textual materials include the Kaurna sentences in Teichelmann and Schu¨rmann (1840), the Yuwaalaraay tapes used by Giacon (2017), and the Gumbaynggirr narratives in Morelli et al. (2017),17 which includes important new grammatical information gleaned from the texts. Other examples include the Dharrawal and Dharumba mythological stories recorded by Andrew Mackenzie in the 1870s (discussed in Besold 2012 vol. 2) and Langloh-Parker’s (1953: 219–21) Emu and Bustard story. Such texts convey valuable aspects of traditional culture. Linguistically, they illustrate grammar in use in ways that paradigms or even isolated sentences in a sketch grammar do not. They typically include conversational material (including questions, greetings and leave-takings, etc.) and discourse features such as the use of particles, repetition, interjections, sound effects, etc. Nevertheless such written texts involve difficulties of interpretation, even if they are provided with a free translation or word-for-word glossing. The translation rarely communicates the total meaning of the sentences. Some of the meaning may be implicit: characters may not be introduced, and it may not be clear who is speaking in dialogues. The words of the text may not be known from any other source (e.g. wordlists). The internal morphological structure of words may be impossible to analyse precisely, even with the help of a sketch grammar that may be available. It is important to recognize that all written historical material, even text, is not complete language. The philologist is working with an interpretation and selection, mostly by a non-fluent speaker. Features such as intonation, tempos, pauses, vocal effects and accompanying body movements (gestures and facial expressions) are generally not captured. 17 It was only Morelli’s experience of over 25 years in working on this language that gave him the background to recognize the new information. Long-term involvement is key to fuller philological analysis.

74

8.4 Philology and language revival 8.4.1 The completeness of philological descriptions Since the aim of revival grammars and dictionaries is to allow the most complete possible reconstruction of the language, they must aim for the fullest possible description of the language. Giacon’s (2017: 7) observation on what is very common in rebuilt YG can be generalized to any revival. Any features of traditional YG that are clearly stated can potentially be part of rebuilt YG. Any features that are not explicitly stated, taught and well learnt will not be part of rebuilt YG unless they also happen to be part of English.

A standard language description will capture many features of that language, generally focussing on the more prominent and easily describable, and following the pattern of existing descriptions. For some other purposes, however, a less comprehensive account may be sufficient; for example, studies relevant to historical-comparative issues or typological studies. The description of complex areas of language is extremely challenging; see for example the long discussion—but incomplete description—of demonstratives in Giacon (2017). The tendency to keep to the simpler parts of language is even stronger in philology since the sources of information are so limited. They are limited in quantity, but also consist of interpretations, whether analysis (grammatical description) or even transcribed speech, rather than being actual realized language. Many studies neglect some of the most common and most important features of language. Enfield (2017) looks at numerous studies of conversations, including focussing on the importance of time gaps in conversation and the existence of standard repair and turn-taking structures. These are important in the everyday use of language (and so are relevant to language revival), but are not topics studied in standard linguistics (cf. also Gale, Chapter 64, this volume). Philology for language revival must try to describe as much of the language as possible. If language revival (in Australia) is based primarily on word forms, the ‘revived’ language will be a relexified English. The challenge is, while working with quite limited materials and no fluent speakers, to develop as comprehensive as possible a description of the language. Revival grammars must, in particular, look for those features which are easily overlooked—those features which are perhaps unique to the language or to a group of languages— and for that very reason are so easily overlooked in earlier

philological methods for australian languages descriptions, for instance ‘associated eating’ and the inclusory construction in Yuwaalaraay-Gamilaraay (Giacon 2017: 326, 374). The result of this philology may be a growing language description that gives a set of word forms and standard grammatical descriptions (such as case and pronoun paradigms, verb morphology and so on), but also one that works on topics which may not be easily recoverable from texts, such as stress, intonation patterns, more complex verb forms, demonstratives and detailed semantics. The description of such areas will be based on the historical sources and on typology.

8.4.2 The products of philological analysis Since the aim of much philology is to assist in language revival, philological language descriptions must be as complete as possible, given the constraints of time and expertise. Such descriptions will only be arrived at progressively and not as a first analysis. They need to interpret the original sources in light of current linguistic knowledge, particularly knowledge of Australian languages. They will need, at times, to alert the reader to the limitations of the sources and give them relevant information. For example, wordlists can easily be misinterpreted. When a list has one word for ‘hit’, it is easy for the reader to assume the Indigenous and English words are equivalent. The relationship is much more complex. Eckert and Hudson (1988: 75) give at least four Pitjantjatjara words that translate English ‘hit’,18 and other languages are similar. If the sources do not have features that are common in Australian languages, such as an inclusive/exclusive distinction or inalienable possession, the reader should be alerted, and also if the sources do not show less common but widespread features such as distinctive proper noun marking or ‘non-English’ ways of asking about a name. It will be difficult to retain the original pronunciation and intonation of the language. A phonemic writing system that doesn’t fully specify flattens the variation of the original. For instance, if there is phonemic vowel length the writing system might have only two options—long or short—which may not capture the actual variation. Contemporary readers will mostly pronounce stops via an English interpretation—voiced or voiceless, depending on the orthography. Perhaps the writer should warn the reader that, as Reid (2010) says, revived languages will retain very little of the original phonology that differs from English (cf. Gale, Chapter 64, this volume). 18 Their examples include pungu ‘hit (with stick)’, atunu ‘hit (with a thrown stone)’, rungkanu ‘hit (with a thrown stick)’ and wakanu ‘hit (with a spear or arrow, i.e. pierced)’.

Language descriptions generally say little about poetry, wordplay, and similar topics (though cf. Walsh, Chapter 59, this volume; Green et al., Chapter 50, this volume). Yet it is clear from the YG data that these were distinctive features of language use. Langloh Parker (1905: 132) gives a number of riddles, and the Emu-Bustard Yuwaalaraay story she recorded (as spoken by her consultant Buudhaa) contains many word plays. For instance, the emu pair cut off [garra-li] each other’s wings while the Bustard tucks hers in [wa-li]. The emu keeps [garrawa-li] her many children and the Bustard keeps [garrawa-li] her wings. Such features are not important in typological comparisons. They are very important in language use, however.

8.4.3 The ‘philological spiral’ Philological examination of historical sources is an openended process. Early descriptions can be expected to come up with fairly accurate word forms (particularly when there is audio information). Depending on the sources, other features, such as the main cases, suffixes and their allomorphs, can also be described. As linguistic knowledge develops, more can be incorporated into grammatical descriptions. For instance Stockigt (2017: 2) points out that the description of the case systems, of ergativity and of bound pronouns challenged the early, classically-trained grammarians. These are now the bread and butter of Australian grammatical description. Introductions to Australian languages describe many other features which are now relatively well understood. One challenge for philologists is to describe features of the language that are not well recorded, for instance subordination. As heard from the tapes, Williams asked Arthur Dodd to translate a number of sentences with relative clauses. Dodd is reasonably quick and confident where the relativized noun is the subject or object of the main clause (e.g. ‘The girls caught the dog that was biting the baby’), but there are long pauses as he clearly struggles when other arguments are relativized, or when, as in the next sentence, the main clause is passive (‘The girls who were seen by the men hit the baby’). The transcription needs to show the consultant’s words, but also the pauses and uncertainty. Another challenge for the philologist is to describe features that are generally not well described—perhaps because they are found in one or few languages, or because they have not generally been described. An example of a rare, if not unique feature, is the verbal suffix -guwa-y. Donaldson, working with Eliza Kennedy, was struggling to find its function. She made many suggestions—tense, mood, aspect, number. The light went on when Kennedy said something like ‘it is a pitying sort of thing’ (heard by Giacon

75

john giacon and harold koch on Wangaaybuwan tapes). Donaldson (1980: 184) glosses the suffix pity. Williams (1980: 74) may well have known of the -dha-y ‘associated mouthing’ (Giacon 2008) suffix from Donaldson, a contemporary at ANU working on a related language, but she did not find it in the tape sources. In later work, with more time, Giacon has found numerous examples. Dixon (2002: 206, 532) discusses a valency decreasing suffix which has a number of functions and ‘is found in languages right around the continent’. His description is consistent with the situation in Wangaaybuwan, where detransitivizing (the mainly transitive) L-class verbs results in a form identical to (the mainly intransitive) Y-class verbs. Donaldson (1980: 168), whose study was based on a 3–4 year PhD project, describes the resulting form as intransitive. Williams (1980), whose study results from an honours thesis, did not describe the process. This construction is common in the Yuwaalaraay tapes (Giacon 2017: 289–302), but is more complex than Donaldson describes. On occasions the derived form is intransitive, at other times transitive, and some examples have a reflexive meaning. Giacon recognizes the complexity, but does not have a description of the principles governing the behaviour. He refers to this process as production of a middle verb rather than detransitivization—since the latter term describes only one result. There are other Yuwaalaraay verbal derivations which are more complex and even less well described. Giacon (2017: 300) has a number of examples which seem to include both detransitivization and a reflexive suffix. Use of just the reflexive would presumably mean ‘he tied himself ’ and it may be that the middle plus reflexive allows the addition of a further argument, ‘leaves’.

Mathews glosses as ‘am beating’. Giacon’s original interpretation was that the transcription was likely an error. The fact that similar examples are found in recordings from seventy years later suggests that the error may be in the translation. The point is not the specific example, but rather that the process of philological analysis is ongoing. More will be understood as new information from the same and other languages sheds light on previously obscure areas. However it is important to be aware of the limitations of linguistics based on historical materials. Even the best historical descriptions of Australian languages were not produced in ideal conditions and we should not be surprised they are incomplete. Stockigt (2017: 373) alerts us to the difficulty of analysis from speakers and the much greater limitations of analysis from historical sources:

(2) [Prompt:] He was pretending to be a tree. AD/JM 8187 84 dhayn-duul=bala nhama-li / yulaa.y-ngiili-ngindaay / man-one=ctr 3.def-?? / tie.middle?-refl-sub / girran.girraa-gu leaves-erg ‘He tied himself up with leaves.’ AD ‘But the man who had tied himself up with leaves.’ JG

Philological methods are required to interpret historical records of Australian languages, using information derived from them to contribute to a description, as complete as possible, of a language, either supplementing other materials or when necessary providing the only available data. Although the goal of producing a description of the language is shared with the practice of field-based work with fluent speakers, somewhat different methods are required for interpreting older documents. We have called attention to the many pitfalls in using such data, provided some pointers on how to use such sources, and given some indications of what possible results can be expected.

There are similar forms in Mathews (1903a: 266; see Giacon 2017: 285) with bumaingilda, presumably buma-y-ngiilda-[nha] ‘hit-Middle-Reflexive-Continuous-[Present]’ which

76

That it took the Lutherans almost thirty years to come closer to describing the complexity of the system of case marking on different nominal types is of consequence to the certainty upon which some other early analyses can be relied. That after thirty years of intense engagement, the different marking of cases on female personal names appears to have remained undocumented at Bethesda should alert the linguist engaged in the reclamation of other languages from older sources written with much less exposure to the structure of a language, to the type of material that is likely to have never have been recorded.

8.5 Summary and conclusions

PART II

Structures

A. Phonetics and phonology

chapter 9

Articulatory and acoustic phonetics Marija Tabain

9.1 Introduction In this chapter I consider some key articulatory and acoustic phonetic characteristics of consonant production in Australian languages. The articulatory data presented here consist of a case study of a single speaker of Arrernte (for purposes of clarity), based on multiple articulatory techniques (electro-palatography, electro-magnetic articulography, and ultrasound). By contrast, the acoustic data are based on recordings of 20+ speakers of three Central Australian languages: Arrernte, Pitjantjatjara, and Warlpiri. The reader is referred to Fletcher and Butcher (2014), and the various references cited there and throughout the present chapter, for further information on the phonetics of Australian languages. The phonetic study of Australian languages is largely driven by the extensive coronal contrasts which are a defining characteristic of this language family, and so unusual in the world’s languages (Maddieson 1984; Evans 1995b; Ladefoged and Maddieson 1996; Gordon 2016). The complexity of the contrasts is such that many aspects of articulation are recruited to preserve the cues to consonant place of articulation, at times at the expense of typical coarticulatory processes in running speech. This phenomenon has been termed the place-of-articulation imperative (Butcher 2006), which is considered to be a driving force in the articulatory patterns of Australian languages. For instance, whilst cross-linguistically languages have been shown to control CV sequences more carefully than VC sequences (e.g. Byrd 1994), the Australian languages Arrernte, Yanyuwa, and Yindjibarndi have been shown to control the CV sequences and VC sequences equally carefully (based on formant transition and durational data—Tabain, Breen, and Butcher 2004). This may reflect a strategy to maximize formant transition cues both into and out of the consonant, thereby preserving place of articulation information. Table 9.1 shows the consonant phoneme inventory for Arrernte, which is a typical ‘maximal’ consonant inventory for Australian languages (Breen and Dobson 2005). This

inventory has six places of articulation and is particularly rich in sonorant consonants (nasal and lateral). A salient feature of Australian languages is that the coronal consonant inventory is divided into apical and laminal consonants. The apical consonants (articulated with the tongue tip) are the alveolars and retroflexes; the laminal consonants (articulated with the blade of the tongue) are the dentals and palatals. Although there is an articulatory basis to this classification into apical and laminal (see more below), the distinction is largely driven by phonology. However, some languages (such as Kalaw Kalaw Ya) have only one apical place of articulation (see Gasser and Bowern 2014 for further discussion of these distributions); and some languages (such as Pitjantjatjara or Warlpiri) have only one laminal place of articulation. In these situations, the single apical or single laminal may have properties intermediate between the two categories given in Table 9.1 (see Butcher 1995 for further information on such articulations). In this chapter, I provide an overview of the articulation and acoustics of the various Australian consonants, in particular the lingual consonants (coronals and velars), by focussing on the Central Australian language Arrernte, which has a maximal Australian phoneme inventory. This phoneme inventory can be described as ‘long and thin’—it has many places of articulation, but is lacking in manner and voicing contrasts (i.e. there are no fricatives or affricates, and there are no voicing or voice quality contrasts— although some languages from the central northern and north-eastern parts of Australia do show voicing and fricative contrasts). On the other hand, each oral stop is accompanied by a nasal at the same place of articulation, and the coronal consonants also have a lateral at the same place of articulation. Such characteristics make Australian languages an excellent test case for the phonetic principles of gestural economy and articulatory symmetry (Maddieson 1996; Recasens 2010): it is generally believed that whenever possible, the same articulatory gestures may be reused for different manners of articulation. Australian languages also provide an excellent test case for principles of phonetic dispersion in consonant systems (Manuel 1990; Schwartz, Boe, Vallee,

Marija Tabain, Articulatory and acoustic phonetics. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Marija Tabain (2023). DOI: 10.1093/oso/9780198824978.003.0009

81

marija tabain Table 9.1 The phoneme inventory for Arrernte, which represents a typical ‘maximal’ consonant inventory for Australian languages (note that the pre-stopped nasals are not shown here). Note that /w/ is a labio-velar glide. Also, in Arrernte, the velar glide /ɰ/ is phonemic in the speech of older speakers. peripheral

apical

laminal

bilabial velar alveolar retroflex dental palatal stop

p

k

t

ʈ

t̪

c

nasal

m

ŋ

n

ɳ

n̪

ɲ

lateral

l

ɭ

l̪

ʎ

trill

r

glide

w

ɻ

j

and Abry 1997), since the presence of so many coronal contrasts is likely to prevent any individual consonant from ‘encroaching’ on another consonant’s place of articulation.

9.2 Articulation of lingual consonants In this section I illustrate the articulation of Arrernte lingual consonants via EMA (electro-magentic articulography), EPG (electro-palatography), and ultrasound data from a single female speaker of Arrernte, Mrs Janet Turner (born 1973— see Tabain 2009a, b; 2011; 2012; Tabain and Beare 2018 for articulatory data from additional speakers of Arrernte). EMA is a technique that allows the tracking in time of selected points on the speech articulators—in the data I present here, the selected points are the Tongue Tip (with a sensor placed about 1 cm from the tip of the tongue), the Tongue Body (with a sensor placed about 2–3 cm from the tip of the tongue), and the Jaw. By contrast, EPG is a technique that provides excellent information on contact between the tongue and the entire hard palate; it uses a custom-made artificial palate embedded with 62 sensors to register the contact. The Articulate palate, which is used in the data presented here (Wrench 2007), extends from the base of the upper incisors, to the border between the hard and soft palates. As such, it provides some information on velar articulations, and extensive information on coronal consonant articulations. Finally, ultrasound is a technique that allows for imaging of the tongue root and the tongue body.

82

It provides no information on the tongue tip-blade; however, given the volume-preserving nature of the tongue, and thus the high degree of coupling between various portions of the tongue, many factors relevant to lingual consonant place and manner of articulation are actually seen in more posterior regions of the tongue. Figure 9.1 shows EMA data for the coronal consonants, and Figure 9.2 shows ultrasound data from the same speaker (recorded 11 years apart). Looking first at the laminal data in Figure 9.1b (dental and palatal stop and nasal, right column), it can be seen that the jaw position is quite high for this class of sounds, in particular the palatals. This is because laminal articulations usually recruit a substantial portion of the tongue body for their articulation, and as such the tongue position is quite constrained during consonant closure. As a result, the jaw position is quite high due to the coupling between the jaw and the tongue. The jaw is however higher for the palatal consonants than for the dental consonants, since the palatal requires a much higher position for the tongue body than does the lamino-dental. In Figure 9.1a, the extensive movement in the jaw is visible for the palatal, compared to lesser movement for the dental (as measured by the distance between the symbols marking each time-point on each trajectory); and in Figure 9.1b, the high jaw position for the palatal is evident for the stops and nasals (due to speaker fatigue during the recording, there are no palatal lateral data for EMA). It should also be noted that for the palatal stop /c/,1 the endpoint for the jaw trajectory is higher than the midpoint (normally the jaw and tongue lower at consonant release). This is because the jaw continues its upward trajectory into the affricated release of this particular stop consonant, with a high jaw position necessary for turbulence creation at the lower teeth as part of the acoustic requirements for sibilant affrication. Figure 9.2 confirms the high tongue position for palatals; this high tongue position is accompanied by a very forward tongue root position (on the left of each panel), which is typical of palatal articulations including front vowels. The tongue trajectories in Figure 9.1a also show the extensive movement that is typical of palatal sounds as part of the high target. The dental provides a point of contrast with the palatal in that whilst the tongue tip moves upwards for both of these laminal consonants, the tongue body has a noticeable downwards movement for the dental consonants in contrast to the upwards movement for the palatals. This low tongue body position for the dentals is particularly noticeable in 1 Throughout this chapter I use the symbol /c/ and refer to this sound as a palatal consonant. However, note that in some Australian languages, including Arrernte, ‘palatal’ consonants are better described as alveo-palatals (Keating and Lahiri 1993), and that they are strongly affricated. For these reasons, a more accurate symbol for the sound might be /ʨ/.

articulatory and acoustic phonetics

(a)

Tongue Tip

Tongue Body

−800

s

−720

m

e

−830

e

−760

e m

−840

e s m

e m

s

m

300

310

320

330

510

520

Tongue Tip

530

e s m

s

ms

s

−800

e m

−810

s

−730

m

−820

m

ee

−740

e

320

340

360

380

−750

s

e

−1730

m

−1740

e m

−1750

530

540

550

560

570

−1700 s

e

−65

−950

m

300

420

Tongue Tip

stop lateral nasal

m s e s

−1740 −1760

440

460

480

500

520

e m s

–100

–80

–60

Tongue Body

–40

e m

m

−665

e

e m

−1690

−670

−800

–20

0

Jaw

−780 −790

−55

−1740

e

s

−830

m

−60

'HQWDO

m e

−1720

m

−900

−820

280

−70

s

e m

260

−75

Jaw

m

−850

240

nasal rhotic stop lateral

e

s

e

s

s

me

−80

580

e s

−800

−810

Retroflex

s

−750

se

−60

m

m

−800

me

Tongue Body

m

−65

m s

e

Tongue Tip

−780

−1720

em m

−70 Jaw

e

s

−720

s

−75

−1710 s

−710

−780 −790

−80

540

Tongue Body

−770

−790

s

−1740 290

−770

s

−1730

−850

−830

rhotic lateral nasal stop

em

−780

s

e s

−1720

s

Alveolar

m e

−1710 m

e

s

e

e

−740

m

−820

m s

s m

−810

Jaw

−1700

3DODWDO

m

−675

−810

s

−820

s

210

220

230

240

250

−680

m

−685

e

460

e

s

−1720

s

470

−1710

480

490

s

e

nasal stop s

–110 –105 –100 –95 –90 –85

Figure 9.1 (Continued on next page)

83

marija tabain (b)

Tongue Tip

−770

−790

Tongue Body

m e m

s

e

s

s

s

s

m

m

e

300

350

s 450

400

500

−650

m

m e

e

−1700

−750

s

−810

m

m e

−830 200

250

s

e

300

−850

e m

es

350

500

520

540

e m

−840 280

e

s m

−1735

−1745

s −100

360

−90

−80

e m s

50 0

52 0

54 0

−70

l l̪ ɭ s m e s

−1765

−800 340

es e s me

m

−1755

s m

320

e

e m

e

−60

Jaw

e

−760

m

s 300

−110

560

s −720

s

ɲ n ɳ n̪

Tongue Body

e

−820

m m s

−1740

480

−70

e −1720

s m

−80

m

s

m

Tongue Tip

−800

−90

s

e

m

s

s

Jaw

s

s

e

−100

550

Tongue Body

m −790

m e m e

s

−1730

Tongue Tip −770

s

m

e −1000

250

t c ʈ t

e

−1710

s

e

−900

e m

−1690

m

m

–830

s e

m e

−800

s –810

m e

−700

Jaw

56 0

580

−70

−60

−50

−40

−30

−20

−10

Figure 9.1 EMA (electro-magnetic articulography) data recorded in 2004 for coronal consonants as produced by a female speaker of Arrernte. Data are shown for the Tongue Tip (left column), Tongue Body (centre) and Jaw (right column). Data are averaged and time-normalized. ‘s’ marks the acoustic start of the consonant; ‘m’ marks the midpoint; and ‘e’ marks the acoustic endpoint of the consonant. All data are inter-vocalic in a central vowel context. For the data shown here, number of tokens for a given consonant varies between 9 and 43. (a) Data plotted according to consonant place of articulation (including the rhotics); (b) data plotted according to consonant manner (not including the rhotics). On the y-axis, higher numbers indicate a higher articulator position; and on the x-axis, higher numbers (to the right of each panel) indicate a more posterior articulator position. Color versions of plots available online.

Figure 9.1b, and it is also shown clearly in the ultrasound data in Figure 9.2, where the dentals have a noticeably lower tongue body position compared to the other places of articulation. This low tongue body position for the dentals is likely due to a need to keep the articulation purely dental, in contrast to the alveolars in the language. Turning now to the apical places of articulation—alveolar and retroflex—it can be seen that the distinction between the two is not as clear as the distinction between the two laminals. Figure 9.2 shows that these apical sounds are intermediate between the dental and the palatal in terms of tongue body position (i.e. the front part of the tongue visible on the images), and pattern somewhat with the dentals in terms of tongue root position. However, the tongue body position is higher for the retroflex than for the alveolar,

84

and there is some evidence that the tongue root may be more forward for the retroflex. These observations regarding the tongue body position for apicals being intermediate in height between the dental and the palatal are confirmed by the EMA data (Figure 9.1b). In addition, the observations regarding alveolar vs. retroflex differences extend to the rhotic manner of articulation, with the trill /r/ having a lower tongue body position and a more posterior tongue root position than the glide /ɻ /. The particularly retracted tongue root position for the trill is believed to facilitate bracing of the tongue in order to initiate trilling. The other point to note regarding the apical distinction is the extensive movement in the tongue trajectories for the retroflexes (with the exception of the rhotic), compared to the alveolars (see in particular Figure 9.1a). This reflects the

articulatory and acoustic phonetics lateral

nasal

60 40 20 place

0

alveolar dental palatal retroflex

stop

rhotic 60 40 20 0 –68

–48

–28

–8

12

32

52

–68

–48

–28

–8

12

32

52

Figure 9.2 Ultrasound data recorded in 2015 for coronal consonants as produced by a female speaker of Arrernte. Data are sampled at the acoustic midpoint of the consonant, and are rotated to the bite-plane of the speaker. Each spline represents a Smoothing Spline ANOVA analysis, calculated using polar co-ordinates. The lips are to the right of each image. The data presented here represent a total of 1,132 coronal consonant tokens, taken from intervocalic, word-initial, and word-final position, as well as from homorganic consonant clusters. Color versions of plots available online.

ballistic forward movement of the tongue during consonant closure: the tongue tip is retracted into a post-alveolar position at the onset of consonant closure, and is released at a more forward, alveolar location. This is facilitated by a lower jaw position for the retroflex articulation, which allows for the tongue tip retraction. It can also be seen in Figure 9.1b that the tongue tip and tongue body positions for the retroflex are routinely more posterior than for the alveolar—however, this difference is not always significant, a point which is taken up a little further below. Before leaving the EMA and ultrasound data, it is worth highlighting some important differences between the various manners of articulation. Although it can broadly be said that the jaw position is highest for the palatal and lowest for the retroflex, with the dental and alveolar in between (Figure 9.1b), Figure 9.1a shows that jaw position is consistently lower for the laterals, and consistently higher for the stops (with nasals in between). This is a cross-linguistic finding (Keating, Lindblom, Lubker, and Kreiman 1994). The high jaw position for the stops is believed to be due to acoustic requirements for the stop burst (i.e. the lower teeth serve as an obstacle to airflow, and therefore shape, increasing the amplitude of the burst spectrum—Mooshammer, Hoole, and Geumann 2007). In the case of Arrernte, this has been shown to be relevant for the alveolar, retroflex, and palatal

stops, but not for the dental, which has a broad—as opposed to a peaky—stop burst spectrum (Tabain 2012). The low jaw position for laterals, by contrast, is believed to be due to aerodynamic requirements for lateral airflow (note that while the alveolar rhotic trill has a jaw position in between the alveolar nasal and lateral, the retroflex rhotic glide has quite a high jaw position). It is notable that jaw position for the laterals does not quite follow the place-ofarticulation pattern mentioned above—jaw position is lower for the dental lateral than for the alveolar and retroflex laterals. This is likely due to a general incompatibility between the requirements for lateral airflow and the requirements for different coronal places of articulation, a point which will be mentioned again below. Figure 9.3a shows EPG data for the lingual stop, nasal, and lateral consonants of Arrernte. It can be seen that the dental consonants involve contact on the front two rows of the EPG palate—this indicates a clearly dental articulation. By contrast, the alveolar consonants involve closure at rows 2 and 3 of the palate, indicating a clearly alveolar articulation. The retroflex consonants show contact between rows 2 and 5, with row 5 being a post-alveolar point of contact. However, what is particularly noticeable for the retroflex consonants (and to a lesser extent for the alveolar consonants) is the extreme variability in the data—the light grey

85

marija tabain shading shows that the contact is not consistently at any one location across repetitions of the sound. This point is taken up further below when the apical consonants are considered in the light of lexical stress. Whereas the dental, alveolar, and retroflex consonants involve a relatively thin band of contact in the relevant region, the palatal consonants involve quite extensive contact across the palate. This is typical of palatal articulations (Recasens 1990; 2013). The velars, by contrast, show contact only at the back row of the artificial palate—EPG does not capture the entire extent of velar contact, but there is

doubtless a good deal of lingual contact posterior to the contact captured using this technique. Based on Figure 9.3a, a few observations can be made regarding manner of articulation differences. Firstly, the lateral consonants show evidence of lateral airflow along the sides of the tongue, with electrodes usually not showing contact along column 8 for this speaker. Also, it is often the case that nasals show more contact overall than stops do (though this trend is somewhat obscured by the great variability in the case of the apicals). This is likely due to airflow considerations, since the build-up of intra-oral pressure behind

(a)

Figure 9.3 (Continued on next page)

86

/t/

/n/

/l/

/c/

/ɲ/

/ʎ/

/t/

/n/

/l/

/ʈ/

/ɳ/

/ɭ/

/k/

/ŋ/

articulatory and acoustic phonetics (b)

/j/

Stressed

Unstressed

/ɻ/

Stressed

Unstressed

/w/

Stressed

Unstressed

Figure 9.3 EPG (electro-palatography) data recorded in 2008 for the lingual consonants of Arrernte. Data are sampled at the acoustic midpoint of the consonant. The top of each palatogram represents the dental region, and the bottom of each palatogram represents the front of the velar region. There are eight rows in each palatogram, and each row contains eight electrodes, with the exception of the first row which contains six electrodes. Each electrode is shaded according to frequency of contact across the repetitions—the darker the shading, the more frequently that electrode is contacted. (a) Lingual stops, nasals, and laterals. Each palatogram represents average contact over multiple repetitions (between 15 and 179 tokens, median 48). (b) Approximants, according to lexical stress (738 tokens in total, ranging between 14 for unstressed /j/ to 314 for stressed /ɻ /).

the oral stop closure may prevent more extensive linguopalatal contact. It is also one of the reasons (freedom in jaw variability being another) that nasals are often more prone to variability according to vocalic context or according to prosodic context (see Tabain, Fletcher, and Butcher 2011 for data on palatal nasals in Arrernte and Warlpiri). In general, consonants in Arrernte do not show evidence of articulatory strengthening in strong prosodic positions, such as in positions of lexical stress (though see Fletcher and Butcher 2014 for discussion of durational correlates of different prosodic positions). This is in contrast to extensive evidence of articulatory strengthening, depending on segmental identity, in other languages of the world (see Fletcher 2010; Fougeron and Keating 1997). The possible exceptions, at least in Arrrernte, are the glides /j, w, ɻ /, and the apicals (the apicals are discussed further below). Figure 9.3b shows the glides plotted according to lexical stress (assigned according to the stress status of the following vowel—durational evidence for Arrernte shows that the CV syllable is lengthened under stress, not the VC syllable as might be expected based on the phonology—Tabain 2016; Breen and Pensalfini 1999). It can be seen that the

palatal glide /j/ has extensive lateral contact from the alveolar region to the velar region, while the rhotic /ɻ / has contact from the post-alveolar region through to the velar region, with a noticeable constriction in the post-alveolar region. The labio-velar glide /w/ shows lateral contact in the last three rows of the palate (once again, a large amount of posterior contact is probably not being captured for the velar). The important point to note is that all three glides show more extensive contact when stressed than when unstressed. Such prosodic strengthening is believed to be largely due to greater duration, which allows the segment more time to reach its ideal target. In the case of the glides, the articulation becomes more consonantal, with greater tongue-palate contact (most vowels do not show any tongue-palate contact). Thus, in strong prosodic position, the glides—or semi-vowels—become more consonant-like, and less vowel-like. It is notable that the glide inventory of Australian languages is not typologically unusual, and not ‘crowded’—this may be an important factor which permits such prosodic variability. However, it should be remarked that these observations on glides are quite tentative and still require further investigation.

87

marija tabain Before leaving this section, a few observations are needed regarding the velar consonants. Velars have a low jaw position, consistent with a more posterior region of the tongue being used as active articulator (velars are not shown in Figure 9.1 above, but velars and retroflexes generally have a similarly low jaw position—cf. Keating, Lindblom, Lubker, and Kreiman 1994 for velars in Swedish and English). Importantly, in Australian languages, there is both acoustic and articulatory evidence from a variety of languages that velars in a non-front vowel context (i.e. in the context of /a, u/) are articulated further back than in Germanic languages (Butcher and Tabain 2004). Figure 9.4 shows a schematic of this principle based on acoustic data from the Australian languages Yanyuwa, Yindjibarndi, and Arrernte; and on the Germanic language Australian English. To the right of each column is the F2 (second formant) value at the midpoint of the vowel, and to the left of each column is the F2 value at the onset of the vowel (representing the velar articulation at consonant release). F2 is an important perceptual cue to consonant place of articulation: in the case of velars, a higher F2 value is indicative of a more forward velar articulation. It can be seen that the velar is articulated quite far forward in the context of an /i/ vowel—this is a crosslinguistic tendency due to coarticulatory bio-mechanical constraints between the velar and the front vowel (Recasens and Espinosa 2010). In the case of English, the velar is similarly articulated relatively far forward in the context of /a/, as well as in the case of /u/. By contrast, for the Australian languages, the velar is articulated further back as the ‘place’ of the vowel moves further back: front, central, then back.

As a result, the velar articulations in the context of /a/ and /u/ are quite far back in the Australian languages. This is possibly due to a need to keep the velar from encroaching on the articulatory and acoustic space of the palatal, since historically, fronted velars may become affricated palatals (Keating and Lahiri 1993—cf. the palatals described above). This is thus a possible consonant example of the principle of sufficient dispersion.

9.3 A special consideration of apical consonants The apical consonants presented above show great variability compared to the laminal consonants and compared to the velars. In Arrernte, and possibly many Australian languages, the apical contrast is somewhat marginal, with very few minimal pairs and a very low functional load (Tabain 2009a). In addition, it has been observed that retroflexion may behave as a word-level prosody, with evidence that it crosses syllable boundaries (Butcher 2006). Figure 9.5 shows the same apical EMA data as above, but this time plotted according to lexical stress of the following vowel. Figure 9.6a shows the same apical EPG data as above, but also plotted according to lexical stress of the following vowel. Figure 9.6b shows this same EPG data according to stress, but across time, using the Centre of Gravity measure which captures the point of contact along the palate (the higher the CoG value, the more forward the contact).

Australian Aboriginal languages

Australian English

24

24

23

23 k

22

k

i region of F2 /F3 ‘cross-over’

20 k

18

Vowel onset

20 19

a

17

k

18 k

16

15

15

F2

(ERB)

14

u Vowel target

Vowel onset

a

17

16 14

i

21

21 19

22

F2 (ERB)

u

Vowel target

Figure 9.4 Schematic of velar place of articulation based on acoustic data from the Australian languages Arrernte, Yanyuwa, and Yindjibarndi; and from the Germanic language Australian English. ‘F2’ is the second formant (an acoustic cue to consonant place of articulation), and ‘ERB’ is Equivalent Rectangular Bandwidth, a non-linear perceptual warping of the frequency scale.

88

articulatory and acoustic phonetics Tongue Tip

Tongue Body

−760

s

s

−780 −800

m

s m

−820

300

320

m e

−740

m

360

380

400

m s

500

520

280

s

−750

e

300

320

340

360

380

−800

320

s

530

540

550

−725

s m s m e

560

570

m e

360

380

s sm

nS nU KS KU

e

s

−85

−80

−1730

e

−1750

m

560

580

m

m

m e

540

s me

e

s

e e

−75

−70

Jaw

−1770 520

e

Nasal

580

−735 340

−60

e m −1750

520

m

s

s −715

300

−65

−1730

s

e e m s m e e m s

−1710

em

Tongue Body

−780

m

−70

s

s

Jaw

m 510

e

Stop −75

580

m

e −770

es me

s

Tongue Tip

−840

560

s

mm m e e s

−840

−820

540

−730

s e s

ʈU ʈS tS

e

−1715

s

m

m

em

Tongue Body

−780

−820

e

−1725

Tongue Tip

−800

s

m

m e

−780

340

−1705

s

−760 e

−840

s

−720

e

e

Jaw

−700

−75

lS lU ȷS ȷU

s

Lateral −70

−65

s

e m s

−60

−55

−50

−45

Figure 9.5 EMA data for apical consonants according to lexical stress. For the data shown here, there is a total of 124 tokens. Color versions of plots available online.

Several observations can be made based on these plots. First of all, both the tongue tip and the tongue body show a lot more movement overall for the retroflexes compared to the alveolars. This movement is in a forward direction. Secondly, the tongue tip and body are overall in a more posterior position for the retroflexes as compared to the alveolars. However, there are instances of alveolars showing retroflex-like behaviour, for instance in the tongue movement trajectories for the stop (note that there are no unstressed alveolar tokens for the stop /t/ in the EMA data, only stressed tokens). The final important point in the EMA data is that the jaw is noticeably lower for the unstressed retroflex. The other three categories—stressed alveolar, unstressed alveolar, and stressed retroflex—all show a higher jaw position. This low jaw position for the unstressed retroflex is particularly noticeable in the case of the nasal data. As mentioned above, the retroflex requires a low jaw position in order to allow the tongue tip to retract into a post-alveolar place of articulation. The intrinsically lower jaw position for nasals facilitates this tongue tip retraction. Looking now to Figure 9.6a, it can be seen that by presenting the apical data according to lexical stress, the differences

between alveolar and retroflex become much clearer. The unstressed stop and nasal retroflexes /ʈ, ɳ/ clearly show a single row of contact at row 4, which is post-alveolar. This very thin contact is indicative of a purely tongue-tip articulation (as opposed to the involvement of a slightly larger portion of the tip-blade complex). A similar pattern is seen for the stressed retroflex nasal /ɳ/, but not for the stressed retroflex stop /ʈ/, which shows much greater variability (as indicated by the lighter grey shading across a larger area of the palate). The alveolar stop and nasal /t, n/ show some variability, but overall the contact is more forward than for the retroflexes, with the stressed alveolars having slightly greater contact than the unstressed alveolars (which may be evidence of articulatory strengthening in the strong prosodic position). It should be pointed out that these various observations do not hold for the lateral manner of articulation. As can be seen on the palatograms, there is greater variability for the laterals than for the stops or nasals, even when data are plotted according to stress. This is true even for the unstressed retroflex /ɭ /. Some clarification regarding this pattern is seen in Figure 9.6b, which shows the EPG contact profile across time (i.e. across the entire duration of

89

marija tabain the consonant). It can be seen that overall, for both alveolars and retroflexes, EPG contact patterns for the lateral are more forward than for stops or nasals. This suggests that there is some incompatibility between a lateral manner of articulation—with air flowing along the sides of the tongue—and a more posterior apical articulation. Figure 9.6b also clearly shows that the unstressed retroflex articulations are, for any given manner, the most posterior articulations, with a ballistic forward movement of the tongue (indicated by the rising lines). The stressed retroflex stop and nasal also show this ballistic forward movement, although the point of articulation is slightly more forward (the palatograms in Figure 9.6a were sampled at 0.5 on the time-scale of the plots in Figure 9.6b). The stop and nasal data also show that the stressed alveolars have a more forward articulation than the unstressed alveolars (and of course the alveolars are more forward than the retroflexes). The alveolars also show a more plateau-like articulation pattern, with closure and release at the same location. This is particularly true for the nasals, while the stops show some evidence of a movement during closure for the alveolars. Moreover, the closure for the alveolar stop is at a more posterior location as compared to the alveolar nasal. The laterals do not show as clear a difference between alveolar and retroflex as do the stops and nasals. What these time-plots do not make clear is the extreme variability between alveolar and retroflex tokens, even within repetitions of the same word, and even in the case where a word is part of a minimal pair (see Tabain 2009a for specific examples). There is tremendous overlap between the three categories: stressed alveolar, unstressed alveolar, and stressed retroflex. This is particularly true for the lateral manner, as can be seen in Figure 9.6b. It is worth pointing out that comparisons of lexical frequency or token frequency often show that the nasal alveolar and retroflex are the most frequent apicals. This may be part of the general preference for sonorant consonants over obstruents in Australian language phoneme inventories (Butcher 2006), but it may also be thanks to the general compatibility between the nasal manner of articulation and the retroflex articulation (and thus facilitation of the phonemic contrast), with the low jaw position allowing for maximal retraction of the tongue tip. By contrast, these articulatory data would predict that the alveolar–retroflex contrast has a lower functional load for laterals due to the incompatibility between lateral airflow and a more posterior placement of the tongue tip.2 2

It is generally noted that the alveolar-retroflex contrast is neutralized in initial position across the world’s languages (Steriade 1995; 2001). This is largely due to the fact that perceptual cues to the alveolar-retroflex distinction occur at the left edge of the segment (i.e. at consonant onset), where

90

Before leaving this section on articulatory data, it may be observed that while the prototypical alveolar articulation is more likely before a stressed vowel, the prototypical retroflex articulation routinely occurs before an unstressed vowel, and not before a stressed vowel. This latter observation is not in line with typical cross-linguistic tendencies towards articulatory strengthening in prosodically strong positions (cf. Fletcher 2010). However, it is in line with mostly prosodic or durational data from Australian languages, which suggests that post-tonic strengthening is a key feature of Australian languages (e.g. Butcher and Harrington 2003b for Warlpiri). Fletcher and Butcher (2014) have suggested that this allows for maximizing of cues to (a)

/n/ Stressed

Unstressed

/ɳ/ Stressed

Unstressed

/t/ Stressed

Unstressed

/ʈ/

Stressed

Unstressed

/l/ Stressed

Unstressed

/ɭ/ Stressed

Unstressed

Figure 9.6 (Continued on next page) the prototypical alveolar has an alveolar point of contact, and the prototypical retroflex has a post-alveolar point of contact. By contrast, at the right edge of the consonant (i.e. at consonant release), both of these sounds have a more alveolar point of contact, although there may be subtle differences (see acoustic data below).

articulatory and acoustic phonetics JT

(b) Stop

Nasal

Lateral

EPG Centre of Gravity

4.5

4.0 alveolarS alveolarU retroflexS retroflexU

3.5

3.0

0.00

0.25

0.50

0.75

1.00 0.00

0.25

0.50 time

0.75

1.00 0.00

0.25

0.50

0.75

1.00

Figure 9.6 EPG data for apical consonants according to lexical stress. (a) Palatograms sampled at the temporal midpoint of the consonant. (b) Normalized time-plots showing EPG Center of Gravity. The higher the Center of Gravity, the more forward the contact. For the data shown here, there is a total of 896 tokens. Color versions of plots available online.

consonant place-of-articulation, by strengthening the consonant that has both VC and CV transitions available. In the case of the retroflex consonants, the VC sequence contains the main cues to consonant place (namely a lower spectral centre of gravity, including a lower F3, before the consonant). However, this explanation is not likely to be as useful in a language like Arrernte, where almost half of all words begin with a vowel (Henderson and Dobson 1994). It is more likely that the extra length associated with a lexically stressed vowel (assuming the presence of lexical stress, which is true for Arrernte—Tabain 2016) allows time for jaw lowering and the retraction of the tongue tip into a posterior place of articulation.

9.4 Some acoustic correlates of coronal consonant production in Australian languages We turn now to a consideration of some salient acoustic characteristics of the extensive coronal contrasts found in Australian languages. Figure 9.7 shows the coronal stop

spectra for Arrernte, based on data from two female speakers (Janet and Sabella Turner). It can be seen that the dental stop /t̪/ has a relatively flat spectrum overall, as compared to the other coronal stops which have a relatively more ‘peaky’ spectrum. This is the one coronal stop for which the jaw does not need to remain high at stop burst release. Since the consonant is already being articulated at the upper teeth, for most speakers it would not be possible to channel the airflow towards the lower teeth as part of a strategy to enhance any spectral peaks. As a result, the spectral output does not have any noticeable, high-amplitude peaks for the dental stop. By contrast, the palatal stop /c/ has a peak at around 4 kHz, with a sharp drop-off in amplitude to the right of this peak. Although it is not shown here, the affrication which follows the stop burst for the palatal /c/ has quite a lot of energy (Tabain and Butcher 2014; 2015).3 3 The duration of the stop burst plus aspiration or affrication tends around 50 ms for the palatal and velar stops, around 20 ms for the bilabial and dental, and around 15 ms for the alveolar and retroflex stops. Crucially, only the palatal and velar stop bursts show significant lengthening when the following vowel is stressed (and to a lesser extent the bilabial and the dental), suggesting that a short burst is an important cue to apical consonant identity.

91

marija tabain Arrernte t t t c

90

Magnitude (dB)

80

70

60

50 0

2000

4000 6000 Frequency(Hz)

8000

10000

Figure 9.7 Averaged coronal stop spectra for Arrernte centred at the onset of the stop burst. Data are from two female speakers, and there is a total of 372 tokens in the data presented here. Color versions of plots available online.

The apical stops show a different spectral output.4 They both have a very broad peak between 2 and 4 kHz (in the case of coronal consonants, any peaks below 1 kHz are usually ignored since they are thought to mainly show influence from voicing), and there is likewise a sharp drop-off in amplitude to the right of this broad peak. Crucially, the right edge of the broad peak is at a slightly higher frequency for /t/ than for /ʈ/. This suggests that, on average, the alveolar stop has a slightly more anterior release than the retroflex stop, since the frequency of the spectral peak in stops is largely determined by the size of the resonating cavity in front of the constriction at the moment of release (other factors such as the size of the cavity behind the constriction, and the presence of any secondary obstacles such as the teeth, also play a role). Although not shown on this plot, the spectrum for /k/ has a strong and relatively narrow peak at around 2 kHz. The spectrum for /p/ does not have any noticeable peaks beyond the one below 1 kHz, and it has a gradual drop-off in energy 4 Note that the apical consonant spectra in Australian Aboriginal languages seem to be very different to the alveolar consonant spectra for languages such as English (Tabain et al. 2016), in that they have much more energy at higher frequencies. Although the reason for these differences is not entirely clear, a more purely apical articulation in the Aboriginal languages (as opposed to a broader tip-blade articulation in English) may be a contributing factor.

92

above this frequency range (see Tabain et al. 2016 and Tabain and Butcher 2014 for further details). As already mentioned, Arrernte is a language which is dominated by the central vowels /a/ and /ə/, and the spectra presented in Figure 9.7 are taken only from these central vowel contexts. However, many Australian languages have a three-vowel system /i a u/ (and in other cases, a five-vowel system /i, e, a, o, u/), and what happens to the spectra in different vowel contexts can be instructive. In general, an adjacent rounded vowel such as /u/ lowers the spectral centre of gravity (i.e. the mean frequency of the spectrum); and an adjacent unrounded vowel such as /i/ raises the spectral centre of gravity (Stevens 1998). This is reflected as a subtle shift upwards or downwards in the spectral peaks. In the case of the apicals examined in the three-vowel language Pitjantjatjara, the shift is particularly noticeable in the context of /u/, with the right edge of the broad spectral peak lowered to about 3 kHz instead of about 4 kHz (but with the small distinction between alveolar and retroflex maintained). This is due to the fact that the rounding of the lips effectively lengthens the front resonating cavity at the moment of stop burst release (the burst is only the first 2–3 ms transient following the moment of release, and it does not include any aspiration or affrication characteristics which may follow). By contrast, in the context of /i/, the right edge of the broad spectral peak is slightly higher (by about 100–200 Hz only) as

articulatory and acoustic phonetics a breathy voice quality would have a significant drop-off in spectral energy with increasing frequency). Given the importance and the frequency of sonorant consonants in the phoneme and lexical inventories of Australian languages, some consideration of sonorant sounds is required here. Figure 9.8 shows average lateral formant values based on data from three Central Australian languages (Arrernte, Pitjantjatjara, and Warlpiri). The data are

Average Lateral Formant Values

F4

F3 15

F2 Frequency (Bark)

compared to the /a/ context. This is because lip-spreading does not reduce the size of the front resonating cavity to such a large extent. One interesting piece of evidence regarding apical articulations comes from a comparison of the stop burst spectra preceding an /i/ vowel. While normally, the spectral centre of gravity for an alveolar /t/ is slightly higher than the spectral centre of gravity for a retroflex /ʈ/ (suggesting a slightly more anterior release for /t/, as mentioned above), the reverse is true in the case of a following /i/ vowel. This suggests a slightly more anterior release for the retroflex /ʈ/ in the context of a following /i/. This may reflect a general incompatibility between the tongue body posture required for a front vowel, and the retraction of the tongue tip required for a retroflex articulation, resulting in an overshoot of the ballistic retroflex articulatory target (cf. Proctor et al. 2010). This may help to explain the cross-linguistic rarity of the retroflex in a front vowel context: if at the right edge of the consonant (i.e. at consonant release), the retroflex is more alveolar than the phonologically alveolar consonant, this may provide contradictory cues to consonant place. Before leaving this section on stop burst spectra, a final important characteristic of Aboriginal language spectra must be mentioned. This relates to the long-term average spectrum. On average, across all segments and over long stretches of speech, Australian Aboriginal languages have much more energy in the higher frequencies than do other languages, as measured in spectral tilt (this is the drop-off in energy as frequency increases, with a lower spectral tilt indicating greater spectral energy in higher frequencies, due to a flatter regression line across the long term average spectrum). This phenomenon has been explored by Stoakes and colleagues (Stoakes 2014; Stoakes et al. 2011), and is deemed to be another reflection of the place-of-articulation imperative: with greater energy in the upper parts of the spectrum, any place of articulation cues will be more salient to the listener. As was seen in the stop burst spectra above, important contrasts are regularly maintained at around 4–5 kHz in the spectrum in Aboriginal languages: the differences between English and Australian Aboriginal languages Arrernte, Pitjantjatjara, and Warlpiri are particularly salient in this region of the spectrum. This greater energy in the higher frequencies is largely due to the particular voice quality which has been considered a characteristic of Aboriginal language speakers. Electro-glottographic observation of speakers of Bininj Gun Wok has shown low levels of glottal airflow, and high glottal contact quotients: the abruptness of the glottal closure, with perceptually high levels of creak or glottal fry, leads to enhancement of the higher frequencies (by contrast

10

5 F1

/l/

/l/

/ȷ /

/λ/

Figure 9.8 Lateral formants based on data from three Central Australian languages (Arrernte, Pitjantjatjara, and Warlpiri). Data are presented in Bark, which is a non-linear perceptual transformation of the frequency scale, and are sampled at the temporal midpoint of the lateral consonant.

93

marija tabain presented in Bark, which is a non-linear transformation of frequency (see Tabain, Butcher, Breen, and Beare 2016a for the original data in Hertz). This plot shows nicely the perceptual distance between F1 and F2, with a comparatively even spacing between F2, F3, and F4. This clustering in the higher frequencies is what is considered to give laterals their characteristic ‘ringing’ sound (lateral spectra are the product of both resonances and anti-resonances, and the large gap between F1 and F2 is likely due to the presence of anti-resonances created by a side-branch to the main lateral airflow). Note that laterals in Australian languages show no evidence of secondary velarization, which would result in lower F2 values. The first point to note is that the laminals /l ̪/ and /ʎ/ have a lower F1 value than do the apicals. This is likely due to the higher jaw position which is inherent for these sounds (see above): since F1 in laterals is considered to be a Helmholtz resonance formed by the constriction and the resonating cavity behind the constriction (i.e. a bottlelike resonance), the longer constriction which characterizes laminals leads to jaw raising, which in turn leads to a tighter constriction (i.e. a smaller cross-sectional area for the constriction). This in turn leads to a lower resonance frequency. The laminals are also characterized by a high F4 (although the perceptual value of this difference would need to be confirmed). The retroflex lateral /ɭ/ is characterized by lower F3 and F4 values, which might be a reflection of the more posterior placement of the tongue tip (F3 is considered a quarter-wavelength resonance of the cavity in front of the constriction, and is thus lowered as the front cavity becomes larger). The palatal /ʎ/ is in addition characterized by a particularly high F2 value: this result is consistent with F2 being a half-wavelength resonance of the back cavity (and thus F2 is raised as this cavity becomes smaller—recall that the palatal has extensive tongue–palate contact, whereas the retroflex shows a very thin band of contact). Crucially, these characteristics of the lateral ring strongly echo the formant transitions which characterize the coronal places of articulation for other manner classes (stop, nasal, and rhotic). In general, the vowels in Australian languages are considered little more than carriers of information for the consonant place of articulation (cf. Tabain and Breen 2011). Retroflexes are cued by a strong lowering of the F2-F3F4 vowel complex before the consonant,5 while palatals raise

F2 in the vowels before and after the consonant. Dentals raise F3 in the vowels before and after the consonant.6 Finally, there is evidence that the spectral properties that characterize the stop contrasts carry over to the other manners of articulation. In addition to the formant commonalities just mentioned, both stop burst and lateral ring spectra show a higher centre of gravity for palatals, a lower centre of gravity for retroflexes as opposed to alveolars, and a greater variance (i.e. a flatter spectrum) for dentals. To a certain extent this is true for nasals as well, although in that case the complexities of the interaction between oral resonances and nasal resonances makes the pattern less clear (Tabain et al. 2016b). However, it is important to note that, in contrast to many other languages of the world, Australian languages do not allow nasalization of vowels adjacent to nasal consonants (Stoakes 2014). Nasalization of vowels shifts oral vowel resonance frequencies due to coupling with the nasal cavity, and would severely compromise the place of articulation cues mentioned above. This is yet another example of the place-of-articulation imperative in Australian languages.

5 Note that work on the northern language Wubuy (Bundgaard-Nielsen et al. 2012; 2015), where the alveolar vs. retroflex contrast is not neutralized in initial position, suggests that there may be cues to retroflex place following the consonant as well, namely via a lowered F2.

6 Although evidence of a higher F3 is not seen in the lateral data for the dentals, it is possible that the transition into and out of the consonant perturbs the F3 standing wave thanks to the lower tongue position for dentals (this is highly speculative).

94

9.5 Conclusion The multiple places and extreme symmetry of the typical phonological system of Australian languages place constraints upon the articulatory strategies available to speakers. Whereas a language with, for example, only one lateral can adapt that lateral to its environment (e.g. vowel context or syllable position), a language with three or four laterals has fewer articulatory options available for adaptation. As a result, the consonant exerts more influence on the adjacent vowel, and is less variable according to prosodic position. Concomitantly, in acoustic terms, the intricate place of articulation cues in Australian languages must be sufficiently robust to survive the transmission of these complex contrasts through time. The discipline of phonetics looks forward to a better understanding of just how these articulatory strategies place cues might work.

articulatory and acoustic phonetics

Acknowledgements Some of the acoustic data presented here were recorded by my long-time collaborators and mentors Gavan Breen and Andy Butcher—I would like to thank them for their generosity, and to also thank the speakers they worked with for their commitment to language research. I am also grateful to the many research assistants who labelled the data presented here. Finally, I would like to thank the language

speakers I have worked with myself over the years, some of whom have taken part in every recording I have ever done of their language: Sabella Turner, Janet Turner, Carmel Ryan, Mia Mulladad, Phyllis Stevens, and Lorrayne Gorey (Arrernte); and Kathleen Windy, Hilda Bert, and Charmaine Coulthard (Pitjantjatjara). Kele! Palya! I wish you all the best in keeping your languages strong, and I hope that our work together has contributed to this endeavour in some way.

95

chapter 10

Segment inventories Erich R. Round

As new tools for large-scale phonological typology emerge, it has become possible to examine and understand continent-level variation among Australian languages in more detail than previously, and to add more insightful nuance to the frequent characterization of Australian phonologies as strikingly uniform (Busby 1980; Dixon 1980; 2002; Evans 1995b; Hamilton 1996b; Butcher 2006; Fletcher and Butcher 2014; Baker 2014). In this chapter I examine segment inventories, and in the next chapter linear segmental phonotactics. In both, the starting point is a classic set of observations about Australian similarity and diversity, but the primary aim is to move beyond these. To approach that goal, I observe variation in terms of major genealogical groupings of languages, comparing variation among them and within them, drawing on a large empirical dataset covering the reported phonemic inventories of nearly four hundred Australian language varieties (Round 2019a, b). These are the kinds of tools and methods which will increasingly dominate the practice of typology and the theoretical disciplines that rely on its findings in this century (Gasser and Bowern 2014; Macklin-Cordes and Round 2015). For Australian languages, they reveal a new degree of precision about phonological diversity, and in some cases qualitatively new insights. Below, Section 10.1 introduces the main parameters of variation in Australian segment inventories. Section 10.2 reviews less frequent, additional consonant types and Section 10.3 some infrequent absences from consonant inventories. Section 10.4 examines diversity among systems with two series of stops. Section 10.5 covers vowels, and Section 10.6 concludes. Data sources are described in Section 10.7. The topic of contour segments, such as prestopped nasals, is covered in Chapter 11, Section 11.3.

10.1 Main parameters of variation in consonant inventories Two thirds of Australian consonant inventories can be described in terms of just five parameters of

variation1,2 : the presence of one vs. two apical places of articulation; the presence of one vs. two laminal places of articulation; the presence or absence of a glottal stop; the presence of one vs. two series of plosives;3 and the presence of one, two, or no laterals at laminal places of articulation. Setting all parameters to their maximal value yields the inventory in Figure 10.1, represented most closely by Nhanda (Blevins 2001a; though Nhanda lacks a vibrant). Setting them to their minimal values yields Figure 10.2, represented by Wargamay (Dixon 1981). Two hundred and fifty Australian language varieties have consonant inventories that sit somewhere between these two extremes.4 Striking in cross-linguistic terms is that these inven tories, which represent the bulk of variation in Australian languages, lack fricatives, even [s] which is found in over 80% of languages outside of Australia (Maddieson 1984: 44), and that even the simplest inventories contrast two coronal places of articulation, with many Australian languages contrasting four. In terms of their phonological behaviour, such as phonotactic combinatorics and propensities to neutralize with one another, the four coronal places divide into two apical places, articulated with the tongue tip, vs. two laminal places, articulated with the tongue blade. Also striking is the large number of liquids, in particular the near-universal presence in Australian languages of two rhotic phonemes, one of which is a retroflex approximant and the other a vibrant, realized as an alveolar trill or tap (Nhanda is a rare exception in lacking the vibrant). No Australian language uses contrastive airsteam mechanisms5 or contrastive aspiration. 1 I do not cover Tasmanian languages here, since little is known about them securely (Bowern 2012c). 2 In Chapter 11.3 I discuss several kinds of contour segment, such as prestopped nasals, that could be analysed as segments or sequences. Counting them as sequences places this figure at 65%; counting them as segments places it at 64%. 3 For convenience I will refer to the contrast as one among stop series. Three types of phonological analysis of the phenomenon appear in the literature (Eather 2011: 14; Evans and Merlan 2004: 190–6): in terms of two series of mono-segmental stops; in terms of singletons vs. geminates; and in terms of syllables with two differing prosodic features. 4 In some single-laminal languages, the laminal phonemes are labelled as dental, not pre-palatal. In most if not all of them, those phonemes also have pre-palatal allophones. 5 Though see Hale and Nash (1997) on the auxiliary language Damiin, used traditionally by male Lardil second-degree initiates, which contains

Erich R. Round, Segment inventories. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Erich R. Round (2023). DOI: 10.1093/oso/9780198824978.003.0010

96

segment inventories

Fortis plosive Lenis plosive Nasal Lateral Vibrant Glide

Labial p b

Apical Alveolar Retroflex t ʈ d ɖ

m

n l r

w

ɳ ɭ

Laminal Dental Pre-palatal t ȶ d ȡ n l

ɻ

ȵ ȴ

Dorsal k ɡ

Glottal ʔ

ŋ

j

Figure 10.1 Inventory most closely approximated by Nhanda (Blevins 2001a).

Plosive Nasal Lateral Vibrant Glide

Labial p m

Apical t n

Laminal ʈ ȵ

w

l r ɻ

j

Dorsal k ŋ

Figure 10.2 Inventory of Wargamay (Dixon 1981).

is typically characterized most prominently in terms of the one-vs.-two apical and laminal parameters, which are accorded approximately equal status, yet Figure 10.3 reveals that the two parameters are in fact quite dissimilar. As we look across genealogical groups, and at the number of languages within them with one laminal place, we find proportions ranging from zero to 100% and covering much of the range in between; and the distribution is relatively balanced: its median is 50%, and mean is 59%. In contrast, languages with only one apical place of articulation are relative rarities in genealogical terms: the median proportion in the genealogical groups in Figure 10.3 is zero; the mean is 14%. As it happens though, singleapical systems are common in two of the largest branches of Pama-Nyungan, meaning that in terms of raw numbers, Australian languages with one-apical systems are numerous. Genealogically speaking, languages with glottal stops are also rare, though no rarer than single-apical languages: the

No Australian language uses post-velar supraglottal places of articulation. None contrasts stops and affricates. And nearly all Australian languages have matching nasals and stops at each place of articulation. To understand the variation within these main parameters at a continental level, it is revealing to study them with respect to genealogically coherent groups. Here I focus on the Australian language families for which I have phoneme inventory data for at least three language varieties. This allows us to view group-internal variation as well as variation across groups. Since the Pama-Nyungan fam(n = 5) ily is very large, I provide separate figures for six principal PN Warluwaric PN Central (n = 25) divisions within it (after Bouckaert, Bowern, and Atkinson PN North-east (n = 107) PN Yolngu (n = 10) 2018): the large Western, Central, North-eastern and South- PN South-east (n = 77) 6 eastern branches, and smaller Yolngu and Warluwaric. It PN Western (n = 86) should be borne in mind, that while the Pama-Nyungan Tangkic (n = 4) Western Daly (n = 7) groups will each have around the same time depth, the depth Maran (n = 4) of the other families will vary to an extent that is not well unWorrorran (n = 8) derstood, and they may be either deeper or shallower than Gunwinyguan (n = 15) Iwaidjan (n = 7) the Pama-Nyungan groups. Mirndi (n = 8) Maningrida (n = 4) Frequencies of phonemic systems with one vs. two apiYangmanic (n = 3) cal places, and one vs. two laminal places, and systems with Nyulnyulan (n = 10) Giimbiyu (n = 3) or without glottals, are shown in Figure 10.3. The picture Garrwan (n = 3) painted by Figure 10.3 may be surprising to linguists familiar Darwin (n = 3) with how Australian phonemic typology has been described in the past. Variation in Australian phoneme inventories segments unseen elsewhere in Australia, including clicks, an ingressive lateral fricative, and an ejective stop. 6 I depart from Bouckaert et al. (2018) and follow Blake (1990b), Evans (1995c), and Round (2017b) in classifying Tangkic as non-Pama-Nyungan.

Laminal

Apical

Glottal

Contrastive places of articulation

two

one

zero

Figure 10.3 Apical, laminal, and glottal places of articulation: proportion of languages within each group with one, two, or no contrastive places.

97

erich r. round median proportion is zero; the mean is 14%, but only small genealogical groups have high proportions of languages with glottals. Australian languages may have one series of plosives or two. The phonological variation among languages with two series of stop is intricate enough to warrant its own section: see Section 10.4 below. All Australian languages have at least one lateral phoneme, and laterals only occur at coronal places of articulation where there also exists a plosive. To appreciate the main pattern of variation in Australian lateral systems most clearly, we can distinguish between subsystems of laterals at apical places of articulation and at laminal places. All languages possess an apical subsystem of laterals, while only some possess a laminal subsystem. In each subsystem that a language has, we can speak of the subsystem being ‘maximal’ if it contains a lateral phoneme at each available place of articulation, and ‘submaximal’ if not. Figure 10.4 shows the proportions of apical and laminal subsystems which are maximal, submaximal, or absent, within families and Pama-Nyungan subgroups with at least three members. Apical laterals PN Yolngu PN North-east PN South-east PN Warluwaric PN Central PN Western

(n = 10) (n = 107) (n = 77) (n = 5) (n = 25) (n = 86)

Tangkic Maningrida Gunwinyguan Maran Iwaidjan Western Daly Yangmanic Worrorran Mirndi Nyulnyulan Giimbiyu Garrwan Darwin

(n = 4) (n = 4) (n = 15) (n = 4) (n = 7) (n = 7) (n = 3) (n = 8) (n = 8) (n = 10) (n = 3) (n = 3) (n = 3)

Laminal laterals

10.2 Additional consonantal segment types Having examined the main parameters of variation in Australian phonemic inventories, now we turn to the remainder, of less common and more wide-ranging variation. Discussed are: fricatives in Section 10.2.1, glides and liquids in Section 10.2.2, fronted velar and prepalatal apical places of articulation in Section 10.2.3, and labialized consonants in Section 10.2.4. Less-common absences from Australian inventories, relative to Figures 10.1 and 10.2, are discussed in Section 10.3.

10.2.1 Fricatives Lateral subsystem maximal

submaximal

absent

Figure 10.4 Lateral subsystems at apical and laminal places of articulation.

Most striking is that submaximal subsystems are relatively rare. Languages tend to have a full complement of apical laterals, and if they have any laminal laterals at all, a full complement of them also. A consequence is that in most instances, the total number of laterals that a language has follows from two parameters: whether or not it has a laminal subsystem at all; and how many apical and laminal places of articulation it distinguishes more generally within the phonemic system. This sheds new light on a classic observation by Dixon (1980: 143) regarding geography:

98

that in roughly the eastern third of Australia, languages have just one lateral, while elsewhere the number tends to be higher. Given the context of the data in Figure 10.4 we can see that Dixon’s one-vs.-many geographical observation is caused in part by languages’ phonemic systems and in part by an accident of phylogeography. The eastern region is populated by Pama-Nyungan languages of the Northeastern and South-eastern clades. Only in these genealogical groups do the languages overwhelmingly both lack laminal lateral subsystems (which is not terribly uncommon), and have just one apical place of articulation (which is decidedly uncommon). Thus, the ‘eastern one-lateral’ generalization follows mainly from the fact that the eastern region is occupied by two language clades with rare, oneapical phonemic systems. In terms of lateral-specific parameters, their systems are not particularly remarkable, though the superficial outcome (one-lateral systems in the east) is eye-catching.

Phonemic fricatives are rare in Australian languages, and mostly localized to the genealogically diverse Daly region in the Northern Territory, and to the Paman subgroup of the North-eastern clade of Pama-Nyungan, on the Cape York Peninsula. Unlike stops and nasals, fricatives do not tend to appear at all available places of articulation. In Cape York, the voiced triplet of /β, ð, ɣ/ is found in many Paman languages, particularly in the Northern Paman, Norman Paman, and Alaya-Athima subgroups; in Anguthimri, a palatal fricative also appears, in the set /β, ð, ʝ, ɣ/; in Lamalamic there is no velar fricative, but voiceless /ϕ, θ, ɕ, h/; in Mbara the two fricatives are both laminal /ð, ʝ/; in Kuthant the sole fricative is velar /ɣ/; and the Western Torres language, spoken on islands between the tip of mainland Australia and New Guinea, is the one Australian language that contrasts voicing in its fricatives

segment inventories /s, z/.7 Marrithiyel (Western Daly) has only a labial fricative8 /ϕ/, while elsewhere in the family Emmi has /β, ʐ / and Marringarr and Marti Ke have /β, ʐ , ð, ʝ/. Ngan’gityemerri (Southern Daly) distinguishes /β, z, ʝ, ɣ/. Outside these areal hotspots, Dalabon (Gunwinyguan) appears to contrast /h/; Adnyamathanha (Thura-Yura subgroup of the Western clade of Pama-Nyungan) contrasts /β, ð/; Larrakia (Darwin group) has /β/; and Warluwarra (Warluwarric clade of Pama-Nyungan) has contrastive retroflex and palatal fricatives. All phonemic fricatives in PamaNyungan languages can be traced historically to the phonologization of positional allophones of stops (see Alpher, Chapter 16, this volume), though this is not so in other families.

10.2.2 Glides, laterals, and vibrants The Iwaidjan languages possess a velar continuant, which has been phonemically analysed as either a glide or a fricative. A comparable segment appears in the isolate Tiwi, located offshore to the west of Iwaidjan, and in the small Giimbiyu family to the south of Iwaidjan. Velar glides also appear in Arandic languages (Western clade of PamaNyungan) and the neighbouring Warluwarra (Warluwaric clade of Pama-Nyungan). Contrastive dental glides appear in Bununba (Bunuban), Kurrama and Yindjibarndi (Ngayarta, Western clade of Pama-Nyungan), and Umbuygamu (Lamalamic, North-eastern clade of Pama-Nyungan). The Iwaidjan languages other than Mawng exhibit large liquid inventories, expanded by the inclusion of flapped laterals at the two apical places of articulation, and a second, retroflex vibrant. Over thirty languages in various clades of Pama-Nyungan exhibit extra vibrants, either a contrastive retroflex vibrant, a contrastive tap, a contrastive voiceless vibrant, or a pair of alveolar and retroflex taps. A single, extra retroflex vibrant appears in Ngardi, Ngardily, and Warlpiri (Ngumpin-Yapa subgroup, Western clade), Narrungga (Thura-Yura subgroup, Western clade) and in the Yolngu clade in Ga¨lpu, Dhay’yi, and Djapu. A single additional tap appears in Yarluyandi, Arabana, Wangkangurru, Pitta Pitta, and Wangkayutyuru (Karnic subgroup, Central clade); Malyangapa, Wadikali, and Yardliyawarra (Yardli subgroup, Central clade); Kok Nar, Kurtjar, Walangama, Kuthant (Norman Pama subgroup, North-eastern clade); Yanda and Kukatj (Guwa-Yanda and Mayi subgroups, North-eastern clade), and Muruwari (Central New South Wales subgroup, 7 These fricatives may also have prepalatal realizations, and descend historically from prepalatal stops (Alpher, O’Grady, and Bowern 2008). 8 At the retroflex place of articulation, Marrithiyel has an obstruent phoneme with realizations ranging from stop to affricate, fricative, and glide. At the palatal place, a lenis obstruent can be realized as stop or fricative.

South-eastern clade). A pair of taps appear in Diyari, Thirarri, Karangura, Ngamini, and Pirlatapa (Karnic subgroup, Central clade), and in Nukunu and Adnyamathanha (ThuraYura subgroup, Western clade). Contrastive voiceless trills appear in the Lamalamic languages (North-eastern clade), and Yaygir (Gumbaynggiric subgroup, South-eastern clade).

10.2.3 Additional places of articulation A cluster of languages south of the Gulf of Carpentaria have seven phonetic places of articulation: the familiar labial, dorsal velar, and four coronal places, plus a fronted velar place, however the phonemic analysis of the latter has been a matter of disagreement. In the late 1970s, the fronted velar place was analysed as contrastive in a series of publications, for Garrwa (Garrwan; Furby 1974), Jingulu (a.k.a Jingili, in the Mirndi family; Chadwick 1975), and Yanyuwa (Pama-Nyungan Warluwaric clade; Kirton and Charlie 1978). However, in all cases, fronted velars descend historically from earlier clusters of prepalatal+velar or velar+prepalatal. On the basis of evidence from still-operative morphophonemic alternations, phonotactic patterns, and heterorganic phonetic variants, later works have argued for fronted velars to be analysed as underlyingly heterorganic clusters (Pensalfini 2003: 26–30; Breen 2003: 428; Mushin 2012a: 16–17).9 More than half of the Arandic languages (Western clade of Pama-Nyungan) have been analysed as possessing a prepalatalized apical alveolar place of articulation, though Harvey (2011b) provides arguments supporting a reanalysis in terms of a sequence of /j/ plus a regular alveolar. Labio-velar segments have been proposed for Anindilyakwa (Gunwinyguan; van Egmond 2012), but these contour segments appear to be much like velar+labial sequences in other Australian languages.10

10.2.4 Labialized consonants Contrastively labialized, or rounded, consonants have been proposed in Arandic languages (Central clade of Pama-Nyungan), Anindilyakwa (Gunwinyguan), and Mbabaram (North-eastern clade of Pama-Nyungan). In both 9 As it happens, similar morphophonological combinations of prepalatals and velars are found in nearby languages, where the surface outcome is not a fronted velar, but a regular prepalatal: in the Tangkic language family to the east (Round 2009: 249; 2017b: §1) and Wambaya (Mirndi) to the south (Nordlinger 1998a: 42). Given the areal context, it is conceivable that these too descend from earlier, phonetically fronted-velar realizations of underlying prepalatal+velar clusters. 10 While it is true that they have particular phonotactic properties, this is true of all places of articulation combinations in Australian languages, see Chapter 11, this volume. Separately, see Smith and Johnson (2000: 376) for phonetic labio-velar realizations of cluster-final phonemic labials in Kugu Nganhcara.

99

erich r. round Arandic and Anindilyakwa, the key empirical observation is that certain consonants reliably cause adjacent schwa vowels to surface as [u], even when the vowel in question belongs to a different morpheme. In Arandic the argument is particularly compelling, given that infixal prosodic morphology allows almost any labialized consonant to be moved away from an underlyingly adjacent vowel, in which case, if it is [u] it reverts to schwa, and if the new vowel placed next to the consonant is normally schwa, it becomes [u] (see Wilkins 1989: 92–4 for a concise account of Mparntwe Arrernte; Breen 2001 for a comparative overview). In Anindilyakwa, labialized consonants at the edges of morphs likewise affect vowels across morphological boundaries, though there is no infixing morphology available to diagnose morph-internal consonants (Leeding 1979; van Egmond 2012). Further consideration of Arandic and Anindilyakwa may yield more insights in the future. In both systems, schwa is also palatal-coloured to [i] and in Anindilyakwa there are many additional interactions between individual consonants and vowels, as well as between vowels themselves. The current analyses postulate phonemically labialized consonants in order to account for [u]-colouring of schwa, deriving additional vowel colourings through other mechanisms. Whether this is ultimately the most satisfactory analysis remains a question of interest. The labialized /dʷ, nʷ, ɡʷ/ in Dixon’s (1991b) analysis of Mbabaram are contour segments, similar to what are analysed in nearby languages as consonant clusters.

10.3 Uncommon absences from consonant inventories Around 2% of Australian languages are analysed as having one or more phonemic nasal consonants missing, when compared to the places of articulation at which plosives occur. Absent nasals are all either dental or retroflex, except for the prepalatal nasal missing in Gugu Badhun (Northeastern Pama-Nyungan).11 In all cases, the missing phonemic nasal still occurs phonetically in the language, within phonetically homorganic nasal+plosive clusters, where it is analysed as an allophone of one of the other phonemic nasals. Unlike most of the cases of ‘extra’ segments discussed above (Section 10.2), there is no major areal or genealogical clustering of missing nasals. Both the dental and retroflex nasals are phonemically absent in Marrithiyel (Western Daly); Miriwoong (Jarrakan); Ngarigu (South-eastern Pama-Nyungan); and Gunggari and Guwamu 11 However, Gugu Badhun does contrast /n̪/ and /jn/ in codas, with the latter being the only /j/+apical cluster in the lexicon. Arguably, /jn/ could be phonemicized as contrastive /ȵ/.

100

(North-eastern Pama-Nyungan). The dental is absent in Warndarrang (Maran); Jaminjung (Mirndi); Wemba Wemba (South-eastern Pama-Nyungan); and possibly in Narrungga (Western Pama-Nyungan) and Garlali and Punthamara (Central Pama-Nyungan). The retroflex is absent in Gudanji (Mirndi) and Bidyara and Gungabula (North-eastern PamaNyungan). Around 2% of Australian languages lack the contrast between the rhotic glide and vibrant. These absences are clustered genealogically, and almost all occur in the eastern clades of Pama-Nyungan. In North-eastern PamaNyungan, the contrast is absent in the Paman languages Kugu Nganhcara, Kuugu Ya’u, Ayapathu, Luthigh, Mpalitjanh; the Western Torres language; and in the Maric languages Dharawala, Wadjabangay, and Yiningay. In Southeastern Pama-Nyungan, it is absent in the Bandjalangic varieties; the Kulin Madhi-Madhi varieties; the Upper Murray language Yorta Yorta; and the Yuin language Ngarigu. Outside of eastern Pama-Nyungan, the contrast is absent in Nhanda (Western Pama-Nyungan) and Warray (Gunwinyguan).

10.4 Diversity in two-series stop systems We saw earlier that one of the main parameters of variation in Australian consonant inventories is the presence of one vs. two series of plosives. The phonetic basis of the twoseries contrast is not entirely uniform among the Australian languages which have it. In the Top End, the phonological contrast can correlate with a range of acoustic parameters, primary among which is closure duration, with voicing playing a lesser role (see Evans and Merlan 2004 for a detailed review), while Butcher (1992) has emphasized the aerodynamic correlate of peak intraoral airflow. In Cape York meanwhile, duration is not a factor in the contrast in Umbuygamu (North-eastern clade of Pama-Nyungan), which is one of voicing alone (Verstraete 2018d), and the same may be true for other languages nearby. However, here my focus is not on the phonetics of two-series contrasts in Australia, but on the similarities and considerable diversity of phonotactic behaviour. The two-series stop contrast is frequently subject to positional neutralization (Dixon 2002: 608, 613; Evans and Merlan 2004: 196–7). The phonotactic positions in which neutralization occurs vary across regions and genealogical groups. The contexts examined here are word-initial and word-final; intervocalic; pre-vocalic and post-liquid, post-nasal, postobstruent; and pre-consonantal. For brevity, I distinguish between phonotactic environments in which two stops are contrasted at any place of articulation, vs. environments

segment inventories Initial PN Central PN North-east PN Yolngu PN Western PN Warluwaric PN South-east Western Daly Maningrida Darwin Southern Daly Gunwinyguan Yangmanic Jarrakan Mirndi Giimbiyu

Inter- Postvocalic liquid

Postnasal

Postobstr.

Preglide

Precons.

Final

(n = 4) (n = 12) (n = 7) (n = 8) (n = 2) (n = 1) (n = (n = (n = (n = (n = (n = (n = (n = (n =

2) 4) 2) 1) 8) 1) 2) 1) 3) Stop series contrast contrastive

neutralized

both absent

Figure 10.5 Positional neutralization and maintenance of two-series stop contrasts.

where there is no such contrast; a more exhaustive study would also examine how place of articulation interacts with positional neutralization. Figure 10.5 shows, only for those languages with a stop series contrast, the proportion of languages in genealogical groups which neutralize or maintain the contrast in various environments.12 Unlike in the figures of previous sections, here I show all genealogical groupings, even those represented by just one or two languages. Some basic cross-linguistic patterns are readily apparent. The contrast is always maintained intervocalically. Between a consonant and vowel, the likelihood of the contrast being maintained correlates with the sonority of the preceding consonant, being greatest after liquids, less so after nasals and very rare after obstruents. On a language-internal basis the same is true. A language will maintain the contrast after a lower-sonority consonant class only if it also does so after higher-sonority segment classes; the only exceptions in the lexical dataset are Warluwarra (Pama-Nyungan, Warluwaric clade) and perhaps Thaynakwith13 (Pama-Nyungan, Northeastern clade), which maintain the contrast after nasals but neutralize it after liquids.

12 Data in this section is drawn from a large lexical dataset described in Chapter 11.5. Because it is lexical, the data is liable to include ‘false’ geminates, i.e. sequences of identical stops that occur across morphological boundaries. To control for this, here I exclude languages with fewer than five lexical items containing long/fortis/geminate stops. The method is not fail-safe, but does filter out several languages, all of whose geminates are false in this sense. 13 This is true if orthographic contrasts in Fletcher (2007) are taken at face-value: of around 170 nasal+stop clusters, a half dozen contain voiceless stop graphemes, the others voiced stop graphemes.

Maintenance of the two-series contrast in coda position is very rare word-finally, being found only in three Paman languages: Oykangand, Olgolo, and Ogh Unyjan (Sommer 1969), plus possibly in Ikarranggal14 (Sommer 1999a). And although nearly all two-series languages do permit stops in word-internal codas, the two-series contrast is always neutralized there. In clusters of stop+stop (shown as ‘post-obstruent’ in Figure 10.5), the contrast is almost always neutralized in both stops, with just two exceptions. In dialectally-related Margany and Gunya (Pama-Nyungan North-eastern clade, Maric subgroup), the two-series contrast is maintained in stop+stop clusters only if the first stop is alveolar /d/, after which both voiced and voiceless non-coronal stops can occur, though it should be noted that while the /d/ segment itself generally patterns as a voiced stop, its phonetic realization is a tap (Breen 1981a: 292–3). In Kugu Nganhcara (Pama-Nyungan North-eastern clade, Paman subgroup) stop+stop clusters can be either voiced throughout the cluster or voiceless throughout (Smith and Johnson 2000: 380). Complex onsets are rare in Australian languages, though interestingly, stop-series contrasts are maintained in them preceding a glide, in the Paman (North-western PamanNyungan) languages Aghu Tharnggala, Thaynakwith, and Umbuygamu.15 Two-series contrasts are rarely maintained 14 Sommer (1999a) is not explicit on this point. Examples contain both voiced and voiceless stops in word-final position, but the instrument suffix, for example, appears without comment as both /-ŋk/ and /-ŋɡ/. 15 Larrakia (Darwin group) shows a contrast between a short and long velar stop before a glide, but seemingly only over morphological boundaries: e.g. /kweɻi-kwa/ ‘bamboo’, /kununuk-kwa/ ‘grey hair’, and, apparently across a reduplication boundary: /kwak~kwak-pa/ ‘crow’.

101

erich r. round in word-initial position, but do occur in Emmi, Patjamalh, and Marrithiyel (Western Daly); Murrinhpatha (Southern Daly); one Yolngu (Pama-Nyungan) variety, Djinang; Nhanda (Pama-Nyungan Western clade, Kartu subgroup); Adnyamathanha (Pama-Nyungan Western clade, Thura-Yura subgroup); and within North-eastern Pama-Nyungan: Aghu Tharnggala (Alaya-Athima subgroup), Thaynakwith (Northern Paman subgroup), and the languages of the Lamalamic subgroup.16

10.5 Vowels Over three quarters of Australian languages have vocalic inventories of between three and six vowel phonemes, and 98% have ten or fewer. No Australian language has contrastively unrounded back vowels, and fewer than 1% have contrastively rounded front vowels. Only one Australian language, Anguthimri (Paman subgroup of North-eastern Pama-Nyungan), is reported to have contrastive nasalized vowels (Crowley 1981). No Australian language contrasts vocalic voicing or voice modality. The primary parameters of variation in Australian vocalic systems are the presence or absence of a length contrast; the presence or absence of mid vowels; and the presence or absence of non-low central vowels such as /ə, ɨ/. Figure 10.6 shows the proportions of vocalic systems according to these parameters, within families and Pama-Nyungan subgroups with at least three members. Looking across genealogical groups, the proportion of languages with mid vowels is quite balanced (median 41%, mean 50%), the proportion with a length contrast less so (median 37%, mean 36%), while the proportion with non-low central vowels is low (median zero, mean 10%). Figure 10.7 is an attempt to convey variation in the overall sizes of phonemic vowel systems. Each row presents a genealogical group, and the relative sizes of the circles on a single row indicate the relative commonness of vowel systems of each size, within the genealogical group. Genealogical groups are ordered from top to bottom according to their mean vowel system size. Only five Australia languages contrast ten or more phonemic vowels: Kwini and Wunambal (both Worroran) with eleven; Wik-Ngathan and Mbabaram (Paman subgroup of North-eastern Pama-Nyungan) with twelve; and Anguthimri (also Paman), a distinct outlier with seventeen. Few Australian languages exhibit contrastive rounding (for vowels at the same height and frontness). Rare, front rounded vowels appear in small groups of related languages: the three languages of the Giimbiyu family; Emmi and 16 I am unable to replicate Gasser and Bowern’s (2014) finding, that 41% of languages with the contrast preserve it in word-initial position.

102

Length contrast

PN Yolngu PN Central PN Western PN Warluwaric PN South-east PN North-east

(n = 10) (n = 25) (n = 86) (n = 5) (n = 77) (n = 107)

Tangkic Garrwan Nyulnyulan Iwaidjan Mirndi Darwin Maran Gunwinyguan Western Daly Maningrida Yangmanic Worrorran Giimbiyu

(n = 4) (n = 3) (n =10) (n = 7) (n = 8) (n = 3) (n = 4) (n =15) (n = 7) (n = 4) (n = 3) (n = 8) (n = 3)

Non-low central V

Present

Mid vowels

Absent

Figure 10.6 Main parameters of variation in vowel inventories. Number of phonemic vowels 10 3 4 5 6 7 8 12

17

PN Central (n = 25) PN Warluwaric (n = 5) PN Western (n = 86) PN Yolngu (n = 10) PN South-east (n = 77) PN North-east (n = 107) Garrwan Maran Mirndi Iwaidjan Nyulnyulan Western Daly Maningrida Yangmanic Giimbiyu Darwin Gunwinyguan Tangkic Worrorran

(n = 3) (n = 4) (n = 8) (n = 7) (n = 10) (n = 7) (n = 4) (n = 3) (n = 3) (n = 3) (n = 15) (n = 4) (n = 8) Frequency within group 0.25

0.50

0.75

1.00

Figure 10.7 Frequencies within genealogical groups of vowel system sizes.

Patjamalh (Western Daly); Matngele (Eastern Daly); and ten Paman languages. In languages of the Arandic subgroup of the Western Pama-Nyungan clade, most surface [u] vowels are rounded allophones of /ə/, with the rounding feature associated underlyingly with neighbouring consonants and clusters (see Section 10.2.4). Similarly, many surface [i] vowels are also

segment inventories allophones of /ə/ conditioned by neighbouring consonants. Nevertheless, in most Arandic languages there remains a residue of phonetic [i] vowels, and in some lects also [u], that cannot be accounted for in this way, and so their vowel inventories are not strictly vertical as one might expect, but contain /i, ə, a/, and in some, /u/ (see Breen 2001 for an overview). Australian languages are typically not analysed as having phonemic diphthongs, though see Verstraete (2018d) for a careful argument for phonemic / ı͡a/ and / ʊ͡a/ in Mbarrumbathama (Pama-Nyungan North-eastern clade, Lamalamic subgroup). Nevertheless, tautosyllabic diphthong-like sequences do occur, when vowels appear adjacent to glides. In the lexical dataset described in detail in Chapter 11.5, over half of Australian languages permit coda /j/ and over one third permit coda /w/; around half allow sequences of /j/+vowel following a consonant, and eighty percent allow /w/+vowel following a consonant. Justification for an analysis of these sequences as vowel+glide and glide+vowel comes from the relative lack of restriction on which nuclear vowels appear adjacent to the glides, and in the case of Pama-Nyungan coda /j/, from the fact that in stem-final position, the coda glide behaves like other consonants with respect to phonologically sensitive allomorphy (e.g. Dixon 1977a: 77–8).

10.6 Prospective For two or three decades our empirical, phonological understanding of Australian segment inventories and phonotactics has been at a standstill, after seminal insights by Busby (1980), Dixon (1980), Harvey (1991), and Hamilton (1996b). But new tools for continent-scale phonology are emerging (Gasser and Bowern 2014; Macklin-Cordes and Round 2015), and with them new progress can begin. In this chapter and in Chapters 11 and 12, I have conveyed some of what we can hope to look forward to in Australia and beyond. Here and in Chapter 11 I have attempted to select core topics which are both important to our understanding of continental phonological diversity in Australia, and on which it is possible to shed a few, new rays of light within the space available. For a more complete treatment within a theoretically articulated typological framework, see Round (in prep), also Baker and Harvey (to appear). For more detail on classic phonological topics, see Busby (1980) on segment inventories and read Hamilton (1996b) cover to cover on phonotactics. On phonological systems in general, see other chapters in this volume, also Dixon (1980; 2002), Evans (1995b), Fletcher and Butcher (2014), and Baker

(2014). Given the slow progress in recent decades, even the older references are valuable and largely up-to-date. Dixon (2002) has met with poor reviews generally (Evans 2005; Sutton and Koch 2008) but the phonology chapter is detailed, extensive, and generally valuable, notwithstanding Dixon’s general resistance to sensible reasoning about language genealogy.

10.7 Sources This chapter would not be possible without the enormous contribution of scores of linguists and hundreds of Indigenous language experts over many decades. Sources for phonemic inventories (Round 2019a, b) are listed below. The lexical sources used in Section 10.4 to examine positional neutralization in stop series are described in Chapter 11.5. Pama-Nyungan Arandic Alyawarr (Breen 2001), Antekerrepenhe (Breen 2001), Ayerrerenge (Breen 2001), Central Arrernte (Breen 2001), Eastern Anmatyerre (Green 2010), Eastern Arrernte (Breen 2001), Kaytetye (Turpin and Ross 2012), Lower Southern Aranda (Breen 2001), Western Anmatyerre (Green 2010), Western Arrernte (Breen 2001), Bunganditj Buwandik (Blake 2003a), Warrnambool (Blake 2003b), Central NSW Gamilaraay (Giacon 2014), Muruwari (Oates 1988), Ngiyambaa (Donaldson 1980), Wayilwan (Donaldson 1980), Wiradjuri (Wafer and Lissarrague 2008), Wiriyaraay (Giacon 2014), Yuwaalaraay (Williams 1980), Yuwaliyaay (Williams 1980), Dyirbalic Dyirbal (Dixon 1972), Warrgamay (Dixon 1981), Guwa-Yanda Guwa (Blake and Breen 1990), Yanda (Breen 1990b), Kalkatungic Kalkatungu (Blake 1979a), Yalarnnga (Breen and Blake 2007), Karnic Arabana (Hercus 1994), Diyari (Austin 1981a), Garlali (Holmer 1983), Karangura (Austin 1981a), Kungardutji (McDonald and Wurm 1979), Kungkari (Breen 1990a), Mithaka (Breen 1997), Ngamini (Breen 1997), Nhirrpi (Bowern, Hercus, and Wurm 2002), Pirlatapa (Breen 1997), Pirriya (Breen 1990b), Pitta Pitta (Blake 1979b), Punthamara (Holmer 1983), Thirarri (Austin 1981a), Wangkangurru (Hercus 1994), Wangkayutyuru (Blake 1979b), Wangkumara (McDonald and Wurm 1979), Yandruwandha (Breen 2004a), Yarluyandi (Hercus n.d.-a), Yawarrawarrka (Breen 2004a), Kulin Boonwurrung (Blake 1991), Daungwurrung (Blake 1991), Djabwurung (Blake 2011a), East Djadjawurung (Blake 2011a), Jardwadjali (Blake 2011a), Kolakngat (Blake, Clark, and Reid 1998), Ladji-Ladji (Blake 2011b), Madhi-Madhi (Hercus 1986), Nari Nari (Hercus 1978), Piangil (Blake 2011a), Wathawurrung (Blake, Clark, and Krishna-Pillay 1998), Wathi Wathi (Blake 2011b), Wemba Wemba (Hercus 1986), West Djadjawurung (Blake 2011a), Woiwurrung (Blake 1991),

103

erich r. round Wotjobaluk (Hercus 1986), Yari-Yari (Horgen 2004), Kurnic Bidhawal (Wafer and Lissarrague 2008), Muk-Thang (Fesl 1985), Nulit (Fesl 1985), Thangguai (Fesl 1985), Lower Murray Keramin (Horgen 2004), Ngayawang (Horgen 2004), Ngintait (Horgen 2004), Yaraldi (McDonald 2002), Yitha Yitha (Horgen 2004), Maric Barada (Terrill 1998), Barna (Terrill 1998), Bidyara (Breen 1973), Biri (Terrill 1998), Dharawala (Breen 1990b), Dharumbal (Terrill 2002), Gabalbara (Terrill 1998), Gangulu (Terrill 1998), Ganulu (Terrill 1998), Garingbal (Terrill 1998), Gudjal (Sutton 1973), Gugu Badhun (Sutton 1973), Gungabula (Breen 1973), Gunggari (Holmer 1983), Gunya (Breen 1981a), Guwamu (Austin 1980), Margany (Breen 1981a), Miyan (Terrill 1998), Wadjabangay (Breen 1990b), Wangan (Terrill 1998), Warungu (Tsunoda 2011), Wiri (Terrill 1998), Yambina (Terrill 1998), Yandjibara (Breen 1973), Yangga (Terrill 1998), Yetimarala (Terrill 1998), Yilba (Terrill 1998), Yiningay (Breen 1990b), Yuwi (Terrill 1998), Marrngu-Wati Antakirinya (O’Grady, Voegelin, and Voegelin 1966), Karajarri (McKelson 1989b), Kartujarra (Marsh 1976), Kukatja (Hansen and Hansen 1978), Mangala (Sharp 2004), Manjiljarra (Burgman 2009), Ngaanyatjarra (Glass and Hackett 1970), Ngalia (O’Grady, Voegelin, and Voegelin 1966), Ngarlawangka (O’Grady, Voegelin, and Voegelin 1966), Nyangumarta (O’Grady 1963), Pintupi (Huttar 1976), Pitjantjatjara (Glass and Hackett 1970), Putijarra (Wangka Maya Pilbara Aboriginal Language Centre and Australian Institute of Aboriginal and Torres Strait Islander Studies 2004), Wangkajunga (Jones 2011), Warnman (O’Grady, Voegelin, and Voegelin 1966), Yankunytjatjara (Goddard 1985), Yulparija (Burridge 1996), Mayi Kukatj (Breen 1992), Mayi-Kulan (Breen 1981b), Mayi-Kutuna (Breen 1981b), Mayi-Thakurti (Breen 1981b), Mayi-Yapi (Breen 1981b), Ngawun (Breen 1981b), Wunumara (Breen 1981b), Ngumpin-Yapa Bilinarra (Meakins and Nordlinger 2014), Gurindji (Meakins, McConvell, et al. 2013), Jaru (Tsunoda 1981), Malngin (Ise 1999), Mudburra (Nash et al. 1988), Ngardi (Jagst 1975), Ngardily (Nash 1979a), Ngarinyman (Meakins and Nordlinger 2014), Walmajarri (Hudson 1978), Wanyjirra (Senge 2015), Warlmanpa (Nash 1986), Warlpiri (Nash 1986), Warumungu (Simpson 2002b), Nyawaygic Nyawaygi (Dixon 1983), Wulguru (Donohue 2007), Paman Aghu Tharnggala (Jolly 1989), Agu Aloja (Rigsby 1976), Agwamin (Sutton 1976a), Alngith (Hale 1997b), Angkamuthi (Crowley 1983), Anguthimri (Crowley 1981), Aritinngithigh (Hale 1976d), Atampaya (Crowley 1983), Awu Alaya (Rigsby 1976), Ayapathu (Verstraete and Rigsby 2015), Bakanh (Hamilton 1997b), Djabugay (Patz 1991), Djangun (Patz 2002), Flinders Island (Sutton 1980), Gugu Wakura (Patz 2002), Guugu Yimidhirr (Haviland 1979b), Ikarranggal (Sommer 1999a), Kaantju (Thompson 1988), Kok Nar (Breen 1976), Kokiny, Koko Bera (Sommer 1999b), Koko

104

Dhawa (Sommer 1999b), Kugu Nganhcara (Smith and Johnson 2000), Kuku Yalanji (Patz 2002), Kurtjar (Black and Gilbert 1996), Kuthant (Black 1980), Kuugu Ya’u (Thompson 1976a), Lamalama (Verstraete 2018d), Linngithigh (Hale 1997b), Luthigh (Hale 1976d), Malthanmungu (Sutton and Haviland 1995), Mbabaram (Dixon 1991b), Mbara (Sutton 1976a), Mbiywom (Hale 1976d), Mpalitjanh (Hale 1976d), Muluridji (Patz 2002), Ndra’ngith (Hale 1976d), Ngkoth (Hale 1976d), Ogh Alungul, Ogh Angkula (Rigsby 1976), Ogh Awarrangg (Sommer 1969), Ogh Unyjan (Sommer 1969), Olkol (Sommer 1969), Oykangand (Sommer 1969), Rimanggudinhma (Godman 1993), Tableland Lamalama (Verstraete 2018d), Takalak (Sutton 1976a), Thaayorre (Gaby 2017), Thaynakwithi (Fletcher 2007), Umbuygamu (Verstraete 2017b), Umpila (O’Grady 1990), Uradhi (Hale 1976c), Wagaman (Patz 2002), Walangama (Black 1980), Wik Mungkan (Sayers 1976a), Wik-Ngathan (Sutton 1995b), Yadhaykenu (Crowley 1983), Yidiny (Dixon 1977a), Yintyingka (Verstraete and Rigsby 2015), Yinwum (Hale 1976d), Yir Yoront (Alpher 1991), Pilbara Jiwarli (Austin 2006), Kariyarra (Wangka Maya Pilbara Aboriginal Language Centre 2001a), Kurrama (Wangka Maya Pilbara Aboriginal Language Centre 2001b), Martuthunira (Dench 1995), Ngarla (Westerlund 2015), Ngarluma (Kohn 2011), Nhuwala (O’Grady, Voegelin, and Voegelin 1966), Nyamal (Burgman 2007b), Nyiyaparli (O’Grady, Voegelin, and Voegelin 1966), Palyku (O’Grady and Laughren 1997), Panyjima (Dench 1991b), Payungu (Wangka Maya Pilbara Aboriginal Language Centre 2008a), Purduna (Burgman 2007a), Thalanyji (Wangka Maya Pilbara Aboriginal Language Centre 2008b), Tharrkari (Austin 1992b), Thiin (Austin 2015), Warriyangga (Austin 2015), Yindjibarndi (Wordick 1982), Yinhawangka (Wangka Maya Pilbara Aboriginal Language Centre 2008c), Qld-NSW coast Baanbay (Hoddinott 1967), Barunggam (Holmer 1983), Bigambal (Wafer and Lissarrague 2008), Bundjalung (Sharpe 2005), Butchulla (Bell 2003), Duungidjawu (Kite and Wurm 2004), Gabi-Gabi (Holmer 1983), Gidabal (Geytenbeek and Geytenbeek 1971), Gumbaynggir (Eades 1979), Gureng-gureng (Holmer 1983), Guwar (Jefferies 2011), Guweng (Holmer 1983), Jandai (Jefferies 2011), Minjangbal (Sharpe 2005), Turubul (Jefferies 2011), Waalubal (Crowley 1978), Waka Waka (Holmer 1983), Wuli Wuli (Holmer 1983), Yagara (Jefferies 2011), Yaygir (Crowley 1979), Yugambal (Wafer and Lissarrague 2008), South-west Badimaya (Dunn 1988), Balardung (Douglas 1976), Bibbulman (Douglas 1976), Galaagu (O’Grady, Voegelin, and Voegelin 1966), Goreng (Douglas 1976), Kalaamaya (O’Grady, Voegelin, and Voegelin 1966), Kaniyang (Douglas 1976), Malkana (Gargett 2011), Minang (Douglas 1976), Mirniny (O’Grady, Voegelin, and Voegelin 1966), Ngadjunmaya (O’Grady, Voegelin, and Voegelin 1966), Nhanda (Blevins 2001a), Pinjarup (Douglas 1976), Wajuk

segment inventories (Douglas 1976), Wardandi (Douglas 1976), Watjarri (Mackman 2012), Wiilman (Douglas 1976), Wudjari (Douglas 1976), Yingkarta (Dench 1998c), Yuwat (Douglas 1976), ThuraYura Adnyamathanha (Schebeck 1974), Kaurna (Amery and Simpson 2013), Kukata (Platt 1972), Narrungga (Eira 2010), Nukunu (Hercus 1992a), Parnkalla (O’Grady 2001), Wirangu (Hercus 1999), Upper Murray Dhudhuroa (Blake and Reid 2002), Pallanganmiddang (Blake and Reid 1999), Yabula-Yabula (Bowe and Morey 1999), Yorta Yorta (Hercus 1986), Warluwaric Bularnu (Breen 1988), Eastern Wakaya (Breen 1974), Warluwarra (Breen 2015), Western Wakaya (Breen 1974), Yanyuwa (Kirton and Charlie 1978), Western Torres Kala Kawaw Ya (Hunter, Bowern, and Round 2011), Kala Lagaw Ya (Hunter, Bowern, and Round 2011), Yardli-Paakantyi Bandjigali (Hercus 1982), Malyangapa (Hercus 1989), Southern Paakintyi (Hercus 1982), Wadikali (Hercus and Austin 2004), Yardliyawarra (Hercus and Austin 2004), Yolngu Dhangu (Wood 1978), Dhay’yi (Wood 1978), Djambarrpuyngu (Wilkinson 1991), Djapu (Morphy 1983), Djinang (Waters 1980a), Djinba (Waters and Horn 1989), Gumatj (Wilkinson 1991), Gupapuyngu (Wilkinson 1991), Nhangu (Wilkinson 1991), Ritharrngu (Heath 1980a), Yuin-Kuri Awabakal (Lissarrague 2006), Darkinyung (Jones 2008), Dharawal (Besold 2012), Dharuk (Troy 1992), Dharumba (Besold 2012), Dhurga (Besold 2012), Djirringany (Besold 2012), Eora (Troy 1992), Gundungurra (Besold 2012), Katthang (Lissarrague 2010), Nganyaywana (Crowley 1976), Ngarigu (Hercus 1986), Ngunawal (Besold 2012), Thanggati (Lissarrague 2007), Thawa (Besold 2012) Other families Bunuban Bunuba (Rumsey 2000), Gooniyandi (McGregor 1990), Darwin Larrakia (Capell 1984), Limilngan (Harvey 2001), Umbugarla (Davies 1989), Eastern Daly Matngele (Zandvoort 1999), Garrwan Garrwa (Mushin 2012a), Kunindirri (Breen 2003), Waanyi (Breen 2003), Giimbiyu Erre (Campbell 2006), Mengerrdji (Campbell 2006), Urningangg (Campbell 2006), Gunwinyguan Anindilyakwa (van Egmond 2012), Dalabon (Evans, Merlan, and Tukumba 2004), Gaagudju (Harvey 2002), Gun-Dedjnjenghmi (Evans 2003a), Gun-Djeihmi (Evans 2003a), Jawoyn (Merlan and Jacq 2005), Kune (Evans 2003a), Kuninjku (Evans 2003a), Kunwinjku (Evans 2003a), Mayali (Evans 2003a), Ngalakgan (Baker 2008b), Ngandi (Heath 1978b), Rembarrnga (McKay 2011b), Warray (Harvey

1986), Wubuy (Heath 1984), Iwaidjan Amurdak (Handelsmann 1991), Garig (Evans 1988b), Ilgar (Evans 1988b), Iwaidja (Evans 1988b), Marrgu (Evans 1988b), Mawng (Singer 2006a), Wurrugu (Evans 1988b), Jarrakan Kija (Taylor and Taylor 1971), Miriwoong (Kofod 1978), Kungarakany Kungarakany (Parish 1983), Maningrida Burarra (Green 1987), Gurr-Goni (Green 1995), Nakara (Eather 1990), Ndjébbana (McKay 2000), Maran Alawa (Sharpe 1972), Mangarrayi (Merlan 1989), Marra (Heath 1981a), Warndarrang (Heath 1980b), Mirndi Binbinka (Chadwick 1978), Gudanji (Aguas 1968), Jaminjung (Schultze-Berndt 2000), Jingulu (Pensalfini 2003), Ngaliwurru (Schultze-Berndt 2000), Ngarnka (Osgarby 2014), Nungali (Bolt, Hoddinott, and Kofod 1971), Wambaya (Nordlinger 1998a), Minkin Minkin (Evans 1990), Nyulnyulan Bardi (Bowern 2012a), Jabirr-Jabirr (McGregor 1996c), Jawi (McGregor 1996c), Djukun (Hosokawa 1991), Ngumbarl (McGregor 1996c), Nimanburu (McGregor 1996c), Nyikina (Stokes 1982), Nyulnyul (McGregor 1996c), Warrwa (McGregor 1994c), Yawuru (Hosokawa 1991), Southern Daly Murrinhpatha (Mansfield 2014b), Ngan’gityemerri (Reid 1990), Tangkic Gangalidda (Round 2014a), Kayardild (Round 2013), Lardil (Round 2011), Yangkaralda (Cook 2017), Tiwi Tiwi (Osborne 1974), Western Daly Emmi (Ford 1998), Malak-Malak (Birk 1975), Marramaninyshi (Tryon 1974), Marringarr (Nambatu et al. 2009), Marrithiyel (Green 1989), Marti Ke (Nambatu et al. 2009), Patjtjamalh (Ford 1990), Worrorran Gambera (Capell and Coate 1984), Kwini (McGregor 1993), Ngarinyin (Rumsey 1982b), Umiida (Capell and Coate 1984), Unggumi (Capell and Coate 1984), Worrorra (Clendon 2014), Wunambal (Carr 2000), Yawijibaya (Capell and Coate 1984), Yangmanic Wagiman (Cook 1987), Wardaman (Merlan 1994), Yangman (Merlan 1994).

Acknowledgements Support for this research is gratefully acknowledged from the Australian Research Council (grant DE150101024) and British Academy (grant VF1_101602) as well as the University of Queensland, Surrey Morphology Group and the Max Planck Institute for the Science of Human History. For helpful comments on the chapter, thanks to Claire Bowern.

105

chapter 11

Phonotactics Erich R. Round

As new tools for large-scale phonological typology emerge, it has become possible to examine and understand continent-level variation among Australian languages in more detail than previously, and thus to add insightful nuance to the frequent characterization of Australian phonologies as strikingly uniform (Busby 1980; 2002; Dixon 1980; Evans 1995b; Hamilton 1996b; Butcher 2006; Fletcher and Butcher 2014; Baker 2014). In this chapter I examine linear segmental phonotactics, and in the previous chapter segment inventories. In both, the starting point is classic observations about Australian similarity and diversity, but the primary aim is to move beyond these. To approach that goal, I observe variation in terms of major genealogical groupings of languages, comparing variation among them and within them, drawing on a large empirical dataset covering 300,000 lexical entries from over 250 language varieties. These are the kinds of tools and methods which will increasingly dominate the practice of typology and the theoretical disciplines that rely on its findings in this century (Gasser and Bowern 2014; Macklin-Cordes and Round 2015). For Australian languages, they reveal a new degree of precision about phonological diversity, and in some cases qualitatively new insights. Below, Section 11.1 introduces the main characteristics and parameters of variation in consonant phonotactics. Section 11.2 examines vowel phonotactics, specifically statistical harmony between vowels in adjacent syllables. Section 11.3 addresses issues at the intersection of segment inventories and phonotactics, namely contour segments such as prestopped nasals. Section 11.4 concludes. The preparation and sources of the dataset are described in Section 11.5.

11.1 Consonant phonotactics: main parameters of variation Here I introduce the classic accounts of Australian consonant phonotactics by Dixon (1980) and Hamilton (1996b) in Section 11.1.1, and by Harvey (1991) on glottal phonotactics in Section 11.1.2. In Section 11.1.3 I examine the variation

and uniformity across the continent in superlaryngeal cluster phonotactics.

11.1.1 Dixon’s and Hamilton’s generalizations Dixon’s (1980: 159ff.) highly influential description of widespread Australian phonotactic patterns was based around a prototypical disyllabic word, as presented in (1). (1) Prototypical phonotactic profile, after Dixon (1980) A. C1 V1 C2 C3 V2 (C4 ) B. C1 V1 C5 V2 (C4 ) Dixon observed that in most Australian languages, all words begin with a consonant and are minimally disyllabic, or at least bimoraic. Typically, all of a language’s consonant phonemes can occur intervocalically, in Dixon’s C5 position. In contrast, positions C1 –C4 admit only subsets of the consonant inventory, and the subset permitted at C1 typically has little or even no overlap with the subset permitted at C4 ; a similar relationship holds between C2 and C3 . A consequence of this is that an adequate characterization of Australian phonotactics cannot be formulated in terms of single syllables, without attending to the consonant clusters created across syllable boundaries (Dixon 1980: 159). Dixon draws attention to two strong tendencies in the composition of clusters. Firstly, in terms of manner of articulation, clusters tend to begin with high sonority segments at C2 and end with low sonority segments at C3 . This is cross-linguistically unremarkable (Vennemann 1988; Clements 1990). What is striking is a second tendency, with respect to place of articulation. Homorganic clusters are common, but so too are heterorganic clusters, whose initial segements are most likely to be apicals, less likely to be laminals, yet less likely to be dorsals, and least likely to be labials, while their final segments show the reverse preferences. For example, np is common but mt is rare. Triconsonantal clusters, Dixon observes, typically consist of a liquid followed by a homorganic nasal+stop sequence at a non-coronal place of articulation, such as lŋk. At word edges, position C1 shows similar preferences in terms of manner and place to position C3 , while

Erich R. Round, Phonotactics. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Erich R. Round (2023). DOI: 10.1093/oso/9780198824978.003.0011

106

phonotactics the preferences in position C4 mirror those in C2 . Dixon emphasizes that these generalizations are encountered in every Australian language. See Section 11.1.3 below for how starkly true this is for word-medial clusters. Dixon’s observations were elaborated and validated against a dataset of over eighty languages in a seminal PhD thesis by Philip Hamilton (1996b).1 Working with more data than was available to Dixon in 1980, Hamilton updates the generalizations about clusters to the form shown in (2) (where ≻ indicates ‘is more harmonic than’), and adds detail regarding the tendencies governing homorganic clusters and positional neutralizations within the apical and laminal classes. Hamilton compares the Australian generalizations to generalizations outside of Australia and considers their possible phonetic motivations, concluding that perceptual cue preservation is particularly important (see also Butcher 2006 regarding place of articulation). (2) Preferred manners in clusters and word edges, after Hamilton (1996b: 110, 154, 228) A. Cluster-initial: Liquid ≻ Nasal ≻ Obstruent, Glide B. Word-final: Liquid, Nasal ≻ Obstruent, Glide C. Cluster-final: Obstruent ≻ Nasal, Glide ≻ Liquid D. Word-initial: Obstruent ≻ Nasal, Glide ≻ Liquid (3) Preferred superlaryngeal places in clusters and word edges, after Hamilton (1996b: 110, 154, 228) A. Cluster-initial: Apical ≻ Laminal ≻ Dorsal ≻ Labial D. Word-final: Apical ≻ Laminal ≻ Dorsal ≻ Labial B. Cluster-final: Labial ≻ Dorsal ≻ Laminal ≻ Apical C. Word-initial: Labial, Dorsal ≻ Laminal ≻ Apical The preferences in (2) and (3) construe generalizations about Australian consonant cluster phonotactics in terms of what occurs in certain positions. Another approach is to focus on permitted transitions, or sequencing, from one consonant to the next, which enables us to express simple generalizations that hold beyond just the initial and final consonants in clusters. The generalization for sequencing of place is (4a). For manner, we can distinguish two variant generalizations, (4b, c), which differ in their treatment of sub-sequences of nasals and obstruents. In Australian languages, most sub-sequences of nasals and obstruents follow the order nasal > obstruent, however many languages also allow obstruent > nasal, including in heterorganic sequences. In Section 11.1.3, I examine the empirical strength of these three generalizations in detail. 1 Hamilton (1996b: 213–33) finds that generalizations about word edges are weaker than those for clusters. Gasser and Bowern (2014) stress that some aspects of Dixon’s generalizations for word edges do not hold as not categorical statements.

(4) Sequencing generalizations The preferred linear order of superlaryngeal consonants in clusters is: a. Place: Apicals > Laminals > Dorsals > Labials b. Manner (strict): Glide > Liquid > Nasals > Obstruents > Glide c. Manner (lax): Glide > Liquid > Nasals, Obstruents > Glide

11.1.2 Laryngeal consonants The sequencing generalizations in (4) pertain to superlaryngeal consonants. The phonotactics of glottals is distinctively different. As seen in Chapter 10 (this volume) glottal segments are genealogically restricted in Australian languages. Phonemic glottal fricatives are even more restricted, reported only in Dalabon (Gunwinyguan; Evans et al. 2008) and in the Lamalamic subgroup (Paman Nyungan, Northeastern clade; Verstraete 2017a; 2018d). In Dalabon, the glottal fricative occurs only in coda-final position, including word-finally.2 The Lamalamic glottal fricatives occur only in onset-initial position, including word-initially and before a following glide. Glottal stops occur in the Gunwinyguan, Maran, Yangmanic, and Maningrida families of the Top End, and in the Yolngu clade of Pama-Nyungan also located in the Top End; outside the Top End, glottals are found only in the Paman subgroup of the North-eastern clade of PamaNyungan, and in Nhanda in the Western clade.3 Harvey (1991) is the classic reference on glottal stops in languages of the Top End. Phonotactically, glottals in these languages can appear only syllable-finally, and can only follow sonorants, not stops. Glottal stops in Rembarrnga (Gunwinyguan), Wagiman (Yangmanic), and the Yolngu languages will move relative to other segmental material in order to maintain their syllable-final position, for example in Rembarrnga /mopalʔ/ ‘knee’ ~ /mopalŋʔke/ ‘your knee’ (McKay 2011b: 38–9). In languages away from the Top End, the phonotactics of glottal stops is more varied. Phonotactic variation can be studied using the lexical dataset mentioned at the start of the chapter (see also Section 11.5). Looking outside of the Top End, the lexical dataset contains eighteen Paman languages and Nhanda, which have glottal stops. Among these, glottal stops can appear word-initially in seven (in 2 Evans et al. (2008) report two contractions in which /h/ can appear nonfinally in a syllable: [kahlŋ] < /kah-yeleng-/ ‘then he/she/it’ and [kahnda ~ kanhda] < /kahnɨnda/ ‘demonstrative’. Possibly, these should be explained as casual/rapid speech forms, which can exhibit distinctive phonotactics in many languages. 3 In Dhanggati (South-eastern clade), Lissarrague (2007: 4) reports that a glottal stop occurs in ‘a couple of Dhanggati words (and is probably an alternative pronunciation for ng)’.

107

erich r. round Umbuygamu, Ayapathu, Bakanh, Ndra’ngith, Thaynakwith, Linngithigh, and Kaantju) and word-finally in eight (in Wik Mungkan, Wik-Ngathan, Bakanh, Umpila, Ndra’ngith, Thaynakwith, Linngithigh, and Yir Yoront). In all eighteen Paman languages and Nhanda, the glottal stop is permitted intervocalically. In all of the languages bar Anguthimri, it is permitted post-consonantally, though only in Kuku Ya’u can it appear after an obstruent. Over half of the languages permit glottals preconsonantally, though only half of those permit glottals before obstruents (Wik Mungkan, Wik-Ngathan in just one root, Kuugu Ya’u, Thaynakwith, and Linngithigh). Across the continent then, glottals appear most often between vowels and sonorant consonants. In cases where they can be bordered by obstruents, they are very unlikely to be tolerated in post-obstruent position.

11.1.3 Superlaryngeal clusters across the continent

To understand phonotactic variation at a continental level, it is revealing to study it with respect to genealogically coherent groups. This allows us to view group-internal variation as well as variation across groups. Since the PamaNyungan family is very large, I provide separate figures for six principal divisions within it (after Bouckaert et al. 2018): the large Western, Central, North-eastern, and Southeastern branches, and smaller Yolngu and Warluwaric.4 It should be borne in mind, that while the Pama-Nyungan groups will each have around the same time depth, the depth of the other families will vary to an extent that is not well understood, and they may be either deeper or shallower than the Pama-Nyungan groups. In this section I examine properties of families and major Pama-Nyungan subgroups that have three or more members in the lexical dataset. Figures are presented as boxplots. Since these may be unfamiliar to some readers, a quick description of how to read them follows. The boxplot for each genealogical group depicts the variation among the languages within that group, for some quantity of interest. For example, Figure 11.1 depicts the proportion of intervocalic consonantal strings which are consonant clusters, as opposed to singletons. The plots are read as follows. A relevant scale appears at the bottom of the figure; for example, for proportions, it covers some amount of ground between zero and one. For each genealogical group, the white box spans the values attached to the middle 50% of languages in the group, known as the inter-quartile range (IQR). The group’s name appears at the left, along with its number of member languages in the dataset. Inside each box, a vertical bar indicates the median value for the group, and the ‘whiskers’ that

The aim of this section is to gain a sharper quantitative view of superlaryngeal consonant cluster phonotactics, by examining lexical frequencies of various cluster types. The main focus here will be on evaluating the strength, in frequency terms, of the consonant cluster sequencing generalizations in (4). As we examine the data, some standard methodological implications of corpus-based research will need to be kept in mind. When quantifying the frequency of cluster types, answers are inevitably shaped by decisions about how segments are labelled, especially contour segments (are they two segments or one?) and fortis or long consonants (likewise, two segments or one?). Here, because the primary interest is the sequencing of place and manner, I count PN South-east (n = 47) Warluwaric (n = 5) contour consonants as two segments, since they typically PN PN Western (n = 62) PN Central (n = 16) contain a sequence of different manners of articulation; I (n = 72) count long consonants as one, since they merely repeat the PN North-east PN Yolngu (n = 9) same place and manner. All decisions about how to count Maningrida (n = 4) Mirndi (n = 5) the data affect the results. One consequence of the particuDaly (n = 3) lar decisions made here, is that for languages where contour Western Worrorran (n = 6) consonants are usually treated in the literature as monoNyulnyulan (n = 8) Giimbiyu (n = 3) phonemic, my evaluation of cluster length may appear relaMaran (n = 4) tively high (since I treat them as sequences, hence clusters). Tangkic (n = 4) Conversely, for languages where long consonants have been Gunwinyguan (n = 10) Iwaidjan (n = 3) treated as diphonemic geminates, my evaluation of clus0.0 0.2 0.4 0.6 ter length may appear relatively low. Importantly though, these decisions are applied uniformly across the languages Figure 11.1 Proportion of intervocalic strings that are clusexamined, so that the comparison is of like with like. Beters (not singletons). cause the underlying data is digital, future studies could easily make alternative decisions and recalculate the results, 4 I depart from Bouckaert et al. (2018) and follow Blake (1990a), Evans an approach to typology that I have advocated elsewhere (1995c), and Round (2017b) in classifying Tangkic as non-Pama-Nyungan. See further Bowern (Chapter 7, this volume). (Round 2017a).

108

phonotactics extend out either side of the box show the remaining extent Warluwaric clades, languages overwhelmingly have short of the variation in the data, out to a maximum distance of clusters, while the Central, North-eastern, and Yolngu clades 1.5 times the IQR from the median; if there are any yet more are dominated by languages with some of the highest proextreme languages in the group (termed ‘outliers’), they are portions of long clusters. In the Central and North-eastern plotted as dots. clades, this high complexity of clusters is driven in part by We begin by establishing some context, regarding clus- large numbers of homorganic stop+sonorant sequences, on ter size and homorganicity in general. Figure 11.1 provides which see Section 11.4. a bird’s eye view of the rate at which intervocalic conAustralian languages are distinctive for their tolerance of sonant strings are clusters rather than singletons, across heterorganic clusters, such as /nk/, both phonemically and most of the genealogical groups of Australia, and gives an in connected speech (Baker 2014: 142; Fletcher and Butcher idea of the variation within them. Broadly, it is a picture 2014: 106). Figure 11.3 provides a quantitative view of hetof variation rather than homogeneity. For example, even erorganicity in clusters across the continent. Because longer the Yolngu language with the lowest proportion of clusters clusters have more capacity to be heterorganic simply by still has a higher proportion than any language in the Man- virtue of containing more segments, and because differingrida, Worrorran, and Western Daly families. On the other ent genealogical groups differ in their cluster complexity, hand, when a language family is large enough, then its lan- it could be misleading to examine heterorganicity without guages tend to span most of the observed range of values. For controlling for cluster length. Accordingly, Figure 11.3 is example, Pama-Nyungan languages represent both the low- based on intervocalic di-clusters only, and shows the variest values (South-eastern Pama-Nyungan) and the highest ation in the proportion that are homorganic, as opposed to values (North-eastern Pama-Nyungan) in the figure, indicat- heterorganic.5 Again, there is considerable variation among ing that given sufficient time depth, even a single language genealogical groups, and inside of the larger groups. The exfamily can come to include member languages that cover treme outlier, visible as a dot in the Pama-Nyungan Western most of the typological space that characterizes all families clade, is Tharrgari, in which homorganic nasal+stop clusters on the continent. The processes by which this occurs, and have recently became fortis stops (Austin 1981a), stripping their ultimate causes remain to be clarified. it of most of its erstwhile homorganic clusters. Figure 11.2 shows variation in the proportion of intervocalic clusters (not singletons) that are longer than PN Yolngu (n = 9) Warluwaric (n = 5) di-clusters. Genealogical groups divide roughly between a PN PN South-east (n = 47) majority in which clusters of three or more consonants are PN Western (n = 62) (n = 72) rare, totalling fewer than 5% of all clusters in all languages PN North-east PN Central (n = 16) of the group, vs. a minority of the groups in which long Gunwinyguan (n = 10) clusters are decidedly more common, at least in some memIwaidjan (n = 3) Maningrida (n = 4) ber languages. The major clades of Pama-Nyungan are split Giimbiyu (n = 3) near the two extremes: in the Western, South-eastern, and Maran (n = 4)

PN Warluwaric PN Western PN South-east PN North-east PN Central PN Yolngu

Nyulnyulan Tangkic Mirndi Western Daly Worrorran

(n = 5) (n = 62) (n = 47) (n = 72) (n = 16) (n = 9)

(n = 8) (n = 4) (n = 5) (n = 3) (n = 6) 0.0

0.25

0.50

0.75

1.00

Figure 11.3 Proportion of intervocalic di-clusters that are homorganic.

Western Daly (n = 3) Giimbiyu (n = 3) Tangkic (n = 4) Mirndi (n = 5) Worrorran (n = 6) Nyulnyulan (n = 8) Maningrida (n = 4) Maran (n = 4) Iwaidjan (n = 3) Gunwinyguan (n = 10) 0.0

0.2

0.4

0.6

Figure 11.2 Proportion of intervocalic clusters that contain three or more consonants.

The next set of figures focusses on the consonant cluster sequencing generalizations of (4), which pertain to the ordering of superlaryngeal consonants. Figure 11.4 shows the proportion of clusters which obey the sequencing generalization with respect to place of articulation, such that the sequences of superlaryngeal places of articulation within a cluster are apical > laminal > dorsal > labial. For this analysis, I have set aside glides that may appear at the edges of 5

In all calculations, /w/ is taken to be labial.

109

erich r. round PN Yolngu PN South-east PN North-east PN Warluwaric PN Western PN Central

PN Central PN Yolngu PN South-east PN North-east PN Western PN Warluwaric

(n = 9) (n = 47) (n = 72) (n = 5) (n = 62) (n = 16)

(n = 16) (n = 9) (n = 47) (n = 72) (n = 62) (n = 5)

Gunwinyguan (n = 10) Western Daly (n = 3) Maran (n = 4) Mirndi (n = 6) Worrorran (n = 4) Nyulnyulan (n = 8) Iwaidjan (n = 3) Mirndi (n = 5) Giimbiyu (n = 3) Tangkic (n = 4)

Gunwinyguan (n = 10) Maran (n = 4) Iwaidjan (n = 3) Maningrida (n = 4) Mirndi (n = 5) Western Daly (n = 3) Nyulnyulan (n = 8) Worrorran (n = 6) Giimbiyu (n = 3) Tangkic (n = 4) 0.5

0.6

0.7

0.8

0.9

1.0

Figure 11.4 Proportion of clusters obeying the place of articulation sequencing generalization.

0.5

0.6

0.7

0.8

0.9

1.0

Figure 11.5 Proportion of clusters obeying the manner sequencing generalization (strict version).

PN Central (n = 16) clusters, since they not infrequently violate the usual place PN Yolngu (n = 9) of articulation sequencing generalization; a fuller study PN Warluwaric (n = 5) PN South-east (n = 47) would examine them also. PN North-east (n = 72) At the most extreme in Figure 11.4 is the small Tangkic PN Western (n = 62) family, in which every cluster obeys the place of articulation Gunwinyguan (n = 10) Nyulnyulan (n = 8) sequencing generalization. However, even the most liberal Maningrida (n = 4) genealogical group, Gunwinyguan, still has a median rate Iwaidjan (n = 3) Maran (n = 4) near 90%, and most groups have medians above 95%. Given Worrorran (n = 6) that these figures pertain to a constraint on the sequenc- Western Daly (n = 3) ing of places of articulation that is unreported outside of Giimbiyu (n = 3) Tangkic (n = 3) Australia, the facts summarized in Figure 11.4 present an unMirndi (n = 5) common challenge to any account of worldwide variation 0.5 0.6 0.7 0.8 0.9 1.0 in phonotactics.6 For more on place of articulation in clusters, see also the discussion of assimilation alternations in Figure 11.6 Proportion of clusters obeying the manner seChapter 12 of this volume. quencing generalization (lax version). Figure 11.5 shows the proportion of clusters which obey the strict sequencing generalization with respect to manner of articulation, so that sequences are glide > liquid > nasal > regular, and most groups come close to having all clusters in obstruent > glide. Again, most groups have medians above all languages obeying the generalization. The Central clade 95%. The Gunwinyguan family is one of the exceptions. of Pama-Nyungan remains distinctive, and the Western and The other exceptions are major clades of Pama-Nyungan: North-eastern clades continue to include many outliers. most prominently, the Central clade whose IQR stretches These departures from the lax manner sequencing generalalmost as low as 80%, and the South-eastern, North-eastern, ization are caused by homorganic stop+lateral clusters, on and Western clades which contain many outliers with low which see Section 11.4. As mentioned at the outset of Section 11.1, the main values. To understand these exceptional groups better, properties of Australian consonant cluster phonotactics we can begin by considering Figure 11.6, which shows the and their ubiquity, cross-linguistically, across the continent proportion of clusters obeying the lax manner sequencing have been known for decades. Figures 11.4–11.6 though, generalization, which allows sequences of stop > nasal. reveal how ubiquitous they are also within Australian lanIn Figure 11.6, Gunwinyguan now appears typologically guage lexicons. To a limited degree, this extreme level of 6 Hamilton (1996b) attempts an account, but since it is couched only in entrenchment in every Australian language offers some exterms of language universals, it is not capable of explaining why the phe- planation for how the cross-linguistic picture could mainnomenon occurs in Australia and not elsewhere. Future accounts in terms tain itself: in order for a language to deviate, it would of diachrony will need to contend with the fact that sound change in Australia appears to maintain these unusual phonotactic properties, which is need to undergo massive, internal lexical change. Neverthenot so on other continents (Miceli and Round 2014). less, it remains to be explained how Australian languages

110

phonotactics could have reached this point, and why massive phonotactic changes to clusters in individual languages evidently don’t occur, despite the fact that some changes—such as the creation of new cluster types following the loss of vowels— are entirely conceivable and readily attested on other continents.

11.2 Statistical harmony of vowels in adjacent syllables

corresponding to chance; a value of +1 is twice the chance rate, +2 is four times the chance rate, +3 is eight times, and so forth; a value of –1 is half the chance rate, –2 is one quarter the chance rate, etc. Our interest here is in the broad pattern across the continent as a whole. Since we will not be concerned here with comparisons at the specific level of individual families and groups, this time I include all data, even from very small groups and isolates. The results are shown in Figure 11.7, and while they are not entirely surprising given previous reports, the near-ubiquity of statistical front–back harmony is striking nonetheless. All genealogical groups have higher than chance levels of harmony as the median among their members, and very few individual languages have a rate below the chance level.

In several Australian languages it has been noted that in adjacent syllables, identical vowels occur at a greater than chance rate, particularly if the vowels are non-low: see Haviland (1979b: 39) on Guugu Yimidhirr, Dixon (1980: 179) on PN North-east (n = 72) Dyirbal and Yidiny, and Breen (1981b: 28) on Ngawun (all PN South-east (n = 47) PN Western (n = 62) Pama-Nyungan, North-eastern clade); McGregor (1990: 88– PN Central (n = 16) 9) on Gooniyandi (Bunuban); Round (2009: 90) on Kayardild PN Yolngu (n = 9) (Tangkic); Bowern (2012a: 98–100) on Bardi (Nyulnyulan); PN Warluwaric (n = 5) Fitzgerald (1997: 288) on 22 Pama-Nyungan languages; and Iwaidjan (n = 3) Nash (2017a) on 21 Pama-Nyungan and non-Pama-Nyungan Giimbiyu (n = 3) languages. Here I examine all languages in the Australian Western Daly (n = 3) Tiwi (n = 1) lexical dataset. There are several ways one could frame Eastern Daly (n = 1) a question about vowel quality in adjacent syllables. In Southern Daly (n = 1) Section 11.2.1 and Section 11.2.2 I examine adjacent, non- Kungarakany (n = 1) Tangkic (n = 4) low vowels in the first two syllables of lexical items and Gunwinyguan (n = 10) their statistical front–back harmony, that is, the rate at Worrorran (n = 6) which they match one another in frontness/backness. In Maningrida (n = 4) Darwin (n = 2) Section 11.2.3 I examine a second pattern, in which adjacent Jarrakan (n = 2) syllables both contain mid vowels at a higher than chance Garrwan (n = 2) rate. Yangmanic (n = 2)

11.2.1 Statistical front–back harmony in non-low vowels To examine statistical front–back harmony, our interest is in the first two syllables of lexical items, and the extent to which a pair of non-low vowels exhibits front–back harmony at a rate above chance.7 This is conveniently measured using an observed to expected ratio, which I find is most easily interpreted when expressed as log-odds. Here I use base-2 log-odds, which for practical purposes means that a value of zero corresponds to harmony at a rate exactly 7 By ‘at chance’, I mean that if /i/ occurs in 40% of first syllables and 30% of second syllables, then by chance, in 40% × 30% = 12% of cases it would appear in both syllables. In reality, the occurrence of vowels in syllables one and two may not be independent, and so the actual rate can be higher or lower. For example, if /i/ occurs in 40% of first syllables and 30% of second syllables, but it only occurs in syllable two if it also occurs in syllable one, then /i/ would occur in both syllables a full 30% of the time, much higher than the ‘at chance’ 12%.

Maran Mirndi Nyulnyulan Bunuban

(n = 4) (n = 5) (n = 8) (n = 2) –1.0

–0.5

0.0

0.5

1.0

Figure 11.7 Statistical front–back harmony in non-low vowels.

11.2.2 The *iCu constraint on non-low vowels A high rate of statistical front–back harmony, as measured by an observed to expected ratio, follows not only from a high incidence of front–front and back–back sequences, but also low incidences of front–back and back– front. Nash (1986: 73) noted that although Warlpiri has back–front sequences, it systematically lacks front–back sequences in intramorphemic underlying forms (except when the intervening consonant is /p/ or /w/), a constraint

111

erich r. round which has subsequently been referred to as the *iCu constraint. And Hercus (1994: 54) reports the near absence of front–back sequences in Arabana and Wangkangurru. Given that we established widespread statistical front–back harmony in Section 11.2.1, we can now assess the relative contributions to it from the rarity of front–back sequences (*iCu) vs. the rarity of back–front sequences (*uCi) in the lexicons of Australian languages, by comparing the difference between observed to expected ratios for front–back and back–front strings. Figure 11.8 shows that difference, expressed as a difference in log odds: this means that a value of +1 corresponds to a scenario in which, for example, that the log odds of iCu-avoidance is 1.6 and the log odds of uCi-avoidance is one point lower, at 0.6. Consequently, positive values tell us that statistical front–back harmony is driven more by iCu-avoidance, and negative values that it is driven by uCi-avoidance. The results, in Figure 11.8, show that across the continent, high vowel harmony is more likely to reflect a strong *iCu dispreference than *uCi. Why this should be so remains to be explained by future research. PN Yolngu PN Western PN North-east PN South-east PN Warluwaric PN Central

(n = 9) (n = 62) (n = 72) (n = 47) (n = 5) (n = 16)

PN North-east (n = 72) PN South-east (n = 47) Kungarakany (n = 1) Maningrida (n = 4) Worrorran (n = 6) Western Daly (n = 3) Gunwinyguan (n = 10) Eastern Daly (n = 1) Yangmanic (n = 2) Giimbiyu (n = 3) Southern Daly (n = 1) Darwin (n = 2) Nyulnyulan (n = 8) Maran (n = 4) Iwaidjan (n = 3)

Iwaidjan (n = 3) Kungarakany (n = 1) Yangmanic (n = 2) Jarrakan (n = 2) Southern Daly (n = 1) Tangkic (n = 4) Gunwinyguan (n = 10) Giimbiyu (n = 3) Mirndi (n = 5) Eastern Daly (n = 1) Worrorran (n = 6) Tiwi (n = 1) Nyulnyulan (n = 8) Maran (n = 4) Maningrida (n = 4) Bunuban (n = 2) Western Daly (n = 3) Darwin (n = 2) Garrwan (n = 2)

0

1

2

3

4

Figure 11.9 Statistical metaphony.

–2

0

2

4

Figure 11.8 Difference between *iCu and *uCi (positive = *iCu is stronger).

11.2.3 Statistical mid-vowel metaphony Smith and Johnson (2000: 381) use the term ‘metaphony’ to refer to a phonotactic constraint on the vowels of adjacent syllables in Kugu Nganhcara (Pama-Nyungan, North-eastern

112

clade), according to which the mid vowel /o/ may co-occur only with other mid vowels in monomorphemic roots. Here I use ‘statistical metaphony’ to refer to the tendency for mid vowels to occur adjacent to one another, and not adjacent to either high or low vowels, within the first two syllables of a lexical item. To examine whether statistical metaphony occurs at a greater than chance level, I use the same metric as for statistical front–back harmony above (Section 11.2.1): an observed to expected ratio expressed as base-2 log-odds. Because not all Australian languages have mid vowels, the results in this section are based on a smaller pool of data than the front–back harmony results above (Section 11.2.1), however they are still striking, as shown in Figure 11.9. Abovechance rates for statistical metaphony (scores of greater than zero in Figure 11.9) are just as nearly ubiquitous as above-chance rates were for statistical front–back harmony in Section 11.2.2. The highest rates for statistical metaphony are also markedly higher than for statistical front–back harmony: a score of +4 in Figure 11.9, for example, corresponds to metaphony occurring at sixteen times the rate expected by chance.8

11.3 Australian contour segments: at the edge of segment and sequence Any survey of segment inventories and phonotactics will of necessity be influenced by decisions about how to count segments. This is particularly acute for contour segments, such as prenasalized stops, which some linguists analyse as single segments, and others as sequences. As Hyman (2017: 144) 8 Though it should be borne in mind that such high numbers are only possible if the ‘at chance’ rate itself is very low. This is more likely for mid vowels (Section 11.2.3), which are relatively rare, than for non-low front vowels (Section 11.2.1).

phonotactics has observed, we want to typologize languages, not linguists. Discrepancies in how different linguists categorize a phenomenon often arise, not because of any deficiency in the phonological analysis, but because neither of the traditional options available (e.g. segment vs. sequence) fully suits the empirical facts. This is a quandary that arises in all areas of typology. In response, typologists are increasingly turning to multivariate methods, in which a phenomenon is not merely shoehorned into a category A vs. B (segment vs. sequence), but is assessed along multiple dimensions as to its conformity with, or departure from, canonical instances of A and B (Round and Corbett 2020). Here I consider phenomena which exhibit precisely this kind of behaviour with respect to being single segments vs. sequences of segments. Although such phenomena are typically bundled into the category of ‘contour segment’, this is a disparate, catch-all category which conceals multiple axes of empirical variation, as the discussion below reveals. The backdrop for an assessment of contour segments in Australian languages consists of the common patterns in segment inventories and phonotactic patterns introduced above in Section 11.1, with the following points being particularly salient. Typically in Australian languages, words begin with single consonants (Dixon 1980: 159). When words do begin with a phonetic sequence of consonants such as [mb], that phonetic sequence resembles a canonical single segment in terms of its word-initial position. Consequently, there is a motivation to phonemicize it as a monophonemic ‘prenasalized stop’ contour segment /ᵐb/. Similarly, in Australian consonant clusters, stops typically appear only after, and not before, nasals and laterals (Section 11.1.3). Some languages however do permit phonetic stop+nasal or stop+lateral sequences. Because these sequences can behave phonotactically within clusters much like a single segment, there is a motivation to phonemicize them as single contour segments termed ‘pre-stopped nasals/laterals’. Likewise, phonetic apical stop+trill sequences have been treated as single contour segments, ‘trill-released stops’. One constraint on the positing of contour segments, following a long linguistic tradition (e.g. Trubetzkoy 1969; see also Devine 1971) and found throughout the Australianist literature, has been that the phonetic sequences in monophonemic contour segments must be homorganic. In Section 11.1.3, when I evaluated the proportions of clusters that conform to the consonant cluster manner sequencing generalizations, I treated all of the abovementioned ‘contour segments’ as sequences rather than unitary segments. This has the advantage of ensuring that potential counterexamples to Australian generalizations about sonority sequencing are not hidden inside a single segment, and ensuring that this hiding-away is not idiosyncratically dependent on whether a linguist decided

to posit a monophonemic contour segment or not. However, it has the disadvantage that it fails to distinguish those sequences from heterorganic sequences such as [tm], [kl], which are analysed as monophonemic in no Australian language. In this section, I focus on phonetic sequences of nasal+stop, stop+nasal, and stop+lateral. Keeping homorganic and heterorganic sequences distinct, I examine their presence, absence, and diversity of phonotactic distributions. The results contribute to a more nuanced understanding of the phonotactics of Australian languages, and also to an understanding of the kinds of empirical facts that can repeatedly lead well qualified linguists to make differing decisions from one another.

11.3.1 Initial nasal+stop Phonetic nasal+stop sequences occur in word-initial position in a few groups of Australian languages: the Maningrida family, some Maran languages, and several subgroups of Pama-Nyungan. Heterorganic, initial nasal+stop sequences are rare, occurring only in Njebbana and Nakkara (Maningrida), in fewer than a dozen attested roots of Kurtjar (Pama-Nyungan North-eastern clade, Norman Pama subgroup of Paman), and in a single recorded root in Aghu Tharnngala (Alaya-Athima subgroup of Paman) /nkiə-/ ‘sit’ (Jolly 1989: 41) and Yanyuwa (Pama-Nyungan Warluwaric clade) /npa-/ ‘fall’ (Kirton 1967: 19).9 In the lexical dataset, all other word-initial phonetic nasal+stop sequences in Australian languages are homorganic. The following paragraphs outline how these phonetic sequences have been analysed, and note any significant additional empirical details. In the Maningrida languages Njebbana, Nakkara, Burarra, and Gurr-Goni, the nasals in word-initial nasal+stop clusters are analysed as syllabic consonants (Glasgow 1981b: 77; Green 1995: 25–6; McKay 2000: 177; Eather 2011: 65–8; see also Carew and Beltran, Chapter 70, this volume). It can be noted though, that the syllable projected by these nasals is never stressed, and in Nakkara is explicitly analysed as extrametrical by Eather (2011: 36). Eather (2011: 65) comments that ‘[t]here is no straightforward way of proving their status as syllabic consonants, as against, say, the first segment of a consonant cluster’. In all four Maningrida languages, the word-initial nasal+stop sequences can occur across a prefix– stem boundary (the prefix being just the nasal), which typically would militate against a linguist analysing them as part 9 Two other Yanyuwa roots are phonetically homorganic though phonologically analysed as heterorganic: nykalanykarri- and nykarri- ‘listen’ (Kirton 1967: 19). If the second part of Eastern Anmatyerre ngkwerlp rnpernp ‘rock pituri’ (Green 2010) is considered word-initial, it constitutes another instance, however phonetically it will always be preceded by a vowel, owing to rules of epenthesis.

113

erich r. round of a single phonemic contour segment. Only a few stems begin with nasal+stop in Njebbana, Nakkara, and Burrara, and none in Gurr-Goni. The Maran languages Marra (Heath 1981a) and Alawa (Sharpe 1972) allow phonetic homorganic nasal+stop sequences in word-initial position. In Marra they are analysed as clusters by Heath (1981a: 12), but in Alawa as monophonemic prenasalized stops by Sharpe (1972: 16). In the other Maran languages, Warndarang has stems that begin with nasal+stop, but the nasal deletes if word-initial (Heath 1980b: 19–20); Mangarrayi has neither word-initial nor stem-initial clusters (Merlan 1989: 188, 195). In Pama-Nyungan, word-initial nasal+stop clusters occur in several individual languages and subgroups. In Paman (North-eastern clade), they occur in most languages of the Alaya-Athima and Lamalamic subgroups, in Anguthimri and Linngithigh (Northern Paman), Kok Nar and Kurtjar (Norman Paman), and Mbabaram (Southern Paman). Their phonemic analysis varies. Dixon (1991b: 354), for example, phonemicizes them as homorganic clusters in Mbabaram, while Crowley (1981: 152–5) and Verstraete (2018d: 4–6) posit monophonemic prenasalized stops in Anguthimri and Mbarrumbathama (Lamalamic). Both Dixon (1991b: 354) and Verstraete (2018d: 6) note that the alternative analysis would be possible. In Kugu Ya’u (Middle Paman subgroup), certain words are realized with initial /ŋk/ only when encliticized to a preceding word; otherwise they begin with /ŋVŋk/ (Thompson 1976b: 223–4). Elsewhere in Pama-Nyungan, word-initial homorganic nasal+stop sequences appear in the Arandic languages (Western clade), Kalkatungu (North-eastern clade) and in a single root in Pitta Pitta (Central clade). In Arandic they are analysed monophonemically (Breen 2001). In Kalkatungu (Blake 1979a: 10) and Pitta Pitta (Blake 1979b: 190) they are analysed as clusters. In Yanyuwa (Warluwaric clade) word-initial /nt, np, ɲk/ occur (/ɲk/ is phonetically homorganic, at a fronted velar place of articulation, cf. Chapter 10.2.3). These have been analysed both as clusters (Kirton 1967: 19) and monophonemically (Kirton and Charlie 1978: 188–90; Tabain and Butcher 1999).

11.3.2 Stop+lateral In the past decade, phoneticians and phonologists have become increasingly aware of the phenomenon of allophonic prestopping of lateral segments in Australian languages (see e.g. Round 2014b; Harvey et al. 2015). This section is not about allophonic prestopping, but about languages in which phonetic stop+lateral sequences are in contrast with plain lateral segments. I begin with a coverage of heterorganic stop+lateral sequences.

114

As discussed in Sections 11.1.1 and 11.1.3, stop+lateral is a highly disfavoured sequence in all Australian language families. Nevertheless, the lexical dataset shows that a tolerance of stop+lateral clusters is not uncommon across reduplication boundaries, occuring in Jaminjung (Mirndi); Gurr-goni and Burarra (Maningrida); Mawng (Iwaidjan); Rembarrnga, Jawoyn, Ngandi, Ngalakgan, and Dalabon (Gunwinyguan); Wagiman and Wardaman (Yangmanic); Nyulnyul (Nyulnyulan); and in Pama-Nyungan: Wik Mungkan (Northeastern clade, Middle Paman subgroup); Wemba Wemba, Wotjobaluk, Wathawurrung, Woiwurrung, and Djabwurung (South-eastern clade, Kulin subgroup); and in the Western clade: Yulparija, Nyangumarta and Karajarri (Wati–Marrngu subgroup), Warumungu (Warumungic), and Jaru and Walmajarri (Ngumpin-Yapa). Stop+lateral is tolerated to a lesser extent across other morphological boundaries, but occurs in some Gunwinyguan, Nyulnyulan, and Pama-Nyungan languages, although by its nature, a lexical dataset, which does not cover inflection morphology, may well underrepresent this. Just a few languages appear to permit heterorganic stop+lateral in stems, and only in a handful at most: Wardaman (Yangmanic), Linngithigh, and Umbuygamu (Northeastern clade of Pama-Nyungan). Word-initial stop+lateral (and other stop+liquid) sequences, including heterorganic sequences, have been reported in a few Australian languages, though in some instances the apparent clusters have later been confirmed to be sequences of stop–vowel–liquid in careful speech (see e.g. discussions of the phonetics of stop–vowel–liquid sequences, in which the vowel may be very short, resulting in the percept of a cluster: in Watson 1944 cited by Jefferies 2011; Blake 2003a: 28–9; Round 2009: 70–1). In South-eastern Pama-Nyungan, true initial stop+liquid clusters may have been common in Yaraldi (Lower Murray subgroup) and in the Kurnai (Victoria) subgroup, though in the latter case many of the original records analysed by Fesl (1985) do contain intervening vowels. In the languages mentioned just above, there is no indication that homorganic stop+lateral sequences are favoured over heterorganic, either in terms of phonotactic distribution or numerical prevalence. Standing in contrast, however, is the small and genealogically restricted set of languages in which homorganic stop+lateral sequences either far outnumber heterorganic stop+lateral sequences, or are the only stop+lateral sequences found. Languages in the lexical dataset with more than ten items containing contrastive homorganic stop+lateral sequences are Larrakia (Darwin group) and in Pama-Nyungan: Adnyamathanha and Narrungga (Western clade, Thura-Yura subgroup), and Yandruwandha, Kungardutji and Arabana (Central clade, Karnic subgroup). A comment is required regarding Kungardutji and Arabana. McDonald and Wurm’s

phonotactics (1979) study of dialectally related Wangkumarra makes no mention of prestopped laterals, but in Schebeck’s (n.d.) Kungardutji wordlist they appear to be contrastive. Hercus (1994: 37–43) analyses prestopping in Arabana as allophonic, but the presence of lexical exceptions and lexical optionality in the complex contextual rules means that in classic phonemic terms, the stop+lateral sequences are contrastive, even if their distribution is predictable to a large extent (see also Harvey et al. 2019). Homorganic stop+lateral sequences have been analysed as monosegmental in Yandruwandha (Breen 2004a: 16), but as clusters in Narrungga (Eira 2010: 25–6, 32), Adnyamathanha (Schebeck 1974: xvi), and Larrakia (Capell 1984: 56). Given this albeit small set of six languages, we can now ask in what ways and to what extents their stop+lateral sequences resemble single segments or clusters. Taking a multivariate approach, we do not necessarily expect the answers to be in full accord, as we consider these phonetic sequences in terms of various, differing dimensions of their variation. Because stop+lateral sequences have been phonemicized as prestopped laterals (and not laterally-released stops), I will compare them to plain lateral segments when asking how they resemble single segments. The comparisons to be made will be in terms of occurrence and relative frequencies in word-initial and word-final positions, intervocalically, and bordered by consonants. Data is drawn from the lexical dataset. In word-initial position, of the six languages, only Larrakia and Arabana permit plain lateral segments to appear. None of the six languages allows phonetic stop+lateral sequences word-initially, or any other word-initial clusters. So, it can be said that in word-initial position, homorganic stop+lateral sequences are similar to clusters (in being prohibited) in all six languages. They are similar to plain laterals (in being prohibited) in four languages but dissimilar to plain laterals (in not being permitted) in two. In word-final position, only Larrakia permits plain laterals. Larrakia also permits word-final clusters and a word-final stop+lateral sequence is found in one root, /-wətl/ ‘body’ (Harvey 2004: 1). In word-final position then, stop+lateral sequences pattern exactly parallel to both plain laterals and clusters in either being permitted (in Larrakia) or not (the other five languages). In intervocalic position, all six languages allow plain laterals, stop+lateral sequences and other clusters, so that the stop+lateral sequences again pattern just like both plain laterals and clusters.10 In all six languages, between 20%–55% of plain laterals appear in preconsonantal position, yet in none of them do stop+lateral sequences appear preconsonantally. In this 10 Breen (2004a: 25) remarks that in Yandrawandha, a rare allophone of the prestopped lateral, ‘noted on some occasions with the word padla “ground, sand”, is a well-defined stop followed by a syllabic lateral, [padlˌa]’.

sense, stop+lateral sequences are clearly dissimilar to simple laterals, but rather similar to many consonantal diclusters, which cannot combine with other consonants into larger clusters. In none of the languages do plain laterals or stop+lateral sequences appear post-consonantally. There are three languages with contrastive homorganic stop+lateral sequences which do not appear in the lexical dataset: Yawarrawarrka (Breen 2004a, b), Kaurna (Amery and Simpson 2013), and Parnkalla (O’Grady 2001). In each of these, the stop+lateral sequences appear only intervocalically, whereas plain laterals also appear preconsonantally in all three, and word-finally in Parnkalla. In sum, homorganic stop+lateral sequences pattern similarly to both plain laterals and other clusters in wordinitial position (by being absent) in most languages, though in languages that permit word-initial plain laterals, the stop+lateral sequences pattern differently from them (by being absent). In word-final position, stop+lateral sequences pattern similarly to both plain laterals and clusters (all are present word-finally in Larrakia, and none are in most other languages) or similarly only to clusters (in Parnkalla, by being absent). Intervocalically, all are found in all languages. Preconsonantally, stop+lateral sequences pattern distinctly differently from plain laterals (by being absent), but rather like many clusters. If we return to considering languages that permit heterorganic stop+lateral sequences, which are always analysed as clusters, then in those languages, the homorganic stop+lateral sequences pattern just like the heterorganic sequences, that is, like clusters. Overall then, in terms of their phonotactics, homorganic stop+lateral sequences are more like canonical clusters than canonical single segments, though in most ways they are similar to both. What is distinctive about them is that in a small and genealogically restricted set of languages, homorganic stop+lateral sequences are highly frequent, while heterorganic stop+lateral sequences are rare if not entirely absent. Taking a diachronic view, this can primarily be attributed to their historical origins, as intervocalic (and most commonly, post-tonic) allophones of plain laterals that have become phonologized. However, what remains to be clarified is whether it is more than coincidence, that when those allophones phonologized, they tended to do so in a set of phonotactic positions that matches the positions already occupied by clusters, rather than the full set of positions occupied by single laterals.

11.3.3 Stop+nasal A similar analysis can be applied to stop+nasal sequences. The results will be subtly different from those for stop+laterals.

115

erich r. round We saw in Sections 11.1.1 and 11.1.3 that stop+nasal sequences, while not especially preferred in clusters, are nevertheless tolerated in languages of most Australian families; where they occur, they typically can be heterorganic or homorganic. However, Pama-Nyungan is something of an outlier. Only in the Yolngu clade, in the Ngumpin subgroup of the Western clade, and the Kulin and Bungandidj subgroups of the South-eastern clades are stop+nasal clusters tolerated like they are in most Australian families. Elsewhere in Pama-Nyungan, stop+nasal is either prohibited outright or allowed only when homorganic. Meanwhile, all of the Australian languages in which homorganic stop+nasal sequences are especially frequent are Pama-Nyungan. Here I define ‘especially frequent’ as languages in the lexical dataset with at least ten items containing homorganic stop+nasal sequences, and in which homorganic stop+nasal sequences outnumber heterorganic by at least two to one. The eighteen languages that meet these criteria are found in the Karnic subgroup (Central clade), in the Alaya-Athima subgroup and in nearby Oykangand and Olgol (Northeastern clade, Paman sub-clade), in Kulin (South-eastern clade), Thura-Yura and Arandic (Western clade), and the Djinang language (Yolngu clade). For most of these languages, the homorganic stop+nasal sequences are analysed as monophonemic prestopped nasals, though not in Djinang (Waters 1980a: 4). See Sommer (1969: 59–60) for an early, explicit discussion of the viability of both analyses within several Paman languages. In the multivariate analysis that follows, because stop+nasal sequences have been phonemicized as prestopped nasals (and not post-nasalized stops), I will compare them to plain nasal segments when asking how they resemble single segments. As in Section 11.3.2, the comparisons to be made will be in terms of their occurrence and relative frequencies in word-initial and word-final positions, intervocalically, and bordered by consonants. In word-initial position, all 18 languages permit plain nasals, yet only in languages which permit other consonant clusters word-initially does one find word-initial stop+nasal sequences: in Alyawarr, Central Arrernte, and Antekerrepenhe.11 In word-final position, there is diversity across the subgroups. In the Karnic, Thura-Yura, and Arandic languages (Central and Western clades), none of the plain nasal, stop+nasal sequences or other clusters appears wordfinally,12 in which case the stop+nasal sequences behave like both plain nasals and other clusters, by being absent. In Djinang and the Paman languages (North-eastern 11 In Antekerrepenhe, in just one item (in Breen 2006a): the verb /tnəm/ [tnəmə] ‘to stand’. 12 Here I assume a ‘vowel-final’ analysis of the Arandic languages, see e.g. Breen (2001) on the debate over the analysis of phonetically present, but non-contrastive, word-final vowels in Arandic.

116

and Yolngu clades) all are permitted word-finally: plain nasals, stop+nasal sequences, and other clusters, so again, stop+nasal sequences behave like both plain nasals and other clusters, this time by being permitted. In Kulin (Southeastern clade), represented by Wathawurrung and West Djadjawurung, final nasals and clusters are permitted, but final stop+nasal sequences are prohibited, in which case stop+nasal sequences behave neither like the plain nasals nor like other clusters—though since not all clusters can appear word-finally, it could be said they behave like some clusters and not others. In intervocalic position, all 18 languages allow plain nasals, stop+nasal sequences, and other clusters. In the Thura-Yura and Karnic languages, this is the only position where stop+nasal sequences occur (aside from in one foreign place name in Arabana), meaning that the overall assessment of stop+nasal sequences in those languages will be largely parallel to the assessment of stop+lateral sequences in Section 11.3.2: they pattern more like clusters than like single segments. However, the Thura-Yura and Karnic languages are the exceptions, not the rule. The other 14 languages in the sample are phonotactically more permissive. In preconsonantal position, stop+nasal sequences are permitted rather often. In Arandic, they occur before stops and nasals in Alyawarr, and before stops in Eastern Anmatyerre. In seven of the eight Paman languages, they appear before stops; and in Djinang, before stops and nasals. These are positions where plain nasals also appear and where many clusters cannot, thus in these 10 languages, stop+nasal sequences in preconsonantal position behave more similarly to single segments than to clusters. In the other eight languages, they behave unlike single segments in this respect, and unlike most other clusters. In post-consonantal position, stop+nasal sequences appear even more often: in the eight Paman languages after glides and liquids; in the two Kulin languages after liquids; and in the four Arandic languages after liquids, as well as after the palatal glide in Alyawarr. Moreover, in most of these languages, stop+nasal sequences can appear in interconsonantal position, very much like single nasal segments.13 In summary, in Thura-Yura and Karnic, the behaviour of homorganic stop+nasal sequences resembles that of stop+lateral sequences, which pattern largely like clusters. In contrast, in Arandic, Paman, Kulin, and Djinang, homorganic stop+nasal behaves phonotactically like a single nasal segment. In this section on contour segments, I hope to have made three contributions: one methodological, one theoretical, and one empirical. Methodologically, it will be apparent 13 Though Waters (1980a: 23) reports that in Djinang, when the nasal in a stop+nasal sequence is not bordered by a vowel, it becomes syllabic.

phonotactics in hindsight that had I summarized Australian languages merely in terms of the individual analyses accorded to them in the literature (cluster vs. monophonemic segment), the result would have been not only uninformative but misleading, since the empirical variation does not align with variation in linguists’ decisions about whether or not to posit monophonemic contour segments. A multivariate approach to typologizing is more appropriate, and this will be true for research on segment inventories outside of Australia just as it is inside Australia: merely aggregating phoneme charts may not be the optimal path to true insights. On the theoretical front, we see that any theory which allows the analyst only a binary choice between ‘segment’ and ‘cluster’ is quite possibly not providing adequate tools for representing the full range of empirical variation that exists. And empirically, this short study has revealed real variation in how Australian languages organize homorganic sequences of nasal+stop, stop+lateral, and stop+nasal, which in turn provides an impetus for further investigation. For example, stop+lateral was found to behave more like a cluster than stop+nasal, but then again, in the Thura-Yura and Karnic subgroups, stop+nasal and stop+lateral were more similar to one another—and it is predominantly from those subgroups that all our information about stop+lateral sequences comes, raising the question of whether the crucial difference is not stop+nasal vs. stop+lateral, but rather Thura-Yura and Karnic (and their historical antecedents) vs. other groups.

11.4 Prospective For two or three decades our empirical, phonological understanding of Australian segment inventories and phonotactics has been at a standstill, after seminal insights by Busby (1980), Dixon (1980), Harvey (1991), and Hamilton (1996b). But new tools for continent-scale phonology are emerging (Gasser and Bowern 2014; Macklin-Cordes and Round 2015; Macklin-Cordes, Bowern, and Round 2021), and with them new progress can begin. In this chapter, I have conveyed some of what we can hope to look forward to in Australia and beyond. Still, within a chapter of this size, it has been necessary to be selective. I have attempted to select core topics which are both important to our understanding of continental phonological diversity in Australia, and on which it is possible to shed a few, new rays of light within the space available. For a more complete treatment within a theoretically articulated typological framework, see Round (in prep), also Baker and Harvey (to appear). For more detail on classic phonological topics, see Busby (1980) on segment inventories and read Hamilton (1996b) cover to cover on phonotactics. On phonological systems in general, see

other chapters in this volume, also Dixon (1980; 2002), Evans (1995b), Fletcher and Butcher (2014), and Baker (2014). Given the slow progress in recent decades, even the older references are valuable and largely up-to-date. Dixon (2002) has met with poor reviews generally (Evans 2005; Sutton and Koch 2008) but the phonology chapter is detailed, extensive, and generally valuable, notwithstanding Dixon’s general resistance to sensible reasoning about language genealogy. See Gasser and Bowern (2014) for a study employing a similar approach to the one employed in this chapter,14 examining phoneme frequencies, word-initial phonotactics, and word length, and which serves as a useful check on claims by Dixon (2002: 547–55).15

11.5 Sources This chapter would not be possible without the enormous contribution of scores of linguists and hundreds of Indigenous language experts over many decades. The following describes how the lexical dataset was assembled. Lexical materials were assembled into a digital orthographic form, with over half imported from Bowern’s (2016a) Chirila database, and then converted into a phonemicized representation in a careful process aimed at ensuring accuracy and fidelity to original scholarship. Each language was phonemicized individually and a battery of consistency checks caught errors at various stages. Infelicities were corrected both manually and through semiautomated means, including in hundreds of individual cases with manual reference to original and alternative sources. A further layer of analysis handles contextual neutralization among other issues, in a consistent manner across language varieties. All steps in the process were recorded for full transparency. For further details see Round and Bowern (in prep.). A large electronic database has many advantages, but disadvantages can arise from errors that are still present in the data, and it is inevitable that at this stage, some do remain to be identified and corrected. For such reasons, the results reported here may be subject to minor revisions in the future, though by the same token, owing to the many rounds of error-catching already applied, major changes to results should be unlikely. Not all Australian languages are equally well described. In terms of raw size, lexical datasets vary by a factor of 14 A fore-runner to the methods used here, Gasser and Bowern (2014) examine 120 languages. Compared to the lexical dataset used here, its phonological data are derived by a more direct conversion from raw orthography (see Chapter 10, Section 10.7), which may yield some differences in results compared to the current approach. 15 Though note that Dixon (2002: 547–658) does later qualify some of the simple statements he makes early in his chapter, and which are criticized by Gasser and Bowern (2014).

117

erich r. round thirty: the smallest used here is 150 entries, the largest is over 8,000. Many Australian language varieties, particularly in the eastern parts of the continent, were documented only prior to the advent of modern linguistic methods. Results pertaining to those languages, in the South-eastern and North-eastern clades of the Pama-Nyungan family, unavoidably come with a corresponding degree of uncertainty. I have not covered Tasmanian languages here, about which little is known securely (Bowern 2012c). Sources are the following. Where possible, I cite the ultimate source of lexical datasets that I retrieved in orthographic form from Chirila (Bowern 2016a). Pama-Nyungan Arandic Alyawarr (Breen 1990a), Antekerrepenhe (Breen 2006a), Central Arrernte (Wilkins 1989), Eastern Anmatyerre (Bowern 2016a, currently unreleased dataset), Western Anmatyerre (Green et al. 2003), Western Arrernte (Breen 2000), Bunganditj Buwandik (Blake 2003a), Warrnambool (Blake 2003b), Central NSW Gamilaraay (Ash, Giacon, and Lissarrague 2003), Muruwari (Oates 1992), Ngiyambaa (Donaldson 1997), Wayilwan (Giacon n.d.), Wiradjuri (Hosking and McNicol 1993), Yuwaalaraay (Ash, Giacon, and Lissarrague 2003), Yuwaliyaay (Ash, Giacon, and Lissarrague 2003), Dyirbalic Dyirbal (Dixon 1972), Warrgamay (Dixon 1981), Guwa-Yanda Guwa (Blake and Breen 1990), Kalkatungic Kalkatungu (Blake 1990b), Yalarnnga (Breen and Blake n.d.), Karnic Arabana (Bowern 2016a, currently unreleased dataset), Diyari (Austin 1981a), Kungardutji (Schebeck n.d.), Kungkari (Breen 1990a), Mithaka (Breen 1968), Ngamini (Breen 1975), Nhirrpi (Bowern 1999), Pirriya (Breen 1990b), Pitta Pitta (Blake 1990c), Punthamara (Holmer 1988), Thirarri (Austin 1981a), Wangkumara (Robertson 1985), Yandruwandha (Bowern 2016a dataset: breyandr), Yarluyandi (Hercus n.d.-a), Kulin Djabwurung (Blake 2011a), East Djadjawurung (Blake 2011a), Jardwadjali (Blake 2011a), Kolakngat (Blake, Clark, and Reid 1998), Ladji-Ladji (Blake 2011b), Madhi-Madhi (Blake 2011b), Piangil (Blake 2011a), Wathawurrung (Blake, Clark, and Krishna-Pillay 1998), Wathi Wathi (Blake 2011b), Wemba Wemba (Hercus 1992b), West Djadjawurung (Blake 2011a), Woiwurrung (Blake 1991), Wotjobaluk (Hercus n.d.-b), Kurnic Muk-Thang (Fesl 1985), Lower Murray Yaraldi (Johnson 1989), Maric Bidyara (Breen 1973), Biri (Terrill 1999), Dharawala (Breen 1990b), Dharumbal (Terrill 2002), Gangulu (Terrill 1999), Gudjal (Santo 2006), Gugu Badhun (Sutton 1973), Gunggari (Holmer 1988), Gunya (Breen 1981a), Guwamu (Austin 1980), Margany (Breen 1981a), Warungu (Bowern 2016a, currently unreleased dataset), Wiri (Terrill 1999), Yambina (Terrill 1999), Yilba (Terrill 1999), Marrngu-Wati Karajarri (McKelson 1989b), Kartujarra (O’Grady 1988a), Kukatja (Peile 1993), Mangala (McKelson 1989a), Ngaanyatjarra (Glass 1988), Ngalia (Spehn-Jackson

118

2013), Nyangumarta (Geytenbeek, Geytenbeek, and Wangka Maya Pilbara Aboriginal Language Centre 1991), Pintupi (Hansen and Hansen 1992), Putijarra (Wangka Maya Pilbara Aboriginal Language Centre and Australian Institute of Aboriginal and Torres Strait Islander Studies 2004), Wangka (Blyth 2001), Warnman (Wangka Maya Pilbara Aboriginal Language Centre n.d.), Yulparija (McKelson 1989c), Mayi Kukatj (Breen 1991), Mayi-Kulan (Breen 1981b), MayiKutuna (Breen 1981b), Mayi-Thakurti (Breen 1981b), MayiYapi (Breen 1981b), Ngawun (Breen 1981b), Wunumara (Breen 1981b), Ngumpin-Yapa Bilinarra (Meakins, Campbell, et al. 2013), Gurindji (Meakins, McConvell, et al. 2013), Jaru (Tsunoda 1981), Malngin (Ise 1999), Mudburra (Nash et al. 1988), Ngardily (Green 1988), Ngarinyman (Jones 2005), Walmajarri (Hudson and Richards 1993), Wanyjirra (Senge 2015), Warlmanpa (Nash, Hale, and Breen 1984), Warlpiri (Schwartz 1996), Warumungu (Bowern 2016a, currently unreleased dataset), Nyawaygic Nyawaygi (Dixon 1983), Wulguru (Donohue 2007), Paman Aghu Tharnggala (Jolly 1989), Agu Aloja (Sommer n.d.-a), Angkamuthi (Crowley 1983), Anguthimri (Crowley 1981), Atampaya (Crowley 1983), Awu Alaya (Sommer n.d.-b), Ayapathu (Hamilton 1997a), Bakanh (Hamilton 1997b), Djabugay (Robertson and Sommer 1997), Flinders Island (Sommer n.d.-c), Guugu Yimidhirr (Haviland 1979b), Ikarranggal (Bowern 2016a, currently unreleased dataset), Kaantju (Sommer n.d.-d), Kok Nar (Sommer n.d.-e), Kokiny (Sommer n.d.-f), Koko Bera (Black 1999), Koko Dhawa (Sommer n.d.-g), Kugu Nganhcara (Smith and Johnson 1989), Kuku Yalanji (Hershberger and Hershberger 1986), Kurtjar (Black and Gilbert 1996), Kuugu Ya‘u (Thompson 1988), Linngithigh (Hale 1999), Malthanmungu (Haviland 1992), Mbabaram (Dixon 1991b), Ndra’ngith (Bowern 2016a, currently unreleased dataset), Ogh Alungul (Sommer n.d.-h), Ogh Angkula (Sommer n.d.-i), Ogh Awarrangg (Sommer n.d.-j), Ogh Unyjan (Sommer n.d.-k), Olkol (Hamilton 1997c), Oykangand (Hamilton 1997c), Rimanggudinhma (Sommer 1999c), Tableland Lamalama (Sommer n.d.-l), Thaayorre (Foote and Hall 1995), Thaynakwithi (Fletcher 2007), Umbuygamu (Verstraete 2017b), Umpila (O’Grady 1988b), Wik Mungkan (Kilham et al. 2011), WikNgathan (Sutton 1995b), Yadhaykenu (Crowley 1983), Yidiny (Dixon 1991c), Yintyingka (Verstraete and Rigsby 2015), Yir Yoront (Alpher 1991), Pilbara Jiwarli (Austin n.d.a), Kariyarra (Smythe and Lockyer n.d.), Kurrama (Dench n.d.-b), Martuthunira (Dench 1995), Ngarla (Brown and Geytenbeek n.d.), Ngarluma (Hale 1989a), Nyamal (Burgman 2007b), Nyiyaparli (O’Grady 1988c), Panyjima (Dench 1991a), Payungu (Austin n.d.-b), Purduna (Burgman 2007a), Thalanyji (Austin n.d.-c), Tharrkari (Austin 1992b), Warriyangga (Austin n.d.-d), Yindjibarndi (Anderson, Richards, and Summer Institute of Linguistics n.d.), Yinhawangka (Wangka

phonotactics Maya Pilbara Aboriginal Language Centre 2008c), Qld-NSW coast Bundjalung (Sharpe 1998), Butchulla (Bell 2003), Duungidjawu (Kite and Wurm 2004), Gidabal (Crowley 1978), Gumbaynggir (Murrbay Aboriginal and Culture Cooperative 2001), Jandai (Iselin and Minjerribah Moorgumpin 2011), Waalubal (Crowley 1978), Yagara (Jefferies 2011), Yaygir (Morelli 2012), South-west Badimaya (Marmion 1995), Bibbulman (Dench 1994), Malkana (Gargett 2011), Minang (Dench 1994), Mirniny (O’Grady and Curr 1988), Ngadjunmaya (Wangka Maya Pilbara Aboriginal Language Centre 2008d), Nhanda (Blevins 2001a), Wajuk (Dench 1994), Wardandi (Dench 1994), Watjarri (Mackman 2012), Wiilman (Dench 1994), Yingkarta (Austin n.d.-e), Thura-Yura Adnyamathanha (McEntee and McKenzie 1992), Narrungga (Narungga Aboriginal Progress Association and others 2006), Nukunu (Hercus 1992a), Wirangu (Hercus 1999), Upper Murray Dhudhuroa (Blake and Reid 2002), Yorta Yorta (Bowe and Morey 1999), Warluwaric Bularnu (Breen 1988), Eastern Wakaya (Breen 2006b), Warluwarra (Breen 1990c), Western Wakaya (Breen 2006b), Yanyuwa (Bradley n.d.), Yardli-Paakantyi Malyangapa (Bowern 2016a, currently unreleased dataset), Southern Paakintyi (Hercus n.d.c), Yolngu Dhangu (Zorc 2004), Dhay’yi (Wunungmurra 1993), Djambarrpuyngu (Yirrkala School Literature Production Centre 1984), Djapu (Morphy 1983), Djinang (Waters 1988), Gumatj (Bowern 2016a, currently unreleased dataset), Gupapuyngu (Lowe 1976), Nhangu (James 2003), Ritharrngu (Heath 1976b), Yuin-Kuri Awabakal (Lissarrague 2006), Darkinyung (Jones 2008), Dharawal (Besold 2012), Dharuk (Troy 1994b), Dharumba (Besold 2012), Dhurga (Besold 2012), Eora (Troy 1994b), Katthang (Lissarrague 2010), Nganyaywana (Wafer and Lissarrague 2008), Ngarigu (Wafer and Lissarrague 2008), Ngunawal (Wafer and Lissarrague 2008), Thanggati (Bowern 2016a, currently unreleased dataset), Other families: Bunuban Bunuba (Kimberley Language Resource Centre 2010), Gooniyandi (Kimberley Language Resource Centre 1993), Eastern Daly Matngele (Zandvoort 1999), Darwin Larrakia (Harvey 2004), Limilngan (Harvey 2001), Garrwan Garrwa (Furby and Furby 1993), Waanyi (Laughren 2016b), Giimbiyu Erre (Birch 2006), Mengerrdji (Birch 2006), Urningangg (Birch 2006), Gunwinyguan Dalabon (Bowern 2016a, currently unreleased dataset), Gun-Djeihmi (Evans 2003a), Jawoyn (Merlan and Jacq 2005), Kuninjku (Evans 2003a), Kunwinjku (Evans 2003a), Ngalakgan (Bowern 2016a, dataset: mor83), Ngandi (Heath 1978b), Rembarrnga (Bowern 2016a, currently unreleased

dataset), Warray (Harvey n.d.-a), Wubuy (Bowern 2016a, dataset: Nunggubuyu Flora Fauna), Iwaidjan Amurdak (Handelsmann 1991), Iwaidja (Pym and Larrimore 2011), Mawng (Singer et al. 2015), Jarrakan Kija (Blyth 2001), Miriwoong (Kofod 1978), Kungarakany Kungarakany (Bishop 2000), Maningrida Burarra (Glasgow 1994), Gurr-Goni (Green and Nimbadja 2015), Nakara (Eather, Yurrbukka Community, and Bawinanga Aboriginal Corporation 2005), Ndjebbana (Bowern 2016a, currently unreleased dataset), Maran Alawa (Sharpe 2001), Mangarrayi (Merlan 1989), Marra (Heath 1981a), Warndarrang (Heath 1980b), Mirndi Gudanji (Aguas 1968), Jaminjung (Harvey 1986), Ngarnka (Osgarby 2016), Nungali (Bolt, Hoddinott, and Kofod 1971), Wambaya (Bowern 2016a, dataset: RN Wambaya 29,269,292; NE), Nyulnyulan Bardi (Aklif and Kimberley Language Resource Centre 1999), Jabirr-Jabirr (Bowern 2016a, currently unreleased dataset), Jawi (Bowern 2016a, currently unreleased dataset), Djukun (Nekes and Worms 2006), Nyikina (Stokes n.d.), Nyulnyul (Bowern 2016a, currently unreleased dataset), Warrwa (Bowern 2016a, currently unreleased dataset), Yawuru (Bowern 2016a, currently unreleased dataset), Southern Daly Murrinhpatha (Street 1987), Tangkic Gangalidda (Round 2016a), Kayardild (Round 2016b), Lardil (Hale and Ngakulmungan Kangka Leman 1997), Yangkaralda (Cook 2017), Tiwi Tiwi (Lee 2013), Western Daly Emmi (Ford 1998), Marramaninyshi (Tryon 1974), Patjtjamalh (Ford 1997), Worrorran Kwini (Sanz de Galdeano 2005), Ngarinyin (Coate and Elkin 1974), Unggumi (McGregor 1985b), Worrorra (McGregor 1985b), Wunambal (Bowern 2016a, currently unreleased dataset), Yawijibaya (McGregor 1985b), Yangmanic Wagiman (Wilson and Harvey 2001), Wardaman (Bowern 2016a, dataset: Merlan).

Acknowledgements Support for this research is gratefully acknowledged from the Australian Research Council (grant DE150101024), the British Academy (grant VF1_101602), the ARC Centre of Excellence for the Dynamics of Language (grant TIG322015), as well as the University of Queensland, Surrey Morphology Group and the Max Planck Institute for the Science of Human History. For their work on the AusPhon-Lexicon database, my thanks to Jacqui Cook, Jordan Hollis, Edith Kirlew, Jayden Macklin-Cordes, Amy Parncutt, and Genevieve Richards. For helpful comments on the chapter, thanks to Claire Bowern.

119

chapter 12

Morphophonology Lenition and assimilation Erich R. Round

The phonologies of the world’s languages vary not only in their static properties, such as segment inventories and phonotactics, but also in their dynamic, morphophonological alternations. In the study of Australian phonologies, static properties have long held the spotlight, with booklength works appearing already several decades ago on segments (Busby 1980) and phonotactics (Hamilton 1996b). Dynamic phonology in comparison has never really taken centre stage.1 Short discussions of at most a few pages per phenomenon appear on morphophonological topics in overview works by Evans (1995a), Dixon (1980; 2002), and Baker (2014). These have proven invaluable, but the short format lends itself to the citation of particularly striking or well-known data, and since it lacks space to explore diversity in detail, can contribute to an exaggerated discourse of uniformity in Australian languages, where phenomena are rare, pervasive, or absent, but seldom ‘diverse’. To address this, the current chapter presents just two studies, and a third appears in Chapter 13. Each is on a topic chosen for its particular interest with respect to Australian languages, and owing to the state of the literature described just above, each is (at time of writing) the most in-depth survey of that phenomenon in Australian languages to date, and fills a gap in our knowledge that has persisted for too long. Section 12.1 covers materials and methods. Section 12.2 examines lenition, a morphophonological process that is particularly common in Australian languages. Section 12.3 investigates assimilation, and relates it back to key phonotactic generalizations adduced in Chapter 11. Section 12.4 offers concluding remarks. Sampled languages and their source documents—without which, none of this research would be possible—are listed in Section 12.5.

1 Though two major works are imminent, Baker and Harvey (to appear) and Round (in prep).

12.1 Materials and methods As in Chapters 10 and 11 of this volume, the approach here is to mobilize large comparative datasets to provide quantitatively-backed insights into the diversity of Australian phonologies. The studies in Sections 12.2 and 12.3 draw on data in the AusPhon-Alternations database. In brief, AusPhon-Alternations (AA) tracks instances in descriptive grammars where a morphophonological alternation is reported, irrespective of whether the report is in the form of (morpho)phonological formalisms, lists of allomorphs, or in prose. In an initial step, these reports are recorded and marked up to indicate potential interpretations of the data, for example as being amenable to an analysis as lenition, deletion, assimilation, and so forth. As the coverage of the dataset grows and salient dimensions of empirical variation emerge, a second round of more nuanced annotations is added and refined. The studies here are based on analyses of these late-iteration typological characterizations informed by a substantial coverage of Australian languages. The AA methodology is an approximate one, and we view the database as a significant—but nevertheless initial—attempt at surveying and typologizing Australia’s morphophonological variation. Notwithstanding that, the multi-stage procedure has the advantage of moderating the potentially outsized influence of observations that happen to be surveyed (or to have entered the literature) first, and promotes the discovery of new insights based on evidence distributed across many languages, as Sections 12.2 and 12.3 will attest. The sample drawn on for this chapter is 118 doculects (i.e. descriptions of languages, Good and Cysouw 2013).2 These contain close to two thousand alternations in 2 Only one doculect is used per language here. In several cases, AA contains multiple doculects for a single language. This is for future study, that may inform us about how variable the same language can appear through the lens of different sources, and therefore indicate how uncertain we

Erich R. Round, Morphophonology. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Erich R. Round (2023). DOI: 10.1093/oso/9780198824978.003.0012

120

morphophonology total, of which a subset relevant to lenition and assimilation is studied here.

12.2 Lenition Lenition is a common process in the world’s languages (Gurevich 2004). It appears as a synchronic morphophonological process in more than one third of Australian languages in the AA dataset. Definitions of lenition vary, but essential to all of them is a notion of segmental weakening (Honeybone 2008). For reasons of space, I confine myself here to the most common and widespread kinds of lenition processes in Australian languages, namely alternations in syllable onset position between stops and more sonorous oral segments or zero, in which the alternations are phonologically conditioned by the sonority of the segment on the left.3 An example from Wardaman (Merlan 1994) is the dative suffix in (1), which appears as /-ku/ after stops and nasals, and /-wu/ elsewhere. (1) Wardaman (Merlan 1994: 24, 28, 29, 38) a. waɭ p-ku b. lin-ku toilet-dat snake-dat c. ȶer-wu d. wure-wu ground oven-dat child-dat Only contrastive (non-allophonic) alternations are examined here. For expository convenience, I will refer to the alternations of interest as ‘lenition’, and I return to the question of their synchronic analysis as lenition or as fortition in Section 12.2.7. Subsections below cover the frequencies of stops’ participation in lenition alternations according to their place of articulation in Section 12.2.1; attested pairings between stops and lenis alternants in Section 12.2.2; one-to-many and many-to-one pairings in Section 12.2.3; ranges of places of articulation at which lenition occurs in a language in Section 12.2.4; phonological conditioning in Section 12.2.5; morphological conditioning in Section 12.2.6; and the synchronic analysis of alternations as lenition or fortition in Section 12.2.7. should be, when we have only one description of a language. At time of writing, the AA method has been attempted with over 250 doculects. 130 of these were found to report morphophonological alternations according to our procedure. 3 Related phenomena beyond the scope of this study include alternations in coda positions, lenition of sonorants, alternations between fortis and lenis stops, and sonority-conditioned alternations between stops and nasals.

12.2.1 Participation of stops in lenition alternations, by place of articulation. The lenition alternations under study here appear in 46 of the languages in the AA dataset. From a series of different angles, Sections 12.2.1–12.2.5 examine the frequencies with which various stops and lenis alternants participate in these alternations.4 Counts will be in terms of the number of languages in which certain patterns appear. The picture to emerge will be one of considerable variation around a core of common themes. Figure 12.1 shows the number of languages (out of 46) in which lenition alternations involve stops at each of six superlaryngeal places of articulation (see Chapter 10, this volume, regarding places of articulation in Australian languages). The results reveal an essentially bimodal distribution between velar, labial, and palatal stops which are frequent participants in lenition alternations, vs. alveolar, dental, and retroflex stops which are infrequent participants. This is broadly consistent with previous observations (Dixon 2002: 627; Round 2011). However, to appreciate the import of the observation, it is necessary to clarify what Figure 12.1 conveys. The counts in Figure 12.1 are absolute tallies. They do not reveal the proportions of /k/, /p/, /t/ which undergo lenition, but merely the absolute counts.

k p ȶ t t ʈ 0

5

10

15 20 25 Number of languages

30

35

Figure 12.1 Participation of stops in lenition alternations, in 46 languages. 4 Only three of the 46 languages (Ritharrngu, Limilngan, Djambarrpuyngu) also make a fortis/lenis contrast in their stops. In Ritharrngu and Limilngan only the lenis stop series alternates with sonorants; in Djambarrpuyngu both series do. Here I abstract away from the stop series contrast and examine stops only in terms of their place of articulation.

121

erich r. round Consequently, although Figure 12.1 may at first glance appear to support hypothesis (2a), it could be that all stops have a broadly comparable propensity to undergo lenition, but that velars, labials, and palatals are simply more common than the others, as in (2b). Indeed, Figure 12.1 is also consistent with hypothesis (2c). (2) Hypotheses on the propensities for stop lenition, by places of articulation a. Velar, labial, and palatal stops have a higher propensity than other places. b. Stops at all places have broadly comparable propensities. c. Any other combination of relative propensities. In order to clarify the matter, two additional contextualizing counts were conducted. The overwhelming majority of lenition alternations in the AA dataset involve the initial segments of suffixes. Some do involve root- and prefix-initial segments, but it is suffixes that are driving the numbers in Figure 12.1. Accordingly, Figure 12.2 shows the frequencies by place of articulation of suffix initial stops in the complete AA dataset, not only for lenition, but for any alternation.5 It should be borne in mind that AA is limited to suffixes that alternate, so the figures may still be biased: for example, if suffix initial /t/ is numerous overall but rare in suffixes that alternate in any way, it would be underrepresented in Figure 12.2. To provide some insurance against that possibility, Figure 12.3 shows the frequencies of suffix-initial stops in the lexical dataset described in Chapter 11.5. Not all languages in that dataset include suffixes among their listed lexical items, so the coverage is patchy and needs to be interpreted with caution. However, with these caveats in mind, Figures 12.2 and 12.3 provide information that will help us understand lenition in Australian languages more thoroughly. Figures 12.2 and 12.3 do not match exactly. Figure 12.2 shows a little over double the proportion of alveolars, dentals, and retroflexes that Figure 12.3 does, though the proportion is low in both datasets. The gap between alveolars, dentals, and retroflexes on the one hand, and palatals and labials on the other, is about the same in Figure 12.3 (the lexical suffix data) as in Figure 12.1 (the lenition data). Thus, contrary to hypothesis (2a), we find no evidence that alveolars, dentals, and retroflexes have an especially low propensity for participating in lenition alternations; they just have a lower frequency overall. By the same token, hypothesis 5

In counting alternations, I excluded the Pama-Nyungan ergative/locative allomorphy, since its alternations are very widespread and can skew the results if not controlled for.

122

k, g p, b ȶ, ȡ t, d t, d ʈ, ɖ 0.0

0.1 0.2 0.3 0.4 Proportion of suffixes

Figure 12.2 Suffix-initial superlaryngeal stops in the AA dataset: proportion by place of articulation. k, g p, b ȶ, ȡ t, d t, d ʈ, ɖ 0.0

0.1 0.2 0.3 0.4 Proportion of suffixes

Figure 12.3 Suffix-initial superlaryngeal stops in AusPhon Lexicon: proportion by place of articulation.

(2b), which states that stops at all places of articulation have a broadly comparable propensity for lenition, is also challenged by the data, namely by the velars. Dorsal velars have a strikingly high representation in Figure 12.2 (AA suffixes) and Figure 12.3 (lexical suffix data), clearly outstripping palatals and labials, yet this is not so in Figure 12.1 (the lenition data). Thus there is good prima facie evidence that velars have a propensity for participating in lenition alternations that is lower than palatals and labials. To my knowledge these results are new, and multiple explanations for them will now be worth exploring. It could be the case, for instance, that velars undergo historical lenition more rarely than do palatals and labials. But it could also be true that velars, labials, and palatals all succumb to lenition at about the same rate diachronically, after which the resulting synchronic lenition alternations are lost more rapidly for velars than for labials and palatals. Both scenarios could produce the results just above.

12.2.2 Pairings between stops and lenis alternants Figure 12.4 shows frequencies of pairings of stops with their more sonorous lenis alternants. Three salient points are: that /p, ȶ, k/ overwhelmingly alternate respectively with

morphophonology p~w ȶ~j k~w t~j t~ɻ ʈ~ɻ k~j t~l t~j t~ð ȶ~ɻ ʈ~j p~β p~j k~ɰ k~Ɣ 0

5 10 15 20 25 30 35 Number of languages

Figure 12.4 Pairings of stops with their lenis alternants, within lenition alternations.

/w, j, w/; that alternations with fricatives are rare, consistent with the rarity of contrastive fricatives in Australian languages (Chapter 10, this volume); and that a broad spread of minor variants is attested. Figure 12.5 reports frequencies of alternations between stops and zero. To aid comparison, the scale of the numerical axes is the same in Figure 12.4 and Figure 12.5, showing that alternations with zero are rarer than with non-zero segments. k~Ø p~Ø ȶ~Ø t~Ø t~Ø 0

5

10 15 20 25 30 Number of languages

next to back. This creates a non-trivial confound when asking what the result is when an intervocalic stop lenites, since if it is replaced by a glide on the surface, it is not obvious whether the glide is the direct result of lenition, or if lenition results first in a complete loss of the stop, followed thereafter by the insertion of a glide. The procedure here is as follows. If a stop lenites intervocalically to the same glide, e.g. always to /w/, irrespective of the quality of the flanking vowels, I assume that the glide is the outcome of lenition, and the alternation is counted in Figure 12.4. If the stop lenites intervocalically and is replaced on the surface only ever by glides that are homorganic with one of the flanking vowels and which vary according to those vowels’ qualities, I assume lenition to zero. These cases appear in Figure 12.5. If the stop lenites intervocalically and is replaced on the surface by glides that vary according to the adjacent vowels, but which in only some cases are homorganic with flanking vowels, then I assume lenition to zero for homorganic-glide cases and lenition to a glide in the other cases. These cases will contribute to counts in both Figures 12.4 and 12.5.

12.2.3 One-to-many and many-to-one pairings When phonologists come to analyse lenition alternations, the empirical facts around neutralization can be pivotal (Baker 2014: 171–3): if both /p/ and /k/ alternate with /w/, it may be more parsimonious to say that underlying /p/ and /k/ both lenite to a derived segment /w/ than to say underlying /w/ undergoes fortition to /p/ in some cases and /k/ in others. It can be useful therefore, to examine the manyto-one and one-to-many mappings entered into by stops and their lenis alternants. Figure 12.6 tallies sets of stops that within one language alternate with a lenis alternant in common. Unsurprisingly,

35

Figure 12.5 Pairings of stop with zero, within lenition alternations.

Of the stops that do alternate with zero, velars notably outnumber other places of articulation, though it is important to clarify the methodology here. Most analyses of Australian languages propose that phonemic vowels cannot appear in hiatus, and not surprising under this kind of analysis, many languages require rules of glide insertion to break up underlying and derived vowel+vowel clusters. Where either or both of the vowels is non-low, the inserted glide is very often homorganic: /j/ next to front vowels, /w/

{k, p} ~ w {t, ȶ} ~ j {k, ȶ} ~ j {k, p, ȶ} ~ Ø {ʈ, ȶ} ~ j {t, ȶ} ~ j {t, ʈ} ~ ɻ {p, t} ~ Ø {k, ȶ} ~ Ø {k, t, ȶ} ~ Ø 0

5 10 15 20 Number of languages

Figure 12.6 Many-to-one stop-to-lenis mappings.

123

k ~ {Ø, w} p ~ {Ø, w} ȶ ~ {j, Ø} k ~ {j, Ø, w} t ~ {l, Ø} t ~ {j, ɻ} k ~ {j, w} ȶ ~ {j, Ø, ɻ} t ~ {ð, j, Ø} p ~ {Ø, w, β}

0

5

10

15

20

Number of places of articulation

erich r. round

Number of languages

12.2.4 Ranges of places of articulation for lenition Australian languages vary in the range of places of articulation at which their stops ever undergo lenition. Figure 12.8 tallies the raw number of different places of articulation at which there are stops that participate in lenition alternations. Figure 12.9 counts the frequencies of specific sets of places of articulation (note the numerical scales are not the same). Unsurprisingly, Figure 12.9 is dominated by permutations of velar, labial, and palatal. What is perhaps interesting is the evident lack of any stringent universal scale, such that place of articulation a only participates in lenition if b does, and so forth; rather for the three most common stops {k, p, ȶ} it is possible to find both a pair and a triple that lacks one of them while containing the other two. (Since alveolars, dentals, and retroflexes are rare overall, the absence

124

2 3 4 5 0

Figure 12.7 One-to-many stop-to-lenis mappings.

5 10 Number of languages

15

Figure 12.8 Number of places of articulation of stops that participate in lenition alternations.

Places of articulation

/k, p/ both alternate with /w/ often, in 20 of the 46 languages. The next most common set comprises the two laminal stops both alternating with /j/ in six languages, followed by a diverse tail of rarer many-to-one mappings. Figure 12.7 tallies sets of lenis segments that alternate with a common stop; for comparability, the numerical scale is the same as in Figure 12.6. Most involve sets of size two, comprising a lenis segment and zero. It is not obvious what to conclude from any comparison of Figure 12.6 vs. 12.7. Aside from the most common mappings, {k, p}~w and k~{Ø,w}, the counts in Figure 12.6 and 12.7 are low. Moreover, the simple counts in Figure 12.7 may mask more interesting underlying variation, by conflating qualitatively different circumstances under which the multiple lenis variants appear; that is, they could be associated with different morphemes, or with different triggering environments. Exploring those issues in detail will be valuable but is beyond the scope of this study.

1

{p} {k} {ȶ} {k, ȶ} {k, p} {p, ȶ} {k, p, ȶ} {p, t, ȶ} {k, p, t} {k, t, ȶ} {k, p, t, ȶ} {k, p, ʈ, ȶ} {k, p, t, t, ȶ} {k, p, t, ʈ, ȶ} {k, p, t, ʈ, ȶ} 2 4 6 8 0 Number of languages

Figure 12.9 Sets of places of articulation of stops that participate in lenition alternations.

of any specific combinations involving them is only to be expected.6 )

6 To put numbers on it, we can compare the number of attested sets containing at least one of the alveolar, dental, or retroflex places vs. the number of possible sets containing them. For sets of one place of articulation, we find 0 of the 3 possible sets {t}, {t ̪}, or {ʈ}. For sets of two places, we find 0 of the 15 possible. For sets of three, we find 3 of 60. For sets of four, 2 of 180. For sets of five, 3 of 360. Note that not all possible sets are equally probable, so these figures should not be used to draw conclusions like ‘there is a 1 in 20 chance of this happening’; what is meaningful though, is to compare for example, that there are only 3 possible sets of four places of articulation that contain all of {k, p, ȶ}, and 2 of them are attested.

morphophonology p k ȶ t

Consonant-Vowel Ambiguous: C-V or Occl-Cont Occlusive-Continuant Other contiguous split

t ʈ

Non-contiguous

0

10 20 30 40 Number of languages

50

Figure 12.10 Sonority of adjacent segments to the left, conditioning stop vs. lenis alternant (by stop type). Consonant-Vowel

w

Ambiguous: C-V or Occl-Cont

j

Occlusive-Continuant

Ø

Other contiguous split

ɻ 0

10

20 30 40 Number of languages

50

Non-contiguous

Figure 12.11 Sonority of adjacent segments to the left, conditioning stop vs. lenis alternant (by lenis type).

12.2.5 Phonological triggers The alternations under consideration here are alternations between stops in syllable onsets and a more sonorous oral segment or zero, whose phonological conditioning is by the segment to the left, and is determined by its sonority. Overwhelmingly, the stops occur after less sonorous segments to the left, and the lenis alternants occur after more sonorous segments to the left, but there is variation in where the sonority scale is split, and there are a few instances where the sonority ranks of the triggering segments do not fall neatly into a single less-sonorous set and a single moresonorous set. When the triggering segments do fall into a neat division of the sonority scale, the variation attested is tightly constrained. Almost all splits are between consonants vs. vowels (which condition the appearance of a following stop vs. lenis alternant respectively) or between occlusives (stops and nasals) vs. continuants (cf. Baker 2014; Round 2011).7 Some systems are empirically consistent with both of these interpretations, owing to the fact that neither of the alternating segments is observed after a continuant consonant. Figure 12.10 tallies, according to place of articulation of the stop, the counts of languages which condition that stop’s lenition alternation according to a consonant–vowel split, 7 Dixon (2002: 627) inexplicably omits the occlusive–continuant pattern, instead citing a very rare pattern in which vowels and the trill condition the lenis alternant.

an occlusive–continuant split (and cases ambiguous between these), as well as by other simple partitions into lowvs.-high sonority and finally, according to patterns which do not split the sonority pattern in a simple manner, but rather have one or both of the alternating segments conditioned by a non-contiguous region of the sonority scale. Figure 12.11 does the same, but arranged according to the lenis segment, for lenis segments attested in at least three alternations in the AA dataset. For the purpose of the analysis, all vowels are regarded as having equal sonority. Likewise, all liquids are taken to have equal sonority, whether laterals, trills, taps, or flaps. Because a language can exhibit multiple patterns of sonority-based conditioning of lenition for one and the same stop, for instance when different conditions apply to different suffixes, the tallies in these Figures are higher than the simpler tallies back in Figure 12.1, which only counted whether or not a certain stop participated in lenition alternations, and lumped all conditions together. The results show that neat bipartitions of the sonority scale are implicated in the overwhelming majority of lenition alternations’ triggering conditions. In truth, for alternations with zero, this is in part predetermined by the way the data was gathered: alternations were identified as lenition according to the criteria outlined just above, but ‘antilenition’ alternations—such as /k/ after vowels and zero after consonants—were not included even though they certainly exist (they would typically be regarded as deletion). Apart from this though, the methodology is not the source

125

erich r. round for the other striking results that emerge. Non-zero ‘antilenition’, such as /k/ after vowels and /w/ after consonants, is not found and its absence is not an artefact of the method. Among the neat sonority splits, the vast majority fall into just two types: consonant–vowel or occlusive–continuant. These would appear to be natural fault lines in the sonority scale with respect to the conditioning of lenition in Australian languages. Even the ‘other contiguous split’ and ‘noncontiguous split’ categories are often organized around a basic tri-partition of the sonority scale into occlusives, continuant consonants, and vowels. For example, in Bilinarra (Meakins and Nordlinger 2014: 68) the continuative suffix on verbs is /-kara/ after occlusives and vowels, but /-ara/ after liquids. This splits the sonority scale non-contiguously but still uses the familiar, basic tri-partition of it. In other cases, non-contiguous splits are due to a single segment acting unusually in its triggering behaviour, while the rest of the sonority scale is split regularly. For example, in Bunuba (Rumsey 2000: 97) the iterative aspect marker is /-pa/ after occlusives and /-wa/ after continuants, with the single exception that /-pa/ will also appear after the glide /j/. All in all, the conditioning of stops vs. lenis alternants according to the sonority of the segment to their left is remarkably orderly.

12.2.6 Morphological conditioning Morphology plays a significant role in the conditioning of lenition in Australian languages (Baker 2014: 170, 173). Since an extensive survey is beyond the scope of this study, here I list key aspects of it which deserve more research (see Round in prep). Synchronic lenition (or fortition) never applies without exception across the lexicon, rather it applies primarily in derived environments. Most often these are morphologically derived environments,8 within which lenition can exhibit classic derived-environment effects. In Yukulta/Gangalidda, for example, the portmanteau locative-dative suffix /-kurka/ undergoes lenition to /urka/ when it follows a continuant consonant, including a stem-final /r/ as in /wirwir+kurka/ → /wirwirurka/ ‘riverbank-loc.dat’ (Round 2014a: 183). However, only the first /r+k/ cluster, which is morphologically derived, is subject to lenition of /k/, while the second, underived /rk/ sequence is unaffected. Lenition is often subject to morphological idiosyncrasy, which appears in various guises. Idiosyncrasy can operate in terms of which formatives are, or are not, potential undergoers of lenition. In Wanyjirra, suffix-initial velar, labial, and 8 See Chapter 13.6 on the feeding of lenition by nasal cluster dissimilation, for examples of lenition in phonologically derived environments.

126

palatal stops undergo lenition, but only in some suffixes and clitics and not others (Senge 2015: 92–3). Idiosyncrasy can operate in terms of which lenis segment pairs with a given stop. In Gaagudju, /k/ alternates with the glide /w/ at the start of class-marked kin nouns and most human status adjectives, with /j/ in four phratry adjectives, and otherwise with zero in stems longer than two syllables (Harvey 2002: 34). It can operate in terms of the sonority conditions under which lenition is triggered. In Mara (Heath 1981a: 34, 36) the lenition of /p/ to /w/ occurs following a continuant when at the beginning of inflectable auxiliaries, but occurs only after vowels at the beginning of certain nominal and cardinal-directional stems, and in certain verbal reduplications. The examples cited so far are idiosyncrasies which hinge on the morphological identity of the (non-)target of lenition, but idiosyncrasy can also be associated with the trigger. In Djapu, lenition in case suffixes is triggered differently by nominal vs. pronominal stems (Morphy 1983: 53, 59). It can associate with differences in morphological constructional types, such as reduplication and compounding (e.g. Djambarrpuyngu, Wilkinson 1991: 71), or with specific combinations of trigger and target (e.g. in Djaru, Tsunoda 1981). Though this list is only brief, it serves to establish that the intrusion of morphology into the conditioning of lenition is an interesting and complex issue, deserving of further investigation.

12.2.7 Discussion: synchronic analysis as lenition or fortition I have referred to the alternations under study here as ‘lenition’, but phonologists have accorded them synchronic analyses as both lenition and fortition (Baker 2014: 170–5). Rather than review these choices themselves, it may be more useful to clarify some of the inherent logical issues that a phonologist faces when choosing whether to analyse these alternations as lenition or fortition. The central argument is that in most instances, the alternation data itself will not decide which choice is best, rather it leaves both options open (and this is reflected in the analyses that exist in the Australian phonological literature). To see how this comes about, five illustrative schematic datasets will be helpful, referred to just below as languages A to E. The relevant data for language A is in (3). In language A, suffixes of ‘type I’ have an initial segment that surfaces as a stop /k/ after a preceding consonant and a glide /w/ after a preceding vowel. Language A also has ‘type II’ suffixes, which always begin with a stop /k/, and it has no suffixes of type III, that always begin with the glide /w/.

morphophonology (3) Language A Suffix type I: initial k~w ŋaral-kiri ŋara-wiri * Suffix type III: initial w only ŋaral-wara ŋara-wara

Suffix type II: initial k only ŋaral-kara ŋara-kara

When classic phonological reasoning is applied to this dataset, the obvious choice of synchronic analysis for language A is in terms of fortition. The initial stop of the type II suffix does not vary on the surface, so we can assume it is also underlyingly a stop, /k/. Since there are no suffixes with invariant initial /w/ on the surface, we are free to posit /w/ as the underlying form of the initial segment of type I suffixes, and derive the stop variant by a rule of fortition that applies after consonants. Not only is fortition a simple analysis, but a lenition analysis would be gratuitously complicated. We would need to posit some abstract difference between stops that lenite (type I), and stops that do not (type II), and a rule of lenition whose application makes reference to that abstract difference. Thus, in language A the choice in favour of a fortition analysis is clear. It follows from the empirical facts plus the methodological principle of parsimony, that gratuitous abstractness is to be avoided. Similar argumentation would apply to language B in (4), which has suffixes of type I and III but not II: for language B, the obvious choice of analysis would be lenition. (4) Language B Suffix type I: * Suffix type II: initial k~w initial k only ŋaral-kiri ŋara-wiri ŋaral-kara ŋara-kara Suffix type III: initial w only ŋaral-wara ŋara-wara Few Australian languages are like languages A or B though. The relevant data for language C is shown in (5). In language C, not only /k/ but also /p/ alternates with /w/. In general, the issues that arise for language C are relevant for any system with many-to-one or one-to-many mappings between stops and their lenis alternants. (5) Language C Suffix type I: initial {k, p}~w ŋaral-kiri ŋara-kara ŋaral-pa ŋara-wa * Suffix type III: initial w only ŋaral-wara ŋara-wara

Suffix type II: initial k, p only ŋaral-kara ŋara-kara ŋaral-pul ŋara-pul

As in language A, language C has no ‘type III’ suffixes, but the critical new issue is that if a suffix starts with /w/ after a vowel, it does not necessarily start with /k/ after a consonant, rather it might also start with /p/. This makes the use of some kind of abstractness unavoidable, since there is now a need to distinguish leniting stops from non-leniting stops (if a lenition analysis is chosen) or to distinguish /w/ which strengthens to a velar stop from /w/ that strengthens to a labial stop (if a fortition analysis is chosen). Phonological theory provides many technical options for implementing the necessary abstractness: abstract morphophonemic contrasts such as /w1 / vs. /w2 / or /k/ vs. /K/; phonological underspecification; morphological diacritics (such as class I vs. class II suffixes); differences among kinds of phonological junctures or phonological strata; or differences among kinds of morphological or prosodic constituent structures. Irrespective of the formal implementation, the key logical issue is that under the circumstances in language C, lenition and fortition are both viable analyses. Abstractness is required for both, and is gratuitous for neither. Whatever analysis the phonologist does choose, the decision will not be one that follows simply from the empirical facts of the alternation plus parsimony, as it did for languages A and B. Language D, in (6), leads the phonologist to the same conundrum, though by a different route. Since suffixes of types I, II, and III all exist, there are stops that alternate with glides and stops that don’t, and glides that alternate with stops and glides that don’t. Whether a fortition analysis or a lenition analysis is chosen, abstractness will be necessary, and the actual choice of analysis will not follow simply from the empirical facts of the alternation plus parsimony. (6) Language D Suffix type I: Suffix type II: initial k~w initial k only ŋaral-kiri ŋara-wiri ŋaral-kara ŋara-kara Suffix type III: initial w only ŋaral-wara ŋara-wara The relevant data for language E is shown in (7). In language E, the alternation is between /k/ after occlusives (stops and nasals) and zero after continuants. The language also contains non-alternating vowel-initial suffixes of type IV. (7) Language E Suffix type I: Suffix type II: initial k~Ø initial k only ŋaran-kiri ŋaral-iri ŋaran-kul ŋaral-kul Suffix type IV: initial vowel only ŋaran-ara ŋaral-ara

127

erich r. round In language E, suffix initial /k/ after an occlusive might alternate with zero (type I) or not (type II), and a suffix initial vowel /V/ might alternate with /kV/ (type I) or not (type IV), leading once again to abstractness, and a choice of analysis that does not follow simply from the empirical facts of the alternation plus parsimony. All of this has implications for the degree of caution that needs to be exercised when comparing lenition vs. fortition in Australian languages. We saw in Sections 12.2.2– 12.2.3 that many Australian languages share critical traits with languages D and E, namely many-to-one or one-tomany mappings, and alternation with zero. Yet others follow language C. For these languages, the choice of synchronic analysis simply will not follow from the basic facts of the lenition/fortition alternation. Instead, it will necessarily depend on something additional, perhaps information from opaque orderings with respect to other processes; or theoryinternal considerations such as naturalness or markedness; perhaps reference to the form of segments in ‘neutral’ environments like word-initial position; or the productivity of the alternation; or a preference for one’s analysis to be compatible with analyses of related languages. Within this list of considerations, any two may well point the analysis in opposite directions, in which case the final decision may also hinge on the phonologist’s relative ranking or weighting of these considerations. This all makes for a highly complex link between the observable outcome (analysis as lenition vs. analysis as fortition) and the many underlying factors that ultimately lead to it. It also means that ‘lenition languages’ and ‘fortition languages’ are quite unlikely to be coherent or natural or insightful categories. And for that reason, I do not present any tally of such categories here. Instead, it remains a task for future research to examine the factors that shape lenition/fortition choices, and to establish what they reveal about Australian phonological systems.

12.2.8 Summary Over a third of the AA languages have alternations of the kind I have termed ‘lenition’, an alternation between stops and more lenis alternants or zero. Velar, labial, and palatal stops alternate in this way more often than stops at other places of articulation, though this appears mainly to reflect their relative prevalence in suffix-initial position rather than a difference in a propensity to participate in lenition alternations per se. Most commonly, /k, ȶ, p/ alternate respectively with /w, j, w/, though a variety of other mappings are attested, including alternation with zero, and manyto-one and one-to-many mappings are well represented.

128

No stringent scale is apparent, such that place of articulation a only participates in lenition if b does. The conditioning of lenition alternations by phonological context is highly regimented, and overwhelmingly patterns according to whether the adjacent segment to the left is a consonant vs. vowel, or occlusive vs. continuant. Morphological conditioning including idiosyncratic conditioning is also attested, and takes various forms. Finally, the phonologist’s choice between a synchronic analysis in terms of lenition or fortition, in several common sets of circumstances, will not follow simply from the empirical facts of the alternation plus parsimony.

12.3 Assimilation Assimilation is possibly the most common (morpho)phonological process in human language (Brohan and Mielke 2018: 210–11), though its prevalence in Australia has been characterized in strikingly divergent terms. Dixon (2002: 619, 623) has claimed that both consonantal and vocalic assimilation are ‘pervasive’ in Australian languages, whereas Baker (2014: 170, 175) finds consonantal assimilation to be ‘rare’ and vocalic assimilation to amount to but ‘a few scattered instances’. Evans (1995a) discusses dissimilation but does not mention assimilation. Perhaps the best response is to contextualize these claims with empirical evidence. In this section I survey assimilation as found in the 118 Australian languages in the AA dataset. For reasons of space, the study is confined to local assimilation between adjacent consonants, and between vowels in adjacent syllables. As in Section 12.2, the study here is of morphophonological alternations, not allophony, and it reports obligatory processes, not optional ones. Such alternations appear in 73 of the AA languages. Assimilation between consonants is examined in Section 12.3.1 and between vowels in Section 12.3.2. For reasons of space, assimilation between consonants and vowels is not addressed here.

12.3.1 Assimilation between consonants For the purposes of the investigation, I define consonant assimilation as an interaction between two segments in which one becomes more similar to the other, and neither deletes, thus it covers cases like /n+p/ → /nm/ but not /n+p/ → /m/. ‘More similar’ is defined simply: the assimilating target consonant comes to have either the self same place of articulation, or the self same manner of articulation, or both, as the trigger. This excludes more abstract assimilations such as assimilation of continuancy (which

morphophonology could describe lenition, for example) or assimilation of nonretroflexion without full place assimilation (this is an interesting phenomenon in Australian languages, but beyond the scope of this study). Section 12.3.1.2 covers place of articulation assimilation, Section 12.3.1.3 manner assimilation, and Section 12.3.1.4 the assimilation of both simultaneously. Section 12.3.1.1 sets out some initial hypotheses to test against the data.

of major class lexical items (verbs and nouns/nominals) and for C2 I searched for suffix-initials. This is a crude measure on multiple fronts, but it provides some information with which to adjust and improve our expectations. Apical Laminal Dorsal Labial 0.0

12.3.1.1 Hypotheses based on static phonotactics Chapter 11 discusses some strong constraints whose effects are found across the Australian continent, on permissible sequences of consonants in clusters. A sensible hypothesis is that consonant assimilation will often function to convert underlying clusters that violate those constraints into surface clusters that obey them. The essence of these constraints is summarized in (8) and (9): both manners and places of articulation tend strongly to be arranged from left to right in the sequences shown. (8) Left-to-right order of superlaryngeal places: Apicals > Laminals > Dorsals > Labials (9) Left-to-right order of manners: Glides > Liquids > Nasals > Obstruents > Glides A corresponding set of hypotheses then, is that when two consonants C1 +C2 combine, the relative likelihoods of places and manners to undergo assimilatory change, in order to satisfy the constraints in (8) and (9) are those shown in (10–13). (10) C1 most-to-least likely to change place: Labials > Dorsals > Laminals > Apicals (11) C2 most-to-least likely to change place: Apicals > Laminals > Dorsals > Labials (12) C1 most-to-least likely to change manner: Obstruents > Nasals > Liquid > Glide (13) C2 most-to-least likely to change manner: Liquid > Nasals > Obstruents > Glide Of course, (10–13) are hypotheses about relative likelihoods that would be appropriate under the assumption that all possible underlying clusters C1 +C2 are equally likely. That is not true though, and accordingly our expectations about how frequently we will actually see each kind of change needs to be tempered by information about the frequency with which various underlying C1 +C2 clusters actually exist. That information is not straightforward to gauge. However, we can make an attempt at approximating it in a similar fashion to that employed in Section 12.2.1 earlier, namely, for C1 I searched the lexical database for stem-final segments

0.2 0.4 0.6 Proportion of stems

Figure 12.12 Superlaryngeal places of articulation of stemfinal consonants. Dorsal Labial Laminal Apical 0.0

0.4 0.2 0.6 Proportion of suffixes

Figure 12.13 Superlaryngeal places of articulation of suffixinitial consonants. Nasal Liquid Obstruent Glide 0.0

0.4 0.2 0.6 Proportion of stems

Figure 12.14 Manners of articulation of stem-final consonants. Obstruent Nasal Glide Liquid 0.2 0.0 0.4 0.6 Proportion of suffixes

Figure 12.15 Manners of articulation of suffix-initial consonants.

Figures 12.12–12.15 show the results of lexical searches, and their implications for hypotheses (10–13) can be summarized as follows. Figure 12.12 tempers the expectations about C1 place of articulation in (10): though labials should be the most likely C1 to change according to (10), they are also the least likely to appear underlyingly according to Figure 12.12, and though apicals should rarely if ever change, they are vastly over-represented underlyingly, so we may in

129

erich r. round fact see them assimilate. Figure 12.13 would temper the expectations about C2 place of articulation in (11) were it not so evenly balanced, but since it is so even, the hypotheses in (11) should essentially stand. Figure 12.14 complicates the hypotheses about C1 manner of articulation in (12). According to (12), glides are unlikely to change and by Figure 12.14 they are unlikely to occur, so we expect to see virtually no C1 assimilation of glides. For liquids and obstruents, (12) and Figure 12.14 push our expectations in opposite directions, so it is not clear what to expect in the data. For nasals, however, (12) and Figure 12.14 both suggest we may see relatively frequent assimilation, but this should be taken with a grain of salt: general phonological typology suggests that nasals rarely undergo manner assimilation. All in all then, we are relatively empty-handed regarding expectations about manner assimilation of C1 . Finally, Figure 12.15 complicates (13) in a similar fashion to Figure 12.14 and (12), and so it is unclear what to expect regarding manner assimilation at all. In sum, though the exercise leaves us with little in the way of hypotheses, it has proven itself worthwhile insofar as it shows that the aprioristic hypotheses in (10–13) are likely too naïve, with the exception perhaps of those regarding place of articulation assimilation of C2 , in (11).

12.3.1.2 Place assimilation When surveying place of articulation assimilation in Australian languages, the data is liable to be swamped by Pama-Nyungan ergative and locative allomorphy. These case markers are reconstructed back to proto-PamaNyungan as having post-consonantal allomorphs /-Cu/ and /-Ca/ respectively, where /C/ is a stop which assimilates in place of articulation to the stem-final consonant to its left (Hale 1976b; Sands 1996). Since both have been inherited into very many modern Pama-Nyungan languages with their assimilatory allomorphy intact, we are best to control for it by removing them from the dataset, which is what I have done. The AA database contains many instances of place assimilation, so it will be possible here to examine reasonably large samples. The unit of measurement will be the number of languages in which a certain pattern occurs. Our interest will be in C1 as a target of regressive assimilation (for which C2 is the trigger), and in C2 as a target of progressive assimilation (with C1 the trigger). Even though we are currently examining assimilation in place, not manner, of articulation, it is possible of course that the manner of C1 and C2 is important in conditioning assimilation in terms of place. For example, given what is known outside of Australia, we might expect to find that place of articulation assimilation

130

Table 12.1 Number of languages with progressive place of articulation assimilation, tabulated according to trigger and target manner. C2 = Target

is particularly common in nasal+stop clusters. In the following I consider the manners of C1 and C2 first, and then their places. For each, I first consider regressive assimilation (with C1 the target) and then progressive (with C2 the target). For regressive assimilation of place of articulation, all 17 instances in the AA database comprise nasal C1 targets assimilating to stop C2 triggers. In progressive assimilation, C2 is the target and C1 the trigger. Table 12.1 tabulates the combinations of C1 and C2 of manners, and counts the number of languages with progressive assimilations instantiated within such clusters. Shading reflects the counts in the cells. Rows and columns are ordered (top to bottom and left to right) according to their totals. As with regressive assimilation, it is nasal+stop clusters that most commonly host progressive assimilation, though in contrast to regressive assimilation there is an additional, relatively wide range of other manner combinations attested in which progressive assimilation takes place. The overall numbers are also higher: 64 instances of progressive as opposed to 17 instances of regressive assimilation. The results for regressive and progressive assimilation are remarkable in a global typological context. In Australia, a nasal+stop cluster is just as likely to host progressive place assimilation as regressive, whereas elsewhere, regressive assimilation is far more common (Brohan and Mielke 2018: 216–17).9 This deserves further study. It may have implications for theories that assume that in all human languages, codas will be more likely to undergo place assimilation than onsets (Itô 1988). Next we examine clusters according to their place of articulation. Since it is place of articulation which is being assimilated, a question arises as to how we can be sure what the underlying place of articulation was of the target, given that on the surface, its place will have changed to match that of the trigger. Here I have inferred the underlying place of 9 Recall too that the Pama-Nyungan ergative/locative allomorphy, which is excluded from these figures, also involves progressive place assimilation.

morphophonology Table 12.2 Number of languages with regressive place of articulation assimilation, tabulated according to trigger and target place.

articulation in the standard way, by asking what it surfaces as in phonological contexts where no trigger is present, e.g. intervocalically or word-finally. In some cases though, the target segment never appears in such trigger-free contexts, so its surface place of articulation is always the same as an adjacent trigger. In these cases, I label the underlying place of articulation ‘Ø’, since it is underdetermined by the empirical evidence. Table 12.2 shows regressive assimilation, in which C1 is the target. Generally, this table presents few surprises. Retroflexes in C1 do not assimilate, since the acoustic cues to retroflexion are heavily skewed to the vowel– consonant transition, which is left unperturbed in C1 position (Hamilton 1996b; Hamann 2003). Labials in C1 are not seen to assimilate, probably largely because few stem-final consonants are labials (Figure 12.12). Coronals in general rarely assimilate to non-coronals in C2 , since coronal + non-corononal clusters are freely permitted in Australian languages; they are more often seen assimilating to other coronals in C2 . Phonotactically in Australian languages, dorsals in C1 are usually permitted only before other dorsal

or labials. In that context, it is interesting to find them assimilating relatively often, even when C2 is labial. Unsurprisingly, the ‘Ø’-place C1 segments are seen assimilating to all C2 places of articulation. A closing caveat is that individual counts are uniformly low, so that any one or two of them may reflect happenstance as much as a principled pattern. Table 12.3 shows progressive assimilation, where C2 is the target. This is the dataset to which the hypothesis in (11) applies: we expect more assimilatory changes in C2 for apicals > laminals > dorsals > labials. This is largely borne out. After targets with the ‘Ø’ place, the next most frequent targets are the apical places (alveolar and retroflex), followed by the laminals (palatal and dental) whose frequencies are close to dorsals, and trailed by the labials. Arguably, this is overlaid with one unexpected pattern. Palatals are both frequent triggers and infrequent targets of progressive place assimilation, and this was true also for regressive assimilation, in Table 12.2. Phonologists familiar with research into acoustic cues to place of articulation and their apparent impact on the phonology of assimilation (Steriade 2001) may be

Table 12.3 Number of languages with progressive place of articulation assimilation, tabulated according to trigger and target place.

131

erich r. round unsurprised, since palatals have strong, distinctive on- and off-glides with high second formants and depressed first formants both in their vowel–consonant and consonant–vowel transitions, and this should make them robustly perceptible in both C1 and C2 positions. That may be so, but it should be recalled that the only attempt at an explanatory account of Australian consonant cluster phonotactics, by Hamilton (1996b), also invokes spectral cues as an explanatory factor, in order to place palatals squarely in the mid-range of ease of perceptibility in clusters—this is why, according to Hamilton, laminals appear in the middle of the ordering sequence in (8). Absent further refinement, it is not possible for both accounts to be correct, and further examination of the issue is called for.

12.3.1.3 Manner assimilation There are few instances of obligatory, morphophonological processes of manner of articulation assimilation in the AA dataset. In Nyigina (Stokes 1982), initial /n/ in certain verbal roots assimilates to a lateral /l/ on the left (14a) while certain prefix-final trills /r/ assimilate to root-initial /l/ on the right (14b). In Marrithiyel (Green 1981), final /r/ in one set of prefixes assimilates to a root-initial /l/ on the right, as in (15). In Kayardild (Evans 1995b; Round 2009), the glide /w/ assimilates to /m/ when preceded by nasals, as in (16). (14) Nyigina (Stokes 1982: 225–7) a. /wa-la-niɡa-na/ ‘3-irr-follow-pst’ → /wa-l-niɡa-na/ → /walliɡana/ b. /ja-r-luɡa/ ‘1-nonmin-cry’ → /jalluɡa/ (15) Marrithiyel (Green 1981: 137, 142) /pir-li/ ‘3aug.fut-aux’ → /pilli/ (16) Kayardild (Round 2009: 703, 710) a. /ʈaman-wari/ ‘tooth-priv’ → /ʈamanmari/ b. /ŋaɳ-wula-i-ȶ-/ ‘beach-subj.abl’ → /ŋaɳmulaːȶ-/

12.3.1.4 Assimilation of both place and manner Likewise, there are very few instances of the simultaneous assimilation of place and manner in the AA dataset. In Nyigina (Stokes 1982) initial /ɻ / in certain verbal roots assimilates in both place and manner to a nasal /n/ or lateral /l/ on the left, as in (17). (17) Nyigina (Stokes 1982: 208–9) a. /ŋan-ɻa/ ‘1sg-spear’ → /ŋanna/ b. /wal-a-ɻa/ ‘2sg-fut-spear’ → /wal-ɻa/ → /walla/

132

In Ngalakgan (Baker 2008b), the future and irrealis suffixes begin with retroflex /ɻ/ (or with /j/ or a homorganic glide) when they follow vowels, but otherwise surface as a lengthening of the preceding stem-final superlaryngeal consonant, a pattern which might be analysed as total assimilation of /ɻ/ as in (18b–e), though see Baker (2008b: 56) for additional analysis and discussion. The assimilation process ignores the presence of glottal stops (18d, e). (18) Ngalakgan (based on Baker 2008b: 53) a. /munku-ɻa/ ‘follow-fut’ → /munkuɻa/ ~ /munku-ja/ ‘follow-fut’ → /munkuja/ b. /puɭ-ɻa/ ‘drown-fut’ → /puɭɭa/ c. /wulup-ɻa/ ‘bathe-fut’ → /wuluppa/ d. /ʈulʔ-ɻa/ ‘set alight-fut’ → /ʈulʔla/ e. /pawunʔ-ɻa/ ‘leave-fut’ → /pawunʔna/

12.3.2 Assimilation between vowels in adjacent syllables Assimilation between neighbouring vowels is widespread in the world’s languages (Brohan and Mielke 2018: 215) and unsurprisingly is also found in Australia. Because Australian languages mostly have small vowel systems, assimilation typically results in the target becoming identical to the triggering vowel. The study here is of local, non-iterative assimilation between two adjacent syllables.10 It does not cover iterative harmony systems (see Evans 1995a: 741–2; Baker 2014: 175–7). I also exclude patterns that are attested only in a single morphological combination, such as just one root in combination with just one suffix. There are 27 languages in the AA dataset with processes that meet the criteria for inclusion. In some patterns, assimilation is triggered only by certain vowel qualities and not others: Section 12.3.2.1 examines front–back assimilation between high vowels to the exclusion of non-high vowels, and Section 12.3.2.2 examines assimilations of /a/ to certain non-low neighbours and not others. In other patterns, the target assimilates to all vowel qualities, copying the trigger irrespective of its quality; Section 12.3.2.3 examines these. As in the studies just above, this study is confined to obligatory, morphophonological assimilation. 10 Accordingly, I also exclude cases where an iterative harmony system circumstantially causes just one target vowel to change, where in the general case more than one can be affected.

morphophonology 12.3.2.1 Front–back assimilation between i and u In Chapter 11 of this volume, it was seen that the lexicons of Australian languages show a tendency towards front–back harmony in sequences of adjacent high vowels, so that sequences /i-i/ and /u-u/ are considerably more common than /i-u/ or /u-i/. Moreover, /i-u/ sequences generally are somewhat rarer than /u-i/. It would be unsurprising then to also find dynamic assimilation between /i/ and /u/, and a somewhat higher incidence of alternations that remove underlying /i-u/ sequences than underlying /u-i/. Obligatory, local (non-iterative) morphophonological assimilation between /i/ and /u/ is found in twelve of the AA languages. Cases just below include both progressive and regressive assimilation, and assimilation that proceeds both from stems onto affixes and from affixes onto stems. Assimilation serves to avoid surface /i-u/ sequences in eight of the languages, and surface /u-i/ sequences in five. Beginning with cases that avoid surface /i-u/ sequences, certain prefixes in Marrithiyel (Green 1981) and Gooniyandi (McGregor 1990: 107) undergo assimilation of /i/ to a following /u/, as in example (19). (19) Marrithiyel (Green 1981: 137, 139) a. ŋi-tin1min.excl.nonfut-see b. ŋi-ȵar1min.excl.nonfut-paint c. ŋu-mun1min.excl.nonfut-go along Jiwarli dative /-ku/ (Austin 2006: 125) assimilates to a preceding /i/, though not across multiple consonants, as in (20). In Arabana-Wangkangurru, suffixal /u/ vowels assimilate to preceding /i/ except across velar consonants. (20) Jiwarli (Austin 2006: 125) a. wiʈa-wu b. t̪ut̪u-wu boy-dat dog-dat c. kiɭki-ji d. ŋaɭir-ku girl-dat barb-dat In Gooniyandi, Warlmanpa, and Wanyjirra, root-final /i/ vowels sometimes assimilate to a following /u/: this is true for one suffix in Gooniyandi, the dative /-wu/ (McGregor 1990: 99), for just one root in Wanyjirra, /kari/ ‘be, stay’ (Senge 2015: 96–7), and for certain nominal suffixes and verb roots in Warlmanpa (Nash 1979a, Section 12.14).

Bularnu (Breen 1978: 15) exhibits assimilation of both /iu/ and /u-i/ sequences. The operative /-ku/ and allative /-lu/ assimilate to preceding /i/, while dative /-ji/ assimilates to preceding /u/. In four more languages high-vowel assimilation acts to avoid surface /u-i/ sequences: assimilation to a preceding /u/ vowel occurs in the Warluwarra purposive /-ȶi/ (Breen 1970: 58), the Pitta-Pitta proximal /=ji/ (Blake 1979a: 193), the noun-stem-forming /-ri/ of Yandruwandha, and the operative /-li/ of the Strzelecki dialect of Yandruwandha and closely related Yawarrawarrka (Breen 2004a: 29, 55).

12.3.2.2 Assimilatory raising of a In five languages, /a/ assimilates to a neighbouring mid or high vowel. In Australian languages that have mid vowels, Chapter 11 showed that lexicons have a statistical preference for a mid vowel to be adjacent to another mid vowel in the next syllable. In Wardaman (Merlan 1994: 41–2), the vowel of allative /-ɭan/ assimilates across a single consonant to a preceding mid vowel, producing a sequence of two mid vowels on the surface, whereas it does not assimilate to high vowels, as in (21a–e). By the same token, the /a/ vowel of the future suffix follows a different pattern. It raises to /e/ after either of the front vowels /e, i/, and assimilates across multiple consonants, but does not assimilate to the back vowels /o, u/, as illustrated in (21f–k). (21) Wardaman (Merlan 1994: 41, 42, 53, 157, 180, 192) a. kaŋka-ɭan b. peje-ɭen c. ȶoŋo-ɭon upriver-all downrive-all east-all d. ʈami-ɭan e. koroŋ-lan f. ŋa-ga-wa here-all south-all 1sg-take-fut g. ŋa-we-we h. ŋa-ȶiŋi-we i. ŋa-pu-jiŋ-pe 1sg-fall-fut 1sg-sit-fut 1sg-hitrefl-fut j. ŋan-wo-wa k. ŋa-pu-wa 3sg>1sg-give-fut 1sg-aux-fut In four languages, /a/ assimilates completely to adjacent /u/ under certain morphological conditions: in certain classes of verb roots before a following future suffix in Ritharrngu (Heath 1980a: 29), in the roots /muɭa/ ‘this’ and /jala/ ‘that’ before following ergative and dative suffixes in Bilinarra (Meakins and Nordlinger 2014: 70–1), in one root /kara/ ‘keep’ before suffixes whose first vowel is /u/ in Wanyjirra (Senge 2015: 96–7), and in the second person augmented future prefix before auxiliary roots whose first vowel is /u/ in Marrithiyel (Green 1981: 136–7).

133

erich r. round 12.3.2.3 Copying of the trigger vowel, irrespective of its quality In seven of the AA languages, target vowels assimilate and copy the quality of a neighbouring trigger irrespective of its quality.11 In most instances, these target vowels are only ever found in the context of triggers, and so have no ‘elsewhere’ form to speak of, though in Gooniyandi three classifier roots have vowels which, though they typically copy the final vowel of a preceding prefix, will surface by default as /a/ if no prefix is present or when followed by /w/ (McGregor 1990: 196–8). In six languages, suffixes are found whose first vowel only ever copies its quality from the final vowel of the stem: the dative–genitive /-ʔkVn/ and relativizer /-kVn/ in Ngalakgan (Baker 2008b: 145), the ergative–locative–instrumental /-(ŋ)kV/ and topicalizer /-n̪V/ in Ngawun (Breen 1981a: 40-1, 49), the future continuous subjunctive /-C(C)Vr/ in Alawa (Sharpe 1972: 42, 52), the ergative–locative /-kV(l)/ and purposive–allative /Vnp(ik)/ in Kurtjar (Black and Gilbert 1996: 14–15), and the genitive /-nVkan/ in Wambaya (Nordlinger 1998a: 81).

12.3.3 Summary and discussion It makes sense to compare the patterns of assimilation in Australian languages to constraints on the phonotactics of surface forms. However, Section 12.3.1.1 showed that an attempt to formulate reasonable hypotheses in this respect is complicated by the uneven distributions of underlying forms. Even so, Australian languages deliver some stark results. Manner assimilation of consonants is rare. Place assimilation is most common in nasal+stop clusters, but in a notable departure from broader cross-linguistic norms, progressive assimilation is about as common as regressive. Outside of nasal+stop clusters, place assimilation is always progressive in the languages examined. Instances of local, non-iterative vowel assimilation were encountered in one fifth of the AA languages, though no one type was particularly common. Assimilation among high vowels, assimilatory raising of /a/, and total assimilation of targets to triggers were each attested in ten percent or fewer of the 118 languages examined. Before closing, it is worth restating that the alternations examined here were morphophonological, not allophonic, and importantly, were obligatory and not optional. The AA dataset contained 73 languages with such alternations, but it also contains no fewer than 36 languages with optional morphophonological assimilation, in 11 I do not include suffixes whose effect is to lengthen the final short vowel of a preceding, vowel-final stem, though these cases are sometimes analysed as affixation of a single short vowel which copies its quality from the stem (e.g. Blake 1979b: 18 for Kalkatungu; Breen 1973: 97–8 for Bidyara and Gungabula).

134

the sense that the alternation substitutes one contrastive sound for another, but does not do so obligatorily. Explaining why this is so is a task for future research.

12.4 Conclusion Previous reviews of Australian morphophonology have always had only a short format, and with it, a limited capacity to explore and reveal the empirical diversity in Australian phonologies. Here I have conducted two quantitativelybacked studies that reveal a new degree of detail, and at times surprising key generalizations about Australian morphophonology. A third study can be found in Chapter 13 of this volume. Taken together, the studies demonstrate that with modern approaches to phonological diversity, there is still much that can be learnt about Australian languages, and yet more that remains to be explained.

12.5 Language sample The following languages have one or more alternations referred to in the studies above, indicated as [L]enition and [A]ssimilation. Pama-Nyungan Central NSW: Gamilaraay [LA] (Giacon 2014), Muruwari [L] (Oates 1988). Dyirbalic: Wargamay [A] (Dixon 1981). Kalkatungic: Yalarnnga [L] (Breen and Blake 2007). Karnic: Yandruwandha [A] (Breen 2004a), Yawarrawarrka [A] (Breen 2004a), Arabana-Wangkangurru [A] (Hercus 1994), Pitta-Pitta [A] (Blake 1979a). Kartu: Yingkarta [A] (Dench 1998c). Kulin: Wathawurrung [A] (Blake, Clark, and Krishna-Pillay 1998). Maric: Margany [A] (Breen 1981b), Warungu [LA] (Tsunoda 2011). Marrngu: Nyangamarta [A] (Sharp 2004). Mayi: Ngawun [A] (Breen 1981a). Ngumpin: Bilinarra [LA] (Meakins and Nordlinger 2014), Gurindji [L] (McConvell 1988), Wanyjirra [LA] (Senge 2015), Jaru [LA] (Tsunoda 1981), Walmajarri [L] (Hudson 1978; Hudson and Richards 1969). Paman (Middle): Kugu Nganhcara [A] (Smith and Johnson 2000). Paman (Northern): Angkamuthi [L] (Crowley 1983), Atampaya [A] (Crowley 1983), Yadhaykenu [A] (Crowley 1983). Paman (South-West): Yir-Yoront [A] (Alpher 1991). Mantharta: Jiwarli [LA] (Austin 2006). Ngayarta: Yindjibarndi [LA] (Wordick 1982). Queensland-NSW coast: Yugambeh [A] (Sharpe 2005), Gidabal [A] (Geytenbeek and Geytenbeek 1971), Batyala [A] (Bell 2003), Duungidjawu [L] (Kite and Wurm 2004). Thura-Yura: Wirangu [L] (Hercus 1999), Kukata [A] (Platt 1972). Warluwaric: Bularnu [A] (Breen 1978), Warluwara [A] (Breen 1970). Warumungic: Wati Wangkajunga [A] (Jones 2011), Pintupi [A] (Hansen and Hansen 1978), Ngaanyatjarra [A] (Glass and Hackett 1970).

morphophonology Yapa Warlmanpa [A] (Nash 1979a), Warlpiri [A] (Nash 1986). Yolngu Ga¨lpu [L] (Wood 1978), Djambarrpuyngu [LA] (Wilkinson 1991), Ritharngu [LA] (Heath 1980a). YuinKuri Dhanggati [A] (Lissarrague 2007). Other families Bunuban: Bunuba [LA] (Rumsey 2000), Gooniyandi [LA] (McGregor 1990). Daly-Western: Marrithiyel [A] (Green 1981). Darwin group: Larrakia [LA] (Capell 1984), Limilngan [LA] (Harvey 2001). Garrwan: Garrwa [L] (Mushin 2012a). Giimbiyu: Erre [L] (Campbell 2006), Mengerrdji [L] (Campbell 2006), Urningangg [L] (Campbell 2006). Gunwinyguan: Gaagudju [LA] (Harvey 2002), Ngalakgan [A] (Baker 2008b). Tiwi (isolate) [A] (Osborne 1974). Iwaidjan: Amurdak [L] (Handelsmann 1991), Maung [LA] (Teo 2007), Iwaidja [LA] (Teo 2007). Jarrakan: Miriwung [LA] (Kofod 1978). Maran: Mara [L] (Heath 1981a), Alawa [LA] (Sharpe 1972), Mangarrayi [LA] (Merlan 1989). Mirndi: Jaminjung [LA] (Schultze-Berndt 2000), Ngaliwuru [LA] (Bolt, Hoddinott, and Kofod 1971), Jingulu [A] (Pensalfini 2003), Wambaya [LA] (Nordlinger 1998a). Nyulnyulan: Nyigina [LA] (Stokes 1982), Warrwa [L] (McGregor 1994c), Bardi [LA] (Bowern 2012a). Tangkic: Kayardild [LA] (Round 2009). Worrorran: Ungarinyin [LA] (Rumsey 1982b), Wunambal [LA] (Carr 2000), Worrorra [LA] (Clendon 2000). Yangmanic: Wardaman [LA] (Merlan 1994).

Acknowledgements Support for this research over many years is gratefully acknowledged from the Australian Research Council (grant DE150101024), the British Academy (grant VF1_101602), the ARC Centre of Excellence for the Dynamics of Language (grant TIG322015), the National Science Foundation (grant BCS-0844550 to C. Bowern and Yale Unviersity) as well as Yale University, the University of Queensland, Surrey Morphology Group, and the Max Planck Institute for the Science of Human History. For their enthusiasm for Australian phonology and diligent work on AusPhon-Alternations, AusPhon-Lexicon and their predecessors, my thanks to Paula Abul, Cicely Bonnin, Madhawee Fernando, Jordan Hollis, Simone Hubbucks, Phibe Keung, Edith Kirlew, Gretel Macdonald, Sharon Mak, Catherine Moller, Jayden MacklinCordes, Muhib Nabulsi, David Osgarby, Amy Parncutt, Natasha Pepi, Lawrence Pow, Pascal Roth, Jackie Van Den Bos, Glenn Windschuttel, Sophie Whitton, Joanne Yang, Jia Ying, and especially Jacqui Cook and Tom Ennever. For helpful comments on the chapter and for enabling this research to get started a decade ago at Yale, sincere thanks to Claire Bowern.

135

chapter 13

Nasal cluster dissimilation Erich R. Round

One of the more intriguing phenomena in Australian phonologies is nasal cluster dissimilation (NCD), in which two nasal+stop (NC) clusters interact in such a way that one cluster dissimilates, to become something other than nasal+stop. For example, in several Ngumpin languages such as Wanyjirra (Senge 2015), a cluster of /n/+stop undergoes optional denasalization to /t/+stop when preceded by an NC cluster, as in (1), while in Yankunytjatjara (Goddard 1985) an NC cluster, if it is preceded in the word by at least two morae, will optionally lose its stop if the next cluster to the right is nasal+/k/, as in (2).

13.1 Progressive NCD nasal deletion in Gurindji

(1) Wanyjirra (Senge 2015: 91) a. ɲaŋka=n-ku=la ~ ɲaŋka=t-ku=la cond=2.min.sbj-link=3.min.dat b. ɲampa-wu=npula ~ ɲampa-wu=tpula what-dat=2.ua.subj

(3) Gurindji (McConvell 1988b: 137, 138, 140) a. luȶu-ŋka b. wiȵȶi-ka c. pinka-ka ridge-loc spring-loc river-loc d. kajira-mpal e. kaȵȶu-pal north-across down-across f. ȵampa=ta what=2.pl.subj

(2) Yankunytjatjara (Goddard 1985: 16, 77) a. ɲuntu-mpa-nku ~ ɲuntu-ma-nku 2sg-gen-reflex b. kaːɳka-ŋku ~ kaːɳa-ŋku crow-erg The focus of this chapter is on a third kind of NCD, in which the second of two NC clusters loses its nasal. This third kind is an order of magnitude more common than any other NCD pattern in Australian languages, with at least 25 Australian languages exhibiting it, and this chapter provides the missing typological overview which has been called for since McConvell’s original study, primarily of languages in the Ngumpin family (McConvell 1988b). Section 13.1 begins with the classic account of progressive NCD nasal deletion in Gurindji by McConvell (1988b). Sections 13.2–13.6 then examine variation in the phenomenon found in a close to exhaustive sample of Australian languages: Section 13.2 enumerates the dimensions along which the data varies, Section 13.3 examines NCD triggers, Section 13.4 NCD targets, Section 13.5 the application of NCD over long distances and in multiple locations, and Section 13.6 the interactions of NCD with other processes. A summary and discussion are in Section 13.7.

In Gurindji (McConvell 1988b) the locative suffix /-ŋka/, the across suffix /-mpal/, and the second person plural subject enclitic /=nta/, among many others, lose the N from their NC cluster when attached to a base which also contains an NC cluster as in (3b, c, e, f). This is true whether the triggering NC cluster is homorganic (3b) or heterorganic (3c), and the generalization applies without exception to all suffixes and clitics.

NCD deletion can also target NC clusters that sit internal to an affix, and away from its edge, as in (4b) where the trigger is a NC cluster formed across the suffix’s left edge, as well as in (4c) where the trigger is more distant. The suffix illustrated in (4) is comitative /-kuȵȶa/, whose initial /k/ undergoes a separate process of lenition to /w/ when intervocalic (McConvell 1988b: 139–40). (4) Gurindji (McConvell 1988b: 139) a. ŋaȶi-wuȵȶa b. ŋaɻin-kuȶa c. paɳku-wuȶa father-comit meat-comit FZD-comit More broadly, NCD deletion can operate at a distance if the trigger and target are separated only by continuants, that is, vowels, liquids, or glides. This includes glides derived by lenition as in (4c). In (5a) the target 2pl.subj /=nta/ becomes /=ta/ due to an NC trigger /mp/ located five syllables to its left. If a word contains a chain of potential triggers and targets, NC…NC…NC, then in Gurindji the second NC undergoes NCD deletion, but not the third, as in (5b) where the target loc /-ŋka/ undergoes NCD deletion due to the preceding trigger /ȵȶ/, but the following clitic 2pl.subj /=nta/ does not. Oral stops and nasals block NCD deletion, as in (5c, d):

Erich R. Round, Nasal cluster dissimilation. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Erich R. Round (2023). DOI: 10.1093/oso/9780198824978.003.0013

136

nasal cluster dissimilation (5) Gurindji (McConvell 1988b: 140–1) a. ȵampa-wu-waɭa-ji=ta what-dat-now-1.sg.obj=2.pl.subj b. waȵȶi-ka=nta which-loc=2.pl.subj c. ŋampijita-wuȵȶa female-lacking d. ŋanta-ɳa=ŋku optative-1.sg.subj=2.sg.obj Lenition, which converts underlying stops to glides, feeds NCD deletion. For example, in (5a) the dative suffix /-wu/ which sits between the trigger and target derives by lenition from underlying /-ku/, and NCD applies across it. Conversely, NCD deletion counterfeeds lenition. The derived intervocalic /k/ in (3b, c) for example, does not lenite to /w/. Example (4b) shows that the N and C of the NC trigger of NCD deletion need not be tautomorphemic. The target, however, cannot span two morphs in Gurindji. Thus, NCD deletion applies neither in (6a) nor in (6b). Anticipating the cross-linguistic results, I know of no cases in which NCD deletion applies in circumstances like (6a), where the nasals of the trigger and the would-be target are tautomorphemic. However, there are languages in which NCD deletion applies in the equivalent of (6b), where they are in different morphs. (6) Gurindji (McConvell 1988b: 137, 144) a. nuŋkijiŋ-ku, *nuŋkiji-ku relation-dat b. kaɳʈi-yawuŋ-kuɭu, *kaɳʈi-yawu-kuɭu stick-prop-erg

13.2 Dimensions of cross-linguistic variation McConvell (1988b) fleshes out several dimensions of crosslectal diversity in NCD deletion. Two dialects of Gurindji are mentioned in which the constraints on long-distance NCD deletion are somewhat relaxed, in the first by permitting it to occur across intervening /p/, though not other stops; and in the second across /p, k/ but not other stops. Referring to NCD deletion in other Ngumpin languages and in languages farther afield, McConvell identifies several further dimensions of variation. In Gurindji, NC targets are homorganic and may have any place of articulation; in other languages the places of articulation that the target can have may be restricted. Gurindji places no idiosyncratic morphological restrictions on which suffixes and clitics are potential targets, but other languages may. In Gurindji, NCD deletion is possible at a distance though is blocked by certain segments; cross-linguistically, the class of blockers varies, and some languages do not permit NCD deletion at a distance,

rather the trigger and target must be separated by no more than one syllabic nucleus. McConvell also reports potential variation in directionality, in a possible, regressive NCD process in the Ngumpin language Mudburra. Subsequent to McConvell (1988b), short summaries of NCD have appeared in Evans (1995b: 733–4), Blevins (2001a: 32–4), Dixon (2002: 625–6), and Baker (2014: 195–6). Each refers to a subset of McConvell’s dimensions of variation, though Dixon adds that languages may also vary in the ordering of lenition with respect to NCD deletion. In this study I expand the typology, by also examining: whether there are morphologically idiosyncratic triggers; what places of articulation triggers can have; whether targets can span two morphs; what the outcome is of chains of NC…NC…NC; whether NCD deletion is obligatory or optional; and how it is ordered with respect to lenition as well as other processes that may be relevant to it. The aim here has been as complete a coverage as possible of all Australian languages with published, productive NCD deletion processes. The 20 languages identified as meeting the criteria are given with their sources in Section 13.6; in a further five languages NCD exists but the available information was insufficient to contribute meaningfully to the current survey.1 The results that follow are based on the sources’ direct reports about NCD, plus searches of data elsewhere in the source documents.

13.3 Triggers of NCD deletion 13.3.1 Heterorganic triggers In all languages with NCD deletion, triggers can be homorganic, and in none of the 20 is there a demonstrable lack of heterorganic triggers. Blevins (2001a: 32) reports that in Nhanda, items containing heterorganic NCs exhibit lexical idiosyncrasy in their triggering behaviour: whereas /paɳpa/ does trigger NCD deletion, /t̪uɳpa/ ‘dove’ and /wuɲpa/ ‘whistle’ do not. Westerlund (2015: 13) reports that triggers in Ngarla must be homorganic, citing (7c), however the heterorganic example in (7d) also appears twice, where NCD deletion is triggered. The variation in Ngarla possibly reflects lexical idiosyncrasy for heterorganic triggers, like that reported in Nhanda, or alternatively, a degree of optionality. In three further languages, Kurtjar, Kurrama, and Ngarluma, there is no available data for assessing the triggering behaviour of heterorganic NC; in all other languages heterorganic NCs are triggers, much as we saw in Gurindji. 1 In the Ngumpin family, Ise (1999) provides information on Malngin, and McConvell (1988b) mentions a few facts of Mudburra. In both languages, NCD deletion affects multiple suffixes. In the Ngayarta family, Wangka Maya (2001a) establishes that Kariyarra’s locative suffix /-ŋka/ undergoes NCD deletion. In Thura-Yura, Hercus (1999: 112) mentions that Kuyani’s durative–present suffix /-nta, -n̪t ̪a/ undergoes NCD deletion. In Karnic, Austin (1981a) mentions that NCD deletion in Yarluyandi affects participle /-ɳʈa/ and the imperfect same subject marker /-ɳʈa/.

137

erich r. round (7) Ngarla (Westerlund 2015: 102, 149, 385, 388) a. wula-ŋka b. waŋka-ka c. ŋaɳku-ŋka water-loc speech-loc beard-loc d. ȶunku-ka back-loc

13.3.2 Nasal+nasal clusters as non-triggers All languages in the survey permit consonant clusters of two nasals, but owing to their infrequency I have been unable to locate evidence for whether nasal+nasal clusters trigger NCD deletion in seventeen of the twenty languages. In the cases where evidence does exist, nasal+nasal clusters are not triggers, in Yindjibarndi (Wordick 1982: 35), Yanyuwa (Yanyuwa Families and Bradley 2017 e.g. jarranyma-ntharra ‘inciting’, p.215) and Mara (Heath 1981a: 233).

13.3.3 Morphological idiosyncrasy of triggers Morphological idiosyncrasy among NCD triggers is rare. In Nhanda, only stems that contain a heterorganic NC cluster behave idiosyncratically, while heterorganic NC always acts as a trigger. In Yalarnnga, Breen and Blake (2007: 15) report that only two stems trigger NCD deletion in the ergative suffix /-ŋku/: ‘mintji “back” (possibly Kalkutungu) and kuntji “tail”’. They add though, that ‘this is not consistent’. Since Breen and Blake (2007) contains no other instances of /-ŋku/ after a stem containing NC, it is unclear how extensive the lexical idiosyncrasy might be, or whether the data could be adequately accounted for in terms of simple optionality. NCD deletion of the Yalarnnga purposive /-ȵȶata/ is entirely regular. None of the other 18 languages exhibit morphological idiosyncrasy of NCD triggers.

13.3.4 Heteromorphemic triggers NC clusters can be affiliated with a single morph, or in some cases, span two. In three of the languages, morphs cannot end in nasals, and so heteromorphemic NC clusters are not possible. In a further 13, the available data does not contain any instances of NC clusters spanning two morphs and followed by potential NCD targets, and so it is unclear whether heteromorphemic NC can be an NCD trigger. In the remaining four languages, where evidence does exist, heteromorphemic NC clusters do trigger NCD deletion: in Gurindji (cf. (4b) above), Wanjyirra (Senge 2015: 90), Djaru (Tsunoda 1981: 195), and Yindjibarndi (Wordick 1982: 34, cf. p.139, fn.1).2 2

In Yindjibarndi, this hinges on whether one posits the first morpheme break in the pronoun /ȵin-ku-puru-ŋu/ ‘2sg-obj-pl-obj’, in which

138

13.4 Targets of NCD deletion 13.4.1 Restrictions on place and individual morphemes In contrast to triggers, the targets of NCD deletion are frequently restricted, both phonologically and morphologically. In no language in the sample does NCD deletion target heterorganic NC, and in most languages homorganic NC clusters are targeted at only some places of articulation. Some languages impose further, idiosyncratic restrictions on individual morphological items, in which case merely knowing the place of articulation of an NC cluster is not sufficient to predict whether or not it will be a target. For each language in the sample, Table 13.1 presents the places of articulation at which NC clusters potentially can be targets of NCD deletion, and the targets themselves. If there are places of articulation where NC clusters are not targets, one example item is provided, and if there are morphologically idiosyncratic non-targets (at the same place of articulation as targets) they are listed. When evaluating forms’ status as non-targets, I have assumed that the same restrictions hold, with respect to long-distance application and blockers, as is evident from demonstrable targets. For example, if a language never exhibits long-distance NCD deletion, then I only deem a form to be a non-target if it fails to undergo NCD deletion when its NC cluster is separated from a wouldbe trigger by just one syllable nucleus. The focus here is on affixes and clitics. Roots occasionally do function as NCD deletion targets, for example in reduplication, but since the information in sources is sparse, I do not pursue the topic of roots as targets here. Sources’ categorizations of forms as affixes or clitics are taken at face value. A future, more finegrained evaluation might uncover further generalizations, since in some languages, the suffix/clitic divide is also a divide between NCD targets and non-targets. For example, it may be that in some languages, what appears here as morphological idiosyncrasy is actually more principled, once the details of morphological and prosodic word structure are taken into account. In the 20-language sample, 13 languages restrict the places of articulation at which NC clusters can be targets. For three languages the available information is inconclusive, which leaves only four that demonstrably apply NCD deletion to NC targets irrespective of their place of articulation. Eight languages impose further morphologically idiosyncratic restrictions, and for two the evidence is inconclusive, leaving only half in which morphological idiosyncrasy is not a factor. /-mpuru/ has undergone NCD deletion to /-puru/. If not, then Yindjibarndi is one more language that lacks available evidence for whether heteromorphemic NC is a trigger.

nasal cluster dissimilation

Table 13.1 NCD nasal deletion (non)targets, by places of articulation and items. Language

Target PoA’s*

Target items

Non-target items (at targeted PoA’s)

Non-target PoA (example item)

Kurtjar

D, P

erg/loc -Vn̪t,̪ -Vȵȶ

doer -ȵȶer

erg/loc -ŋku

Yanyuwa

D, P, V, L

direc -ŋku, abl -ŋka, loc -ampa, verbal TAM beginning -n̪t,̪ -ȵȶ

—

—

Garrwa

V

pres =ŋka, refl -ŋka; refr -ŋkaȵi, all -ŋkuru

—

loc -ntu

Yalarnnga

P, V

erg -ŋku, purp -ȵȶata

loc -ŋka

all -mpa

Kalkatungu

D, P, V

erg -ŋku, habit -ȵȶaŋu, purp -ȵȶaːni, purp -ȵȶaːja, trans -ȵȶama, pl -n̪ti̪ ti

impf -man̪ti̪

perf -mpa

Bilinarra

V, L

erg -ŋku, loc -ŋka, all -ŋkura, edge -mpa

2min.obj =ŋku

2aug.subj =nta

Gurindji

All

all

—

—

Wanyjirra

A, P, V, L

erg -ŋku, loc -ŋka, all -ŋkawu, lack/comit -kuȵȶa, loc -mpa, 2min.obj =ŋku, 2aug.subj =nta

prop -yaɻuŋ, priv -muluŋ, other -waɻiȵ, pers -jaŋ, d.term -nun, imp -ta

(inconclusive)

Djaru

A, P, V

erg/inst -ŋku, loc -ŋka, all -ŋkawu, lack -kuȵȶa, other -kuȵ, imp -ŋka, 2sg.acc =ŋku, 2pl.nom =nta

other -kaɻiȵ

—

Nhanda

V

erg/inst -ŋku, loc -ŋka, amb -ŋkula, path -ŋkalu

all -ŋku

caus -nta

ArabanaWangkangurru

R

pres -ɳʈa, speed -ɳʈa, dat/all/ loc -kiɳʈa

—

refl -nta

Dhirari

R

ptcp -ɳʈa, seq.ss -ɳʈantu, rel.ss -ɳʈa, rel.ds -ɳʈan̪i

(inconclusive)

prop -n̪tu̪

Mara

R, P

pst.cont -ȵȶi, pres -ȵȶi,-ȵȶa, pres -ɳʈu

—

—

Gooniyandi

V

erg -ŋka

2sg.acc ŋki-

(inconclusive)

Panyjima

V

agt -ŋku, loc -ŋka

—

int =nta

Ngarla

V

erg -ŋku, loc -ŋka

—

rel.ds -ntaŋu

Kurrama

V

inst -ŋku, loc -ŋka

—

int -nta

Ngarluma

V

loc -ŋka

—

priv -jun-

Yindjibarndi

V, L

inst -ŋku, loc -ŋka, pl -mpuru, top =mpa

—

priv -pun̪ta̪ ri

pres -nta, -n̪ta̪ , fut -ȵȶu

(inconclusive)

(inconclusive)

Adnyamathanha A, D, P

* A(lveolar), R(etroflex), D(ental), P(alatal), V(elar), L(abial)

139

erich r. round

13.4.2 Heteromorphemic targets Most languages in the sample are uninformative regarding whether heteromorphemic NC clusters can be targets of NCD deletion. In nine, this is because phonotactic and morphotactic restrictions conspire to preclude any potential heteromorphemic NC targets from arising at the places of articulation that the language otherwise targets. In a further seven languages, potential heteromorphemic NC targets notionally could exist, but I have not found any diagnostic examples. In four languages, it is clear that heteromorphemic NC clusters are not targets of NCD deletion if the nasals in the would-be trigger and target clusters are tautomorphemic. This was seen in Gurindji in (6a). It is shown in Bilinarra in (8c, cf. 8a, b), and below in Wanyjirra (9j) and Djaru (10d). All four of these languages are from the Ngumpin family. (8) Bilinarra (Meakins and Nordlinger 2014: 73, 255) a. kaɻu-ŋku b. kaɳʈi-ku child-erg stick-erg c. naɭampaŋ-ku, *naɭampa-ku, crocodile-erg In Wanyjirra (Senge 2015), heteromophemic NC can be a target of NCD deletion if the target and trigger nasals are in different morphs, as in (9b, c). However, even when these conditions are met, there are some suffixes which still idiosyncratically do not undergo NCD deletion. The final nasal of the origin suffix undergoes NCD deletion (9b, c, cf. 9a), as does the final nasal in the /aŋ/ allomorph of the continuative (9e, cf. 9d). However, final nasals of disyllabic suffixes do not undergo NCD deletion under comparable circumstances (9g, h, i), and nor does the final /n/ of the /an/ allomorph of the continuative (9f). (In Wanjirra, the only blockers of long-distance NCD deletion are oral stops.) (9) Wanyjirra (Senge 2015: 128, 129, 139, 244, 256, 258, 318, 500) a. yalu-wuȵ-ȶawu b. ŋantu-wu-ȶawu dist2-orig-all someone-orig-all c. mawun-ku-ȶawu d. ŋaɳ-aŋ-ku man-orig-all ingest-cont-fut e. wantiȵ-a-ku f. junpaɳ-an-ta fall-cont-fut sign-cont-imp g. ȵampa-jaɻuŋ-kulu h. ȵantu-waɻiȵ-ȶui. what-prop-erg 3.min-oneself-erg i. ŋapaȵȶi-muluŋ-kulu j. nunkijiŋ-kulawu, eye-priv-erg *nunkiji-kulawu family-all In Djaru (Tsunoda 1981), less information is available on heteromorphemic targets, but the monosyllabic ‘from’ suffix is an optional undergoer of NCD deletion (10b), while disyllabic ‘other’ is not (10c).

140

(10) Djaru (Tsunoda 1981: 49, 67, 517) a. wiki-wuȵ-ȶa b. child of man-from-loc c. mawun-kaɻiȵ-ȶa d. man-other-loc

mawun-ku(ȵ)-ȶa man-from-loc ȵantuwuȵ-ȶa, *ȵantuwu-ȶa who-loc

13.4.3 Optionality In (10b) we saw that the ‘from’ suffix in Djaru is an optional undergoer of NCD deletion. Optional NCD deletion is rare in the sample, found only in the Ngumpin languages Djaru and Wanyjirra, and in Yalarnnga (Kalkatungic). In Djaru, NCD deletion applies optionally to the ‘from’ suffix, to pronominal clitics (11b, cf. 11a), within verb complexes (11c) and in the Northern dialect to the privative suffix. It applies obligatorily to all other suffixes (Tsunoda 1981: 49, 179). In Wanyjirra, NCD deletion applies optionally to clitics, and is obligatory otherwise (Senge 2015: 90). As noted above (Section 13.3.3), in Yalarnnga, the application of NCD deletion is ‘not consistent’ in the ergative suffix /-ŋku/ (and is triggered only by two stems). It is obligatorily for the Yalarnnga purposive /-ȵȶata/ (Breen and Blake 2007: 15). (11) Djaru (Tsunoda 1981: 49, 179, 423) a. ŋara=ŋku b. ȵaŋka=(ŋ)ku possibly=2sg.acc if/when=2sg.acc c. /kuȵ buŋ-ka/ → kuȵ bu(ŋ)ka waiting hit-imp

13.5 Distant and multiple application 13.5.1 Distant application Depending on the language, and in some cases on the specific target, NCD deletion may be restricted to applying only locally, in structures where the trigger and target are separated by a single syllabic nucleus. Alternatively, it may apply at a distance. Twelve languages in the sample permit only local NCD deletion. In the Ngayarta languages Panyjima, Kurrama, and Ngarluma, this is trivially the case, due to a combination of morphotactic and phonotactic factors: the only suffixes containing homorganic NC are case allomorphs that have the shape NCV and that attach solely to bimoraic stems, and the only position where an NC cluster can appear in a bimoraic stem is across the boundary of the first two syllables, which always places it at a distance of one syllabic nucleus from the suffix that follows. Aside from these trivial cases, there are a further nine languages which always restrict NCD deletion to applying locally, as illustrated in Ngarla (12) and Kurtjar (13).

nasal cluster dissimilation (12) Ngarla (Westerlund 2015: 85, 126, 151) a. maȴa-ŋku b. ȶinta-ku c. maȴa-ȵȶari-ŋku father-erg others-erg father-pl-loc (13) Kurtjar (Black 1986: 54, 58, 78, 94) a. neːrim-an̪t ̪ b. ɻiŋk-at̪ fingernail-erg cockatoo-erg c. mpiŋ-an̪t ̪ d. jeβaȵȶik-an̪t ̪ lancewood-erg navel-erg In the Karnic language Arabana-Wangkangurru (Hercus 1994a), NCD deletion can apply non-locally to the verbal present tense and ‘speed’ suffixes /-ɳʈa/ (14a–d) but only locally to allative /-kiɳʈa/ (14e–h). The facts here are somewhat complicated, and I return to them in the discussion of blockers, below (Section 13.5.2). (14) Arabana-Wangkangurru (Hercus 1994a: 58, 64, 110, 111, 112, 125, 169, 179) a. mama-ɳʈa b. kaŋkara-ʈa grab-pres burp-pres c. pura-ɳʈa-jaŋu d. pun̪ta̪ -a-ʈa-rakaɳa die-sp-plup drink-tr-sp-pass e. uka-kiɳʈa f. an-kiʈa 3sg-all 2sg-all g. aɻimpa-kiɳʈa h. awaɳʈa-kiɳʈa 2du-all that-all In Garrwa (Mushin 2012a), NCD deletion applies at a distance to the reflexive pronominal suffix /-ŋka/ (15a, b) but only locally to the present tense clitic /=ŋka/ (15c–f). (15) Garrwa (Mushin 2012a: 141, 217, 220, 227, 415; Breen 2003: 450) a. pula-ŋka b. nimpala-ka c. naȶpa=ŋka 3du-refl 2du-refl see=pres d. jaȵpa=ka e. wankiȶpa=ŋka f. lalanpa=ka talk=pres dance=pres watch=pres A further six languages allow long-distance NCD deletion for all targets: the four Ngumpin languages in the sample, Kalkatungu (Kalkatungic) and Nhanda (Kartu).

13.5.2 Blockers In some languages, NCD deletion at a distance will be blocked by some class of intervening consonants (blockers). The most liberal case is Kalkatungu, where no segments appear to be blockers. Illustrative examples are in (16c, f). (16) Kalkatungu (Blake 1979a: 19, 88, 90) a. t̪una-ȵȶaːja b. iŋka-ȶaːja c. t̪ua-nti-ji-ȶaja run-purp go-purp cut-with-a/p-purp d. t̪uni-ȵȶaŋu e. aɳka-ȶaŋu f. api-ȵȶama-ti-ȶaŋu run-habit ail-habit sing-tr-refl-habit

In Wanyjirra, sonorant consonants including nasals are nonblockers, as illustrated in (9b, e) and (17c). Oral stops in Wanyjirra could be analysed as blockers, but need not be. Just two examples that I can find are relevant (17d, e). Both involve the personalizer suffix /-ȶaŋ/ and a wouldbe target NC that is heteromorphemic. Wanjirra does allow NCD deletion to apply to heteromorphemic targets (9b, c), and so it would be reasonable to claim that the stop /ȶ/ in (17d, e) is acting as a blocker. However, for heteromorphemic NC, Wanyjirra also has morphologically idiosyncratic non-targets (9f–i), and so it could equally well be argued that /-ȶaŋ/ is one of them. The upshot is that the available data underdetermines the analysis. Wanyjirra perhaps treats oral stops as its sole class of blockers, or perhaps has no blockers. (17) Wanyjirra (Senge 2015: 148, 156, 324, 428, 574) a. ŋawa-wuȵȶa b. ȵampa-wuȶa water-lack what-lack c. paɳʈawuru-wuȶa d. jaŋkaɳi-ȶaŋ-ku back-lack later-pers-dat e. kampari-ȶaŋ-ku front-pers-dat In Arabana-Wangkangurru, NCD deletion can act at a distance, however all available examples have a distance of maximally two syllabic nuclei between the trigger and target, and the only intervening consonants observed are /k/ and the trill /r/, as in (14b) and (18a, b). Since no other consonants are observed in the equivalent intervening position, it is unknown whether they would act as blockers. In all examples of more distant separations, the would-be targets fail to undergo NCD deletion. All of these instances include at least one consonant other than /k/ and /r/ as in (18c–e). Consequently, the data are consistent with two accounts. Either /k/ and /r/ alone are non-blockers (Hercus 1994a: 58), or NCD deletion in Arabana-Wangkangurru acts at a maximum distance of two syllabic nuclei, and is not subject to blocking. (18) Arabana-Wangkangurru a. min̪ti̪ ka-ʈa b. walk about-pres c. wanka-jiwa-ɳʈa d. rise-tr-pres e. paŋki-waru-t̪i-ɳʈa-ki rub-white-become-pres-emp

wanka-ka-ʈa rise-tr-pres manta-t̪ika-ɳʈa take-return-pres

In Gurindji, nasals are blockers, continuants are nonblockers and stops are blockers or not, depending on the dialect and their place of articulation (McConvell 1988b). Data from Bilinarra, Djaru, and Nhanda are patchy but are consistent with this picture, apart from the trill /r/, which

141

erich r. round is a non-blocker in Gurindji but a blocker in Bilinarra and Djaru. Confirmed non-blockers are /l/ in Bilinarra (19b), the glides /w, j, ɻ/ in Djaru (20c, d), and /ɻ/ in Nhanda (21c). Confirmed blockers are /ɳ, ŋ, t, r/ in Bilinarra (e.g. 19c–e), /ɳ, n, ȵ, ŋ, r/ in Djaru (e.g. 20e, f), and /b, d/ in Nhanda (21d, e). (19) Bilinarra (Meakins and Nordlinger 2014: 74, 148 172, 357) a. kaːɳi-mpal b. kankulu-pal east-edge up-edge c. lanti-ŋaȵȶu d. panpiti-ŋka hip-group placename-loc e. naŋkuru-ŋka placename-loc (20) Djaru (Tsunoda 1981: 49, 163, 208) a. paka-ŋku b. ȵampa-ku splinter-inst what-inst c. ȶiŋkiɻi-ku d. ȵampa-wu-ji-ta laughter-inst what-dat-1sg.acc2pl.nom e. ȵaŋka-ɳa-ŋku f. waȵȶara-nta conj-1sg. why not-2pl.nom nom-2sg.acc (21) Nhanda (Blevins 2001a: 31, 33, 106) a. an̪i-ŋɡula b. aȵȡa-ɡula dance-amb shove-amb c. waȵȡa-ɻa-ɡula d. panda-bi-ŋɡula throw-3obl-amb tired-inch-amb e. waȵȡida-ŋɡula listen-amb

13.5.3 Chains of triggers and targets We saw earlier in (5b), that for a Gurindji word containing a chain of three NC clusters NC1 …NC2 …NC3 , where NC1 and NC2 are potential triggers and NC2 and NC3 are potential targets, NCD deletion will apply to NC2 and not NC3 . The same pattern occurs in Djaru (22) and Yindjibarndi (23). (22) Djaru (Tsunoda 1981: 49) a. /ȵampa-ŋku-ŋku/ b. /ȵampa-kuȵȶa-nta/ → ȵampa-ku-ŋku → ȵampa-wuȶa-nta what-erg-2sg.acc what-lacking-2pl.nom (23) Yindjibarndi (Wordick 1982: 34) /wuntu-ŋka-mpa-ʈu/ → wuntu-ka-mpa-ʈu → wuntu-wa-mpa-ʈu river-loc-top-contrast No language in the sample exhibits a pattern demonstrably different to this. Bearing in mind that the present study

142

confines itself to cases in which the last two NC clusters of NC…NC…NC must appear in suffixes or enclitics and not in roots, there are 15 languages in the sample where morphotactic restrictions prevent any NC…NC…NC chains arising, and so it is not possible to observe how NCD deletion would apply to them. In the remaining two languages, Kalkatungu and Wanyjirra, it appears that such chains could arise, however I have found no relevant examples in Kalkatungu. In Wanyjirra, the one example is (24), but here the final NC is in a clitic, and since NCD deletion applies only optionally in Wanyjirra clitics, no firm conclusions can be drawn from it. (24) Wanyjirra (Senge 2015: 242) /ŋantu-ŋku=ŋku/→ ŋantu-ku=ŋku who-erg=2min.obj

13.6 Interaction with other processes NCD deletion interacts with other phonological processes in 12 of the 20 languages in the sample. If we adopt a serially ordered approach to analysing the data, then in nine of those 12 languages, NCD deletion must be ordered later than the processes with which it interacts. In Kalkatungu (Kalkatungic), Panyjima and Kurrama (Ngayarta), Gooniyandi (Bunuban), and the four Ngumpin languages, NCD deletion counterfeeds lenition,3 which is to say, it converts NC clusters into intervocalic stops and those stops are not subsequently subject to lenition processes that do apply to comparable, underived stops. This is illustrated in (25a). Conversely, in Gurindji, lenition feeds the environments of NCD deletion, since in Gurindji stops are blockers of NCD deletion while glides, including glides derived by lenition, are not, as in (25b). Similarly, in Mara (Maran), lenition of /k/ to zero and subsequent vowel–vowel contraction feeds the environment of NCD deletion (26a), which in Mara does not operate over long distances (26b). (25) Gurindji (McConvell 1988b: 137, 139) a. spring-loc b. underlying /wiȵȶi-ŋka/ lenition k→w — NCD wiȵȶika surface wiȵȶika, *wiȵȶiwa

FaSiDa-comit /paɳku-kuȵȶa/ paɳkuwuȵȶa paɳkuwuȶa paɳkuwuȶa, *paɳkuwuȵȶa

3 Strictly speaking, in languages where lenition is a morphologically idiosyncratic process (as it often is), it would be technically possible to argue that all of the forms that undergo NCD deletion are morphologically idiosyncratic non-undergoers of lenition, and therefore that NCD deletion does not need to be ordered after it. Just how convincing this would be will depend on language-specific factors such as the relative numbers of other (non)undergoers of lenition in the language. For more on lenition in Australian languages, see Chapter 12, this volume.

nasal cluster dissimilation (26) Mara (Heath 1981a: 53, 235, 238) a. rdp-take-pst.cont underlying /rdp-{ka-ȵȶi}/ reduplication kaȵȶikaȵȶi lenition k,ȶ→Ø,j kaȵȶiaȵȶi VV contraction kaȵȶaȵȶi NCD kaȵȶaȶi surface kaȵȶaȶi, *kaȵȶaȵȶi, (27) Yanyuwa (Kirton and Charlie 1996: 15, 71) a. play-abl underlying /l ̪ura-ŋka/ NCD — lenition k→Ø — surface l ̪uraŋka (28) Yindjibarndi (Wordick 1982: 33, 34, 63) a. river-loc underlying /waɳʈa-ŋka/ NCD waɳʈaka lenition k→Ø waɳʈaa V coalescence waɳʈaː surface waɳʈaː, *waɳʈaka

b. rdp-eat-pres /rdp-{ȶi-ȵȶa}/ ȶiȵȶiȶiȵȶi ȶiȵȶajiȵȶa — — ȶiȵȶajiȵȶa,*ȶiȵȶajiȶa

b. foot-abl /maɳʈa-ŋka/ maɳʈaka maɳʈaa maɳʈaa, *maɳʈaka b. one-loc-top /kuȵȶiri-la-mpa/ — — — kuȵȶirilampa

In three languages, NCD deletion is ordered earlier than lenition. In Yanyuwa (Warumungic), NCD deletion feeds the lenition of stops to glides/zero, as shown in (27). In Yindjibarndi (Ngayarta), not only does NCD deletion feed lenition as in (28a), but its environment is counterfed by processes of lenition to zero and vowel coalescence (Wordick 1982: 34): in Yindjibarndi, NCD can only apply across one syllabic nucleus (28b), but it will not apply across a single nucleus which has been derived from two by the application of lenition and vowel reduction, (28c). In Yalarnnga, NCD deletion feeds lenition of /ȶ/ to /j/, as in (29).4 (29) Yalarnnga (Breen and Blake 2007: 19, 20) a. enter-purp b. hold-purp underlying /wirka-ȵȶata/ /n̪inti-ȵȶata/ NCD — n̪intiȶata lenition ȶ→j — n̪intijata surface wirkaȵȶata n̪intijata, *n̪intiȶata

4 NCD deletion of the ergative /-ŋku/ does not feed lenition of /k/, even though lenition of /k/ is found elsewhere in the language. As mentioned earlier, in Yalarnnga NCD deletion in the ergative is triggered ‘inconsistently’ and only by two stems.

c. my-obj-top-contrast /ŋaːɳʈu-ku-mpa-ʈu/ — ŋaːɳʈuumpaʈu ŋaːɳʈuːmpaʈu ŋaːɳʈuːmpaʈu, *ŋaːɳʈuːpaʈu

13.7 Summary and discussion This study reveals significant patterns in the typology and NCD deletion which have not been previously reported. The triggers of NCD deletion tend towards maximum generality: most often they range across all places of articulation, can be both homorganic or heterorganic, are not restricted to specific morphemes, and the triggering NC cluster can either be tautomorphemic or span two morphs. In contrast, the targets of NCD tend to be restricted. Cross-linguistically, NCD deletion tends to target NC clusters only at some places of articulation and not others. And often it is restricted further, to only certain morphological items. With respect to application at a distance and blockers thereof, there are few clear generalizations, other than that glides have not been observed to block NCD deletion in languages where it acts at a distance. Interestingly, while Gurindji has provided the field with its best-known case of NCD deletion, the study in this chapter reveals that it is also a case that is typologically atypical. In no other reported language is NCD deletion so unconstrained by idiosyncrasy, either morphological idiosyncrasy in the target (or more rarely, the trigger) or idiosyncrasy regarding targets’ possible places of articulation. It is worthwhile asking where NCD deletion in Australian languages plausibly fits into a wider understanding of kinds

143

erich r. round of phonological phenomena. At first glance, NCD deletion clearly looks like a type of dissimilation (McConvell 1988b; Evans 1995b: 733–4), since its triggers and targets are both NC clusters. But the triggers and targets are also far from identical. Languages overwhelmingly allow NCD triggers to be heterorganic, yet targets must be homorganic. Triggers are rarely subject to morphologically idiosyncratic exceptions, yet targets frequently are. And languages more readily allow triggers to be heteromorphemic than targets. In all cases, the asymmetries make the class of triggers larger than the class of targets. It is not immediately apparent why this should be case, or what it tells us about NCD deletion. NCD deletion acts at a distance that varies crosslinguistically: at a remove of just one syllabic nucleus, or two, or at long distances. Its distance of application can also vary by target. Blockers may or may not exist. Where they do, lower sonority segments are more likely to be blockers, though the typology remains poorly defined owing to a dearth of detailed evidence. This is unfortunate considering the typology of blocking and its theoretical explanation was a key emphasis in McConvell (1988b), and given that much of the description examined here post-dates McConvell’s paper. Stanton (2018) proposes an analysis of NCD deletion in Gurindji based on a phonetic speculation that segments between triggers and targets are nasalized. Considerations of space preclude a decent discussion, but certainly it is not immediately clear that the empirical premise is valid (no research has suggested such nasalization exists5 ), or how the analysis would extend to cases where a segment’s status as (non)trigger depends on place of articulation, or is morphologically idiosyncratic, or to cases where nasals are not blockers. A satisfying phonological analysis of NCD deletion, beyond the impressive but cross-linguistically unusual case of Gurindji, may still lie over the horizon just as it did when Evans (1995b: 733) singled it out as a ‘problematic phenomenon’ in Australian phonology a quarter of a century ago. In response to this, the contribution here is to provide the first typological study of NCD deletion that

5 McConvell (1988b) refers to the action of the velum, to suggest the equivalent of un-nasalized spreading. How this would extend to cases beyond Gurindji is similarly unclear. Regarding the conjectural nature of assuming NCD deletion is tied to leftwards nasal spreading, see Butcher (1999) for findings of a distinctive lack of leftwards nasalization spreading in Australian languages. Some additional discussion of Gurindji phonetics appears in Ennever (2014) and Ennever et al. (2017).

144

extends beyond the confines of the mostly Ngumpin facts examined by McConvell (1988b), in the hope that a better empirical understanding can provide a stepping-stone towards deeper insight soon.

13.8 Language sample Languages known to exhibit the kind of NCD examined in this chapter are as follows: Pama-Nyungan Kalkatungic: Kalkatungu (Blake 1979a), Yalarnnga (Breen and Blake 2007). Karnic: Dhirari (Austin 1981a), Arabana-Wangkangurru (Hercus 1994b). Kartu: Nhanda (Blevins 2001a). Ngumpin: Bilinarra (Meakins and Nordlinger 2014), Gurindji (McConvell 1988b), Wanyjirra (Senge 2015), Jaru (Tsunoda 1981). Paman (Norman): Kurtjar (Black and Gilbert 1996). Ngayarta: Kurrama (Hill 2011), Ngarla (Westerlund 2015), Ngarluma (Kohn 2011), Panyjima (Dench 1991a), Yindjibarndi (Wordick 1982). Thura-Yura: Adnyamathanha (Schebeck 1974). Warumungic: Yanyuwa (Kirton 1971b; Kirton and Charlie 1996; Yanyuwa Families and Bradley 2017). Other families Bunuban: Gooniyandi (McGregor 1990). Garrwan: Garrwa (Mushin 2012a; Breen 2003). Maran: Mara (Heath 1981a).

Acknowledgements A version of this chapter was presented at the Ninth European Australianist Workshop in Leuven, in April 2019. Thanks to participants for helpful discussion and feedback, and to Claire Bowern for comments on the chapter manuscript. Support for this research is gratefully acknowledged from the Australian Research Council (grant DE150101024), the British Academy (grant VF1_101602) as well as the University of Queensland, Surrey Morphology Group, and the Max Planck Institute for the Science of Human History.

chapter 14

Lexical stress Kathleen Jepson and Thomas Ennever

14.1 Introduction Patterns of lexical stress in Australian languages have been overviewed from a phonological point of view in several publications over the past 30 years (Baker 2014; Dixon 2002; Evans 1995b; Fletcher and Butcher 2014; Goedemans 2010). Within the metrical theory literature, Australian languages have been drawn upon extensively to exemplify rhythmical patterns, foot structure, and other stress-related phonological features. Looking beyond these structural generalizations, however, we are only just beginning to uncover the phonetic properties that instantiate the rhythmic structures of many of the continent’s languages. Acoustic manifestation of stress remains an enigmatic aspect of the phonetics of Australian languages. In this chapter, we proffer a broad overview of stress in Australian languages, highlighting some of the under-reviewed phonological and phonetic aspects of stress and stress-related phenomena, along with a summary, in the traditional vein, of the ranges of metrical structures proposed for the languages of Australia. At the outset, it is first necessary to clarify what we take to mean by the term ‘stress’. This term is used in two ways throughout this chapter reflecting the research within phonology (e.g. Goldsmith 1990; Hayes 1995; Kager 1995; Liberman 1979; Liberman and Prince 1977) and phonetics (e.g. M. E. Beckman 1986; Fry 1955; 1958; Gordon and Roettger 2017; Lehiste 1970). In phonological terms, stress generally describes (abstract) prominence relations between syllables in which one syllable is stronger relative to others within the same (prosodic) word, and this is part of the organizing structural principles of the word which often results in other structural characteristics. In the phonetic domain, acoustic stress describes acoustic prominence or salience of a syllable. Strictly speaking, in a formal phonological approach (e.g. metrical theory), the label of ‘stress’ associated with a syllable may not result in phonetic manifestation of stress on that syllable (see e.g. Goldsmith 1990: 171). In practice, however, formal analyses of stress are often based on some auditory or acoustic analysis. Indeed, auditory perceptions and acoustic measures do not always agree (see e.g. Round 2009; Tabain, Fletcher, and Butcher 2014 on ‘stress ghosting’, discussed further, below).

Tangible aspects of the effects of stress in terms of acoustic correlates of stress realization as well as syllable structure, segmental phonology, and distributional patterns are our focus. We limit ourselves to discussing stress assigned to syllables within phonological words (i.e. ‘lexical’ or ‘word’ stress). We exclude what is sometimes referred to as ‘postlexical accent’, ‘stress accent’, ‘intonational’, ‘sentence’, or ‘phrasal/prosodic’ stress—beyond highlighting where such features bear directly on the analysis of lexical stress (but see Fletcher, Chapter 15, this volume.). The chapter is organized as follows: Section 14.2 provides a summary of the state of the field and previous research; Section 14.3 recapitulates the ranges of metrical stress patterns proposed for Australian languages, Section 14.4 considers key segmental processes that are conditioned by lexical stress, Section 14.5 presents impressionistic phonetic analyses of stress including a discussion of the range of cues that have been claimed to correlate with metrical prominence—pitch, duration, loudness (amplitude/intensity), and so on—and how they are variably utilized in Australian languages, Section 14.6 foregrounds some of the emerging instrumental studies exploring these phonetic correlates. Finally, Section 14.7 summarizes our state of knowledge and looks to future directions within the field.

14.2 Previous surveys of Australian stress systems Research into the stress patterns of Australian languages has generally taken a back seat to other research endeavours in Australianist linguistics in the past forty years. The heyday of phonological research into Australian languages is generally acknowledged to have spanned the late 1960s–1980s. During this time, metrical structure was often a secondary focus to research interest in the establishment of the basic phonotactic structures of Australian languages (Dixon 1972; 1980; Hamilton 1995; 1996b; see also Round, Chapter 11, this volume). Much emphasis was placed on CVCV templates and the role of the syllable was generally downplayed (Dixon

Kathleen Jepson and Thomas Ennever, Lexical stress. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Kathleen Jepson and Thomas Ennever (2023). DOI: 10.1093/oso/9780198824978.003.0014

145

kathleen jepson and thomas ennever 1972). Little descriptive or formal work within Australia capitalized on broader developments in metrical stress theory in the 1970s and early 1980s (Hayes 1980; Liberman 1979; Liberman and Prince 1977, inter alia). Evans (1995b) drew attention to the wide scope for further prosodic research of Australian languages including taking a metrical theory approach to understanding lexical stress. However, our understanding of the lexical stress in these languages and prosodic structure more generally remains minimal in terms of a formal phonological approach, and this is also the case in terms of an experimental phonetic approach. In a survey of Australian stress patterns, Baker (2014: 153) affirmed that the ‘prosodic structure of Australian languages is currently an area in need of much greater empirical research, particularly in complex word structures’. To date there are still only a handful of detailed metrical or phonetic accounts of stress: Sayers (1976b) for WikMungkan; Bishop (2002a) for Bininj Kunwok; Baker (2008b) for Ngalakgan; Round (2009) for Kayardild; Tabain et al. (2014) for Pitjantjatjara. In spite of such a relative paucity of detailed studies in this domain, stress patterns of Australian languages feature regularly in global typologies of stress, including databases such as StressTyp (van der Hulst and Goedemans 2009) and its successor StressTyp2 (Goedemans, Heinz, and van der Hulst 2015), and are drawn upon extensively to exemplify certain rhythmical patterns, foot structure, and other stress-related phonological features (Goldsmith 1990; Gordon 2011; Hayes 1995). Key typological publications that concern Australian languages include Gordon (2002; 2005; 2011) and Goedemans (2010) as well as Evans’s (1995b) overview. Baker (2014), as mentioned earlier in this section, provides a valuable survey of stress patterns of Australian languages, focussing on their interaction with morphological structure, while Fletcher and Butcher (2014) present some of the more recent acoustic investigations into word-level prominence patterns of Australian languages. In addition to typological overviews of metrical stress patterns, a small but significant body of primary research has begun to accumulate in the acoustic domain as mentioned above in Section 14.1 (see overviews in Fletcher and Butcher 2014; see also Tabain, Chapter 9, this volume). A valuable development has been concerted efforts to distinguish the effects of lexical stress from those of post-lexical prosody (Fletcher 2014; Fletcher and Evans 2000; see also Fletcher, Chapter 15, this volume; Fletcher and Evans 2002), an endeavour that is being emphasized in the field more broadly (Gordon 2014; Gordon and Roettger 2017; Roettger and Gordon 2017). Another under-researched and yet fascinating area of research lies in the unpicking of correspondences between formal metre and the rhythmic organization

146

of Australian Aboriginal songs and verses (e.g. Turpin 2008; brief discussion in Bowern 2012a). Hale (1984b), for instance, notes an interesting mismatch between musical metre and syntactic and metrical constituents in a Warlpiri song with an Anmatyerr text. Hercus (1969: 22–3; 1986: 22–3) characterizes a similar disparity in text and tune in Wembawemba songs. Conversely, Turpin and Laughren (2013) demonstrate a preference to align phonological phrase edges with bar edges in Warlpiri yawulyu songs. With regards to smaller constituents, they find greater degrees of freedom—the primary matching being that of a syllable to a metrical (rhythmic) position. Beyond such intriguing snippets, the extent to which the medium of song co-opts and/or modifies the rhythmic structures of Australian languages remains to be explored.

14.3 Stress patterns and their variations From the outset of this section, it should be acknowledged that descriptions and typologies of stress patterns in Australian languages have generally been established with limited instrumental or theoretical analysis of stress in the language under investigation. Nevertheless, as mentioned above in Sections 14.1 and 14.2 and discussed further in this section, these analyses have entered into the wider domain of stress typologies and have been influential in determining the design space of metrical systems of the world’s languages. In this section, we focus on stress from a phonological point of view to highlight some of the key parameters along which Australian languages have been purported to vary. These include the locations of stressed syllables within a word (Section 14.3.1); the degree to which stress assignment is sensitive to syllable quantity (Section 14.3.2) and morphological structure (Section 14.3.3). We also consider cases of ‘variable systems’ where different patterns of stress assignment are observed within a single language (Section 14.3.4). Keep in mind that while there are some generalizations that can be made for some groups of languages, as we discuss throughout this chapter, these can be troublesome, and so we suggest an open-minded and preferably empirical approach is taken when conducting work on other Australian languages. A striking aspect of the typological literature on lexical stress in Australian languages is that it contains wildly competing claims concerning the diversity of structures across the country; ranging from very minimal (Goedemans 2010: 55) to extensive (Evans 1995b: 753). We err on the side of complexity and demonstrate this typological diversity in the following sections. However, there are some select features which can be generalized. Firstly, with two or three

lexical stress exceptions, stress in Australian languages is phonologically non-contrastive—that is to say, stress does not change the meaning of otherwise identical lexical words. This even extends to English-lexified creole languages with Australian Aboriginal language substrates—for example Fitzroy Valley Kriol (Fraser 1977: 168)—a creole spoken in Fitzroy Crossing, WA. The only clear candidates of Australian languages possessing contrastive stress are Ndjébanna (McKay 2000), Mbabaram (Dixon 1991b: 360; 2002: 558) and (possibly) Wik-Mungkan (Sayers 1976a,b).1 Instead, stress in most Australian languages is typically ‘fixed’, that is, it occurs at predictable locations within a word (Baker 2014: 153; Dixon 2002: 558; Evans 1995b; Fletcher and Butcher 2014: 113). Secondly, another commonality, and one which has only been highlighted more recently, is that stress in Australian languages can be analysed as exhibiting a sensitivity to morphological structure (see Baker 2014). Finally, wordfinal syllables are nearly always unstressed. In polysyllabic words with an even number of syllables, this is explained by the fact that Australian languages are analysed as having trochaic foot structure in which the left-most (i.e. first) syllable is the head. In polysyllabic words with an odd number of syllables, the final syllable can be analysed as either extrametrical, that is, invisible to the metrical rules of the language, or forming a dactyl with the preceding two syllables (Baker 2014; for discussion of extrametricality see e.g. Goldsmith 1990; Hayes 1995; Kager 1995; Liberman and Prince 1977). Another feature of stress patterns is that in nearly all Australian languages, the minimum word is bimoraic (Baker 2014; Evans 1995b). Indeed, in morphologically complex languages, the bimoraic minima is diagnostic of the basic prosodic word—see for example Murrinhpatha (Mansfield 2017) and Ngalakgan (Baker 2008b). In these languages, monosyllabic prosodic words typically undergo vowel lengthening in order to satisfy the bimoraic minimum (see discussion in Baker 2014; as well as Baker 2008b; Borowsky and Harvey 1997; Harvey 2001; Harvey and Borowsky 1999). The augmentation of the stem is another strategy. For Pitjantjatjara/Yankunytjatjara, Wilmoth and Mansfield (2021) show that verb roots necessarily must meet a bimoraic minimum and, should they not, an additional syllable termed a ‘stem augment’ is inserted before subsequent inflectional morphology. In Lardil, augmentation to meet a minimal word size of two morae is observed via the addition of a meaningless syllable -Ca to monomoraic stems (Round 2017b: 5–8).

1 Turtle (1977: 48) describes Alyawarra as having ‘contrastive stress’ but it appears to have a very limited functional load.

14.3.1 Location of stress Determining and describing the location of stresses within a word can be parameterized in a number of ways. Aspects to consider are where is the strongest stress in the word (towards the left or right), which syllables within a foot is the head (left or right), in which direction are other stresses assigned or said another way, and in which direction are feet parsed (towards the left or right). One major distinction in the typology of stress systems in Australian languages is where the main, or ‘primary’, stress occurs in the word. The distinction is usually between somewhere near the left word edge or the right word edge. That is, there is a distinction in whether the parsing of a phonological word into metrical feet proceeds from the left or right edge of the phonological word. Setting aside a number of confounding factors, these different alignments give rise to the typology of ‘initial’ and ‘penultimate’ stress patterns in Australian languages which have loosely been associated with Pama-Nyungan and nonPama-Nyungan languages respectively (Goedemans 2010). Both patterns, by their nature of being fixed and close to the word edges, can be considered to delineate the left or right boundary of words (Hyman 2014). A comparison of stress assignment in polysyllabic, monomorphemic words in leftaligned and right-aligned systems is presented in Table 14.1. Table 14.1 also illustrates the preference for the assignment of syllables into dactyls rather than trochees in longer words (five syllables or more) in Jingulu. Many left-aligned systems are also typically described in the literature as involving alternating secondary stresses being assigned rightwards across the word resulting in a rhythmic stress pattern. As mentioned above in Section 14.3, this is often with the proviso that the final syllable in the word is unstressed. This set of facts can be analysed as involving the placement of a quantity insensitive trochee (i.e. a left-headed disyllabic foot) aligned with the left edge of the word. A monomorphemic word is subsequently parsed into successive trochees. Where a word exhibits an odd number of syllables, as discussed in Section 14.3, the final syllable can be analysed as extrametrical (following a trochee), or it can be included in the final foot which would then be a dactyl. In either case, each foot receives a (secondary) stress on its head (i.e. main) syllable which is the antepenultimate syllable in the word. In right-aligned systems, prosodic words are parsed into feet beginning at the rightmost edge. The primary stress is assigned to the head foot which results in the surface observation that disyllabic, monomorphemic words are generally stressed on their initial syllable (just as in left-aligned systems), but trisyllabic or longer words take a primary stress on the penultimate syllable. Note that in right-aligned

147

kathleen jepson and thomas ennever Table 14.1 Stress alignments in Australian languages: left aligned (e.g. Wangkajunga) and right aligned (e.g. Jingulu). Left-aligned

Right-aligned

Syllable count

Wangkajunga (Pama-Nyungan) (Jones 2011: 45)

Jingulu (Mirndi) (Pensalfini 2003: 36–8)

σσ

/cámu/

‘grandfather’

/ŋáwu/

‘camp’

σσσ

/míɲili/

‘type of fruit’

/tankúra/

‘yellow’

σσσσ

/lúkaɻàra/

‘spinifex seed’

/càlurúka/

‘tea’

σσσσσ

/ŋàcalakúru/

‘mouth’

σσσσσσ

/màɳkulukúliti/

‘ear wax’

systems the final syllable is also unstressed because feet are still left-headed. The straightforward system of parsing prosodic words into metrical feet based on a single alignment parameter does not always apply when one turns to morphologically complex words. Considering left-aligned systems, non-primary stresses in multi-morphemic words may be distributed throughout the word in several ways. For some languages, such as Pintupi (Hansen and Hansen 1969), feet are parsed rightward in an entirely regular fashion and ignore any and all morphological structure (i.e. suffixal and clitic boundaries). Pintupi therefore constitutes classical ‘rhythmic’ secondary stress. As Baker (2014) highlights—if this description adequately describes the facts in Pintupi— such morphological insensitivity would be unusual in the Australian context (see Section 14.3.3). In Diyari for instance, feet are prohibited from containing nuclei from different morphemes. This results in a left-aligned stress pattern that shows different secondary stresses dependent on the morphological constituency of the prosodic word as shown by the Diyari data in (1) (Pensalfini 2000: 2) where feet are enclosed in parentheses. (1) Diyari a. (kárna)-(wàra) b. (márda)-la-nhi man-pl hill-char-loc Alternatively, in Dyirbal there is a constraint that a foot cannot comprise a syllable of a root and another from a suffix. Multiple suffixes however are free to be parsed into feet together and may receive secondary stresses as shown in (2) (Pensalfini 2000: 3). (2) Dyirbal a. (wánydji)-ngu motion_uphill-rel

148

b. (wánydji)-(ngú-gu) motion_uphill-rel-dat

It is further possible to observe left-aligned systems that nevertheless involve non-initial word stresses. This situation arises in cases where the prominent foot is aligned not with the left edge of a maximal prosodic word but of a lexical root. This occurs in some non-Pama-Nyungan languages which exhibit extensive prefixing morphology (e.g. Ngalakgan, see Baker 2008b). Further patterns of morphological sensitivity will be returned to in Section 14.3.3. Moving away from morphological sensitivity, a number of further variants of left-aligned systems can be noted. Firstly, there are some Pama-Nyungan languages that do not assign stress to initial syllables if they lack an onset, instead placing a primary stress on the second phonemic vowel (i.e. the peninitial syllable). This is the situation found in Alyawarra (Turtle 1977; Yallop 1977), Antekerrepenh (Breen 1977), and Arrernte, for which the assignment of main stress is usually stated to fall ‘on the first vowel that follows a consonant’ (Breen and Dobson 2005; Breen and Pensalfini 1999). This type of sensitivity to the ‘presence of the onset’ (PO) is argued to result from feet edges having a requirement that they align with an onset (Topintzi 2010). Under this view, initial vowels (i.e. syllables) in these languages would be ana lysed as extrametrical. While some of the world’s languages are further characterized by sensitivity to the segmental quality of the onset (e.g. Pirahã) no Australian languages have been observed with this type of onset sensitivity.2 A further pattern observed in left-aligned systems is the placement of a single secondary stress on the penultimate syllable without intervening stresses which could be assigned in a rhythmical system. Examples of this nonrhythmic assignment of secondary stress has been claimed for Djabugay (Patz 1991), Watjarri (Douglas 1981), and Nakkara (Eather 1990). In some cases it is possible to account 2 For one claimed case, see Gahl’s (1996) and Goedemans’s (1997) revisions of Davis’s (1985; 1988) onset-sensitivity account of stress assignment in MathiMathi.

lexical stress for some stress systems as resulting from bi-directional stress assignment (e.g. Round 2009: 378–81 for rhythmic stress assignment in Kayardild). This can be formalized in an OT approach that ranks both Right-Align and Left-Align as competing constraints. The typology of stress patterns in many reputedly ‘rightaligned’ systems among non-Pama-Nyungan languages is less certain. Chadwick (1978) proposes penultimate stress for all of the West Barkly languages, including Ngarngka, Binbinka, Gudanji, and Wambaya, but Nordlinger (1998a: 33) counters that at least the latter exhibits primary initial stress not unlike the left-aligned system described earlier in this section. Bowern (2012a: 110–14) demonstrates that stress assignment in the Nyulnyulun languages (nonPama-Nyungan) is initial—again a pattern more familiar to the suffixing Pama-Nyungan languages. Murrinhpatha, a non-Pama-Nyungan language, while showing evidence for a penultimate prosodic prominence, is argued to have no other evidence of metrical footing or additional stress assignment (Mansfield 2019b)—a pattern not widely reported for Australian languages. Conversely, penultimate stress is not exclusive to the non-Pama-Nyungan languages. Uradhi (Crowley 1983) is a Pama-Nyungan language which has right-oriented stress, assigned to the antepenultimate syllable, subject to the absence of long vowels elsewhere in the word.

14.3.2 Quantity sensitivity One significant parameter along which stress systems vary cross-linguistically is quantity or weight sensitivity. A language is quantity sensitive when the internal make-up of syllables is relevant to the construction of metrical feet. A syllable’s quantity, or weight is usually considered to be ‘heavy’ or ‘light’. Generally, a syllable is considered heavy if it contains a long vowel, and/or if it has a coda consonant. While quantity has played a fundamental role in the typology of stress systems in the world’s languages, it is often cited as playing little or no role in the metrical systems of Australian languages—that is to say, many Australian languages are quantity-insensitive (Gordon 2002; Kager 1992). This is not always the case, however, as suggested in the description of Uradhi in Section 14.3.1. For a number of other languages, including Nhanda (Blevins 2001a), Ngalakgan (Baker 2008b), Bardi (Bowern 2012a), and Kayardild (Evans 1995c), weight appears to be relevant to a degree in determining metrical structure in monomorphemic words. For at least Kayardild, this claim has been revised by Round’s (2009) explanation of Evans’s analysis as essentially stress ghosting. A simple illustration of quantity sensitivity is observed in Nhanda (Blevins 2001a)

for which it has been analysed that secondary stresses occur on non-initial long vowels. Whereas, in Bardi (Bowern 2012a), while syllable quantity does not predict secondary stress, a heavy syllable in the word-initial (i.e. primary stressed) position, permits secondary stress to occur on the second syllable of the word. That is to say, the extra weight of the initial syllable creates an environment where the following syllable can be stressed. In some languages, heavy syllables may also be a factor in determining the location of intonational pitch accents. For example, in Bininj Kunwok (Bishop 2002b) quantity sensitivity is somewhat observed through a preference for heavy feet of the type CVC or CVCC to attract the pitch peak of a post-lexical tonal event over a word-initial foot of the structure CV.

14.3.3 Morphological sensitivity As alluded to in Section 14.3.1, a characteristic trait of Australian languages is sensitivity to morphological structure (Baker 2014). Morphologically sensitive stress was first noted by Hale for Warlpiri (Hale 1977: 15). The now classic Warlpiri example exhibiting morphologically sensitive stress assignment is shown in (3): (3) Warlpiri Yáparla-ngùrlu Yápa-rlàngu-rlu Hale (1977: 15)

‘father’s mother-ela’ ‘person-also-erg’

The crucial observation of data such as (3) is that polysyllabic morphemes in Warlpiri (suffixes or clitics) are assigned stress independently. Morphologically insensitive stress assignment cannot capture the patterns observed since the polymorphemic words in (3) are otherwise segmentally identical (see also Nash 1980). Not all morphological relations are able to influence metrical stress assignment in this fashion across languages however, leading Baker (2014) to distinguish between productive ‘word-level’ and unproductive ‘root-level’ morphological constituents (Kiparsky 1982). ‘Word-level’ constituents are typically morphemes like case suffixes as in (3) and inflectional prefixes. ‘Root-level’ constituents often correspond to verbal inflectional morphology, (frozen) compounding boundaries and reduplications. The variation in types of morphological constituents which may or may not influence the assignment of stress in Australian languages is one area of promising future research. As of yet, there have been very few systematic investigations of stress and its relationship to morphological structure—it typically being described secondarily to an analysis of morphology. Round (2009) is one notable exception—the construction of metrical feet is argued to be

149

kathleen jepson and thomas ennever primarily driven by morphological structure in Kayardild. The primary foot is aligned with the left edge of roots and additional feet are aligned from the right (differences are noted between nominal and verbal roots). Beyond the root, some suffixes are argued to be underlying stressed and so are relevant to foot placement ‘to a degree’ (see Round 2009: 326 ff.). A general principle of morphological sensitivity to stress assignment is also identified in Murrinhpatha (Mansfield 2017: 362; 2019b) and Ngalakgan (Baker 2005; 2008b). Baker (2014: 155) contends that morphologically conditioned stress assignment is ‘likely’ for many Australian languages, but concedes that we have few thorough descriptions of metrical structures in Australian languages. More cursory mentions of morphologically sensitive phenomena can be found in the descriptions of various Australian languages for which morphological constituency is not argued to necessarily be the primary organizational principle. In the Paman language Kuuk Thaayorre, rightward elements in compounds are assigned stress and certain (but not all) enclitics, including /=(k)aːk/ ‘proprietive’ and /=(k)aːɻ/ ‘privative’, may attract primary stress away from the left edge of the phonological word (Gaby 2017: 69). Beyond brief descriptions such as these, precisely which kinds of morphological constituents may influence stress assignment—and those which do not—still remains to be determined for many Australian languages. Morphologically sensitive stress systems are not unique to Australia (see Kager 2007: 196), yet certain systems analysed for some Australian languages nevertheless pose problems for the standard view of cyclic morphology. Baker (2005), for instance, argues for Ngalakan that stress rules do not apply to increasingly large domains which include the stem, but seemingly to every suffix independently. Theories of phonology that propose constraints on what types of morphology can define their own cyclic domains—e.g. the Strict Cyclicity Hypothesis (Kiparsky 1982)—cannot naturally account for such phenomena. Borowsky (1993) and Baker (2005), however, propose that affixes themselves can behave like miniature stems and this has been supported in work on more comprehensively studied languages such as German and Dutch (Buckler 2009; Buckler and BermúdezOtero 2012). The ability of morphemes to define their own prosodic domain is not just restricted to roots and suffixal morphology but is also observed extending to enclitics as noted for Kuuk Thaayorre earlier in this section. Furthermore, in Ngiyambaa (Donaldson 1980), enclitics are assigned secondary stress in a manner identical to suffix morphology. One concomitant feature arising from the ability of morphemes to define their own stress domains is the possibility for underlying stresses on adjacent syllables.

150

Such underlying structures can surface in different ways, vis-à-vis the (non-)occurrence of ‘stress clash’—the surfacing of stressed, adjacent syllables. Some languages permit stress clash but only across specific morphological boundaries—for instance pronominal enclitics in Wangkajunga (Jones 2011: 45) or preverb–verb boundaries in Walmajarri (Hudson 1976: 7–8). Stress clash is not permitted across any other morpheme boundary in these languages. In Ngalakgan (Baker 2008b) and the Yolŋu languages, two stressed syllables may surface sequentially in instances of reduplication of a monosyllable (see Morphy 1983 on Djapu; Wilkinson 1991 on Djambarrpuyŋu; Wood 1978 on Gaalpu). For Djinang, it has been proposed that if multiple closed syllables occur sequentially each is stressed, suggesting a somewhat quantity sensitive system (Waters 1979). Similarly, Evans (1995c: 82–3) proposes that Kayardild permits stress clash, specifically in circumstances involving certain adjacent morae, (cf. Round’s 2009 re-analysis). However, most descriptions of Australian languages point out that stress clash is avoided, irrespective of morpheme boundaries. From another angle, the variable footing of prosodic words can also provide some insight into morphological constituency and inform analyses of different types of morphological boundaries. In all of the Ngumpin-Yapa languages for instance, there is a class of uninflecting verbal elements (typically termed coverbs or preverbs) which combine with an inflecting (lexical) verb to form a complex predicate (see Osgarby, Chapter 26, this volume; Meakins et al., Chapter 76, this volume). Generally speaking, lexical stress is assigned independently to these two words but certain subtypes of coverbs/preverbs will combine with inflecting verbs in such a manner that they alone receive a primary stress and the immediately following lexical verb will only receive a secondary stress dependent on its syllabic structure (Hudson 1978: 50; Tsunoda 1981: 178). As such this stress pattern is taken as diagnostic of a different kind of morpheme (Osgarby 2018b). Additionally, the preverb–verb morphological boundary in Warlpiri is also described as the only prosodic context in which two adjacent syllables may be stressed—requiring an analysis of an initial ‘degenerate foot’ for monosyllabic preverbs (Nash 1980: 112).

14.3.4 Variable systems In addition to the cross-linguistic variation described in Section 14.3.3, metrical systems in a number of Australian languages are noted for their language-internal variability in the assignment of lexical stress in otherwise predictable systems. Jingulu, for example, possesses a dominant

lexical stress stress pattern of penultimate main stress but nevertheless exhibits a large number of trisyllabic words which exhibit initial stress (Pensalfini 2003: 37). Similarly, in Ngalakgan, any nominal form which takes penultimate primary stress also has a variant with initial primary stress (Baker 2008b: 83). There are further languages, like Yintyingka (Verstraete and Rigsby 2015) or Limilngan (Harvey 2001), which allow main stress to occur on the first or second syllable, conditioned by word length and segmental make-up of the initial and second syllables. Metrical footing also appears to show some internal variation in some Australian languages. The best-documented instances of this are the Yolŋu languages (Morphy 1983 on Djapu; Waters 1979 on Djinang; Wilkinson 2012 on Djambarrpuyŋu; Wood 1978 on Gaalpu; inter alia). In these languages, analysts have used the term ‘stress groups’ to identify the variable selection of trochees and dactyls in the assignment of metrical feet to prosodic words. An extreme case is Gaagudju which Harvey (2002: 55–64) argues has no synchronic evidence for foot structure at all—a divergence with respect to the norm in Australia. Gaagudju represents a situation where diachronic changes have rendered an erstwhile standard alignment of foot and morphological boundaries into a diverse range of possible stress patterns including both penultimate or antepenultimate stress on nominal stems of variable length (Harvey 2002: 59). Another well-known ‘variable’ stress system in Australia is that of Yidiny (Dixon 1977b) which has attracted substantial theoretical work (Crowhurst and Hewitt 1995; Hayes 1980; 1995; 1999; Nash 1979b). Formal analyses arising from Dixon’s statements of stress assignment necessarily had to account for a complex interplay between stress, vowel length, and morphological sensitivity. However, Bowern, Alpher, and Round (2013)—on re-examination of the acoustic data—concluded that long vowels do not attract stress in Yidiny as purported by Dixon (1977b). Yidiny’s typological exceptionalism is thus somewhat reduced and primary stress can be described as invariably initial. The case of Yidiny therefore emphasizes the value in revisiting original stress data as systems of Australian languages have generally only been inferred impressionistically and by nonnative linguists.

14.4 Lexical stress and segmental alternations As discussed in Sections 14.3.1 and 14.3.3, the position of primary or main stress often coincides with the initial syllables of words. Therefore, features that affect the initial segment or syllable of words, or that we see evidence for on those syllables, also apply to primary stressed syllables.

Consequently, an understanding of these phenomena, albeit traditionally associated with word-edges, is important for understanding stress in Australian languages. The relationship between segmental phonology and metrical structure has received much attention in recent phonological literature (J. N. Beckman 1997; Bye and de Lacy 2008; de Lacy 2000; Gordon 2011; J. Smith 2002; Wolf 2012). It has been observed that the surface distribution of sound patterns across languages corroborates a basic notion that different kinds of segmental properties are characteristically favoured in stressed positions vs. unstressed positions (Giavazzi 2010; Gordon 2011). Australian languages are notable in this respect as they demonstrate a number of features that, on a surface level, appear to diverge from widespread patterns of ‘strength’. Here we survey just a selection of some of the distinctive Australian segmental alternations that are found to be conditioned by lexical stress: lenition (Section 14.4.1), initial dropping (Section 14.4.2), and vowel length contrast restrictions (Section 14.4.3). Post-tonic consonant strengthening is discussed in Section 14.6.3.

14.4.1 Lenition Cross-linguistically, consonantal lenition—the ‘weakening’ of a segment to a more sonorous counterpart—has been shown to be sensitive to prosodic structure and the placement of lexical stress (Bouavichith and Davidson 2013; Lavoie 2001). Broadly speaking, lenition is generally found to operate within prosodic domains where it serves to maintain acoustic continuity (Katz 2016; Kingston 2008). Conversely, fortition—the ‘strengthening’ of a segment—is generally understood to operate at prosodic boundaries where it serves to enhance acoustic cues associated with boundaries. Lenition processes in Australian languages are described as co-varying with metrical structure in languages such as Yir Yoront (Alpher 1988), Ndjébbana (McKay 2000), Limilngan (Harvey 2001), and Kuuk Thaayorre (Gaby 2017: 45–6, 74). Recent acoustic studies in this area include Mansfield (2015a), Ennever (2014), and Ennever, Meakins, and Round (2017) on Gurindji. A key finding of the Gurindji studies is the observed permissibility of consonantal lenition in word- (and domain-) initial positions. Recall that wordinitial syllables are often the location of primary stress in Pama-Nyungan languages; consequently, the onset of primary stressed syllables in these languages is lenited. The word initial position was also found to be most conducive to lenition in Murrinhpatha, although it is not the location of primary stress. However, other word medial contexts did not show lenition to such a degree, findings echoed in impressionistic studies elsewhere, for example, Kuuk Thaayorre (Gaby 2017: 45–6, 74). While the Murrinhpatha finding does

151

kathleen jepson and thomas ennever not directly speak to the relationship between stress and lenition, the general observations that in Australian languages, lenition occurs more so in initial syllables than elsewhere runs counter to much of the typological work on lenition (Escure 1977; Kirchner 2001; Ségéral and Scheer 2008) and in detailed acoustic studies of better-documented languages (e.g. lenition in American English, Bouavichith and Davidson 2013; Warner and Tucker 2011). Syllable onsets and word-initial position are generally regarded as locations where segments are phonetically strengthened, not weakened. Widespread morphophonological patterns of lenition appear to reflect the same anomalous patterning of weakening phenomena appearing in so-called ‘strong’ positions. Across Australia, a very widespread pattern of allomorphy is observed whereby suffix initial stops are lenited to corresponding semivowels when attached to stems ending in either a vowel or liquid (Ennever 2014; Round 2010b). These same languages purportedly assign a secondary stress to initial syllables of polysyllabic suffixes (as per Section 14.3.1). While a number of widespread allomorphic patterns involve monosyllabic suffixes (e.g. widespread dative -gu/-wu alternations), these patterns also extend to disyllabic suffixes (e.g. purposive -gurra/-wurra in Wanyjirra, Senge 2015). In fact, in Kuuk-Thaayorre, the likelihood of lenition has been described as increasing proportionally with the ‘size’ of the preceding morphological juncture (i.e. suffix < clitic < compound) (Gaby 2017). Therefore, it is an unusual feature of many Australian languages that morphophonological lenition is permitted in the onsets of (secondary-)stressed syllables. Note, however, that the existence of secondary stress is questioned in experimental acoustic phonetic analyses (see e.g. Jepson 2019; Tabain et al. 2014). A plausible explanation for some of these unusual lenition patterns possibly relates to the specific phonetic cueing of lexical stress (cf. Section 14.5.1). A number of instrumental and acoustic studies have found evidence that pitch peaks occur relatively late in the syllable in Australian languages from three language families and has been referred to as a general preponderance for ‘late-cuing’ (Blevins and Marmion 1994; Dixon 2002: 547; Round 2009: 317ff. for Kayardild; Tabain et al. 2014: 54 for Pitjantjatjara). If the acoustically most prominent position in the syllable is situated on the right edge of the nucleus, or indeed into the following consonant, then it is precisely the ‘post-tonic’ consonant that should resist reduction—both temporally and in magnitude of constriction and/or articulator movement. However, this pattern of late-cuing is not always found. For example, it has not been attested in Djambarrpuyŋu (Jepson 2019): intonational pitch accents are analysed as being associated with the initial syllable of the word (i.e. the primary stressed syllable) and resultant pitch peaks generally occur around

152

the mid-point of the syllable. Further, post-tonic consonant lengthening is not found in words greater than two syllables in length (Jepson, Fletcher, and Stoakes 2021). For those languages where a late peak is attested, the syllabic onset, more distant from the key pitch peak is left open to temporal and acoustic reduction. Word-initial consonant position is also observed to be weak by the collapsing of contrasts either in place (Baker 2014; Dixon 2002), or stop series (as reported for Yolŋu languages, see Wilkinson 2012, inter alia). Acoustic strengthening and weakening of consonants is discussed further in Section 14.6.3.

14.4.2 Initial dropping Another substantive process that has shaped the phonological landscape of Australian languages is so-called ‘initial dropping’ (Alpher 1976; Blevins 2001b; Blevins and Marmion 1994; Sommer 1970). Dixon (2002) provides a comprehensive list of the languages affected by this diachronic process but includes two areal groupings—namely the Arandic languages of Central Australia and a selection of Northern Paman languages (Hale 1964; 1976d) in Cape York along with certain Paman languages to the south, including KukuThaypan (Rigsby 1976). In these languages there has been diachronic loss of initial consonants (and in some cases the first vowel). Hale (1964: 256) contends that this change was, in part, precipitated by a shifting of primary stress from the first to the second syllable of the word. Dixon (1980: 197; 2002: 594) similarly asserts that the ‘prevailing pattern of change’ in these languages involved the loss of initial consonant before any shift of primary stress from initial to second syllable. Nevertheless, Rigsby, quoted in Godman (1993) has pointed out that there are Cape York languages (such as Barrow Point and Flinders Island languages) that have initial dropping without stress shift.

14.4.3 Contrastive vowel length restrictions For some Australian languages, positional constraints exist for contrastive vowel length. That is, vowels are found to be contrastive by length in the first syllables of words, which is typically where primary stress is located in PamaNyungan languages. Positional prominence effects, such as more segmental contrasts in stressed syllables compared to unstressed syllables, are a common property of stress systems (Hyman 2014). Long vowels only occur in the initial (i.e. stressed) syllables of words in, for example, Warlpiri (Nash 1980: 65) and Wangkajunga (Jones 2011: 33); however, they can infrequently occur in other syllables in other languages, as reported for Bardi (Bowern

lexical stress 2012a) and Yukulta (Keen 1983). Of course, some languages show contrastive length across word positions, for example, Nhanda (Blevins 2001a) and Kayardild (Evans 1995c; Round 2009). As discussed in Section 14.3.4, this may disrupt the locations for secondary stresses. A further vowel related restriction is on what vowels can occur in stressed syllables with respect to their quality. In Burarra, for instance, Glasgow (1981b: 73) reports that there is a restriction in what vowels can occur in stressed, secondary stressed and unstressed syllables such that the five vowel system /i, e, a, o, u/ is reduced to a three-way contrast—/i, a, u/—in secondary stressed and unstressed syllables.

14.5 Impressionistic studies of lexical stress Beyond segmental alternations, lexical stress is crosslinguistically associated with perceived and measurable phonetic prominence. Problematically, cues to stress, both impressionistic and acoustic, are also well known to be elusive, or at least show considerable variability, especially those associated with secondary stress (Goldsmith 1990; Gordon and Roettger 2017; Hayes 1995; Kager 1995; Lehiste 1970). The difficulty in pinning down possible prominence encoding features relates to the fact that different languages appear to utilize different combinations of phonetic features to differing degrees. While there is the suggestion within some theoretical literature that there may be a phonetic manifestation of metrical stress, it is not a necessity, as metrical strength is an abstract property. However, the evaluation of the evidence for stress systems is of clear importance in the Australian context (see de Lacy 2014 for a discussion of the strength of formal evidence for stress systems) not least because of the use of Australian languages to exemplify patterns in theoretically driven research. Returning to the data in search of an empirical basis for patterns of stress can prove extremely insightful as shown by Bowern et al. (2013), discussed in Section 14.3.4. Consequently, in this section auditory cues to stress (i.e. what the researcher reports to have heard) in studies of Australian languages are considered. The acoustic studies undertaken to date are discussed in Sections 14.6.1–14.6.3.

14.5.1 Reported cues to stress In the grammatical and phonological descriptions of Australian languages for which an analysis of stress is presented, the researcher(s) may report on acoustic properties which they perceived and attributed to the co-location of

metrical prominence. These features vary widely but often include: a pitch peak or special pitch pattern, greater intensity, longer duration, and more peripheral vowels in stressed syllables and/or more centralized vowels in unstressed syllables. In general, however, there is an acute lack of information about the specific correlates associated with stress reported in grammatical and other descriptive work. Some researchers perform small-scale phonetic analyses (e.g. Bowern 2012a), though quantitative experimental phonetic analyses do not appear in grammars written to date. In this section, we overview the range of phonetic cues explicitly mentioned in relation to stress in studies of Australian languages. The analyses provided in grammatical works can be useful in assessing what features may be salient for listeners in determining the location of stress. Throughout the Australian language research, pitch is the strongest perceptual cue for researchers and is most often reported as a cue. Durational cues (most often vowel duration) are also reported as cues to prominence. Intensity (i.e. loudness), however, is less frequently mentioned, and when it is, it is reported in conjunction with another cue, usually pitch, but sometimes duration. In general, when correlates are mentioned, they are reported to occur in tandem. For example, Hansen and Hansen (1969: 162) write for Pintupi: ‘Primary stress is manifested by increased loudness, with the additional features of high pitch and greater length frequently evident’. This range of cues is echoed for Warlpiri by Nash (1980: 99): The impressionistic phonetic correlates of stress are (i) a relatively greater intensity on the stressed syllable, (ii) a relatively greater duration of the stressed syllable, and perhaps (iii) a somewhat raised pitch on the stressed syllable. Vowel quality is not much affected by stress or the lack of it.

More recently, Bowern (2012a: 111) summarized for Bardi: Stress is manifested as a combination of intensity (loudness), raised pitch and slight lengthening over the vowel. It is also characterized by shortening and some centralization of the vowel in the syllable following the stressed vowel, particularly when that syllable is open.

Not all descriptions involve a neat summary of this sort, rather cues may be dispersed throughout the phonological analysis or mentioned in passing elsewhere within a description (the vowel section of a phonological analysis is usually a fruitful place to look). Nevertheless, impressionistic cues to stress—duration, loudness, pitch, and vowel quality—are mentioned in the majority of descriptions and can be helpful in determining what cues to investigate experimentally.

153

kathleen jepson and thomas ennever Across the board, secondary stress poses the greatest problem in impressionistic analyses due to its lack of consistent auditory cues. Often, cues are not perceived in words uttered at normal speech rate, or outside of isolation (Hudson 1976 on Walmajarri; Kofod 1978 on Miriwung; McKay 2000 on Rembarrnga). Therefore, the following discussion is predominantly about primary stress with some discussion of secondary stress. Note that the languages provided as examples for each correlate may also make use of other correlates for which they are not mentioned. Pitch is the most frequently cited acoustic property claimed to be a correlate of stress in Australian languages. Pitch is reported to be highest on the stressed syllable in such languages as Djaru (Tsunoda 1981), Emmi (Ford 1998), Gaagudju (Harvey 1992a), Kayardild (Evans 1995c), and Yintjingka (Verstraete and Rigsby 2015). Pensalfini’s grammar of Jingulu (2003: 36) is one of only a handful of works to state that pitch was not a cue to stress. Instead, for Jingulu, pitch contours are described as simply falling invariably from the initial syllable of a word, irrespective of the location of primary stress (which includes penultimate primary stresses). Sometimes pitch differences are explicitly stated to be associated with a post-lexical, phrasal level of prosody, but are still associated with, or align to, a stressed syllable—primary, usually—and thus determined to be a useful auditory cue to the location of stress and support for the metrical analysis. This is the case for Bardi (Bowern 2012a: 110; Bowern, McDonough, and Kelliher 2012: 347) for which it is suggested that primary stressed syllables are encoded through the association of an intonational pitch accent. The high pitch accent is associated with the first syllable of content words, which is the proposed location of primary stress. Secondary stresses are found to attract smaller pitch peaks sometimes (i.e. are less prominent). Baker (2014: 90) describes a different situation for Ngalakgan. In compounds, each metrical head is associated with a salient pitch movement termed a ‘pitch accent’. However, pitch accents do not always occur on every metrical head, they are usually assigned to only one head within a single morphological word. Harvey (2001: 32) on Limilngan also states that ‘tone [i.e. pitch] is fundamentally phrase-level and not word-level in its scope’. He does not assume that this provides an insight into stress though he acknowledges there may be an interaction. See Fletcher, Chapter 15, this volume, on intonation. Duration is another recurring acoustic property which has been used to identify the presence of metrical stress in Australian languages. Syllables or vowels that have main or primary stress are reported to be longer. This is described for a range of Pama-Nyungan and non-Pama-Nyungan languages including: Djinang (Waters 1979), Nakkara (Eather 1990), Nhanda (Blevins 2001a), and Wembawemba (Hercus

154

1969). For Bardi, Bowern (2012a: 88) reports that short vowels in initial (stressed) syllables have average duration of 125ms, while short vowels in second (unstressed) syllables are 95ms. In Walmatjarri (Hudson and Richards 1969: 183), in contrast, duration is cited as a cue to secondary stress but not primary stress, which is impressionistically correlated only with higher pitch and greater loudness. There are also instances where phonological vowel length and stress are not disentangled. Such a situation is found in Gaagudju (Harvey 1992a: 84) where ‘most long vowels are stressed, and conversely nearly all stressed vowels are long’. Further support is offered to the location of main of stress through the restriction of the vowel length contrast usually to the initial syllable of the word (discussed in Section 14.4.3). Amplitude, intensity, or loudness is said to be higher or greater for stressed syllables in Alawa (Sharpe 1972), Garrwa (Furby 1974; Mushin 2012a), Kuku Yalanji (Hershberger and Pike 1970; Patz 1982), Limilngan (Harvey 2001), Walmatjarri (Hudson and Richards 1969), and Wangurri (McLellan 1989). Of Alawa, Sharpe (1972: 14) writes that ‘[s]tress is relatively even on all syllables of words, except for an initial build-up of intensity on the first syllable’ which then gradually declines over the following syllables until the end of the word. Pintupi (Hansen and Hansen 1969) uses greater loudness to distinguish secondary stressed syllables from contiguous syllables. Vowel quality is rarely cited as a cue to stress, the implication being that vowel quality is independent of stress. However, when it is reported as a correlate of stress, it can operate in both the expected direction (i.e. metrically strong/stressed vowels are more peripheral in the vowel space), as well as in the opposite direction (i.e. metrically weak/unstressed vowels are more peripheral in the vowel space). In the first case, unstressed vowels are more centralized in the vowel space, or even reduced to schwa in contrast to their stressed vowel counterparts which are more peripheral. This is described for Bilinarra (Meakins and Nordlinger 2014), Kayardild (Evans 1995c), Wambaya (Nordlinger 1998a), and Wergaia (Hercus 1969). There are also reports of phoneme specific differences in stressed and unstressed positions. For Miriwung, Kofod (1978) reports that [ə] occurs in unstressed syllables in free variation with [ʌ], and similarly, the /a/ vowel may be more schwa-like in unstressed syllables. Ndjébbana (McKay 2000), on the other hand, has a neutralization of unstressed vowels to [a] when stress shift has occurred. As an example of the second case, where stressed vowels are less peripheral, Waters (1979) reports for Djinang, that /i/ and /u/ are realized as [e] and [o] when in stressed syllables. Similarly for Gaagudju, Harvey (2002) explains that long cardinal vowels in stressed syllables tend to be replaced by corresponding mid vowels, a phenomenon referred to

lexical stress as ‘vowel grade’ in which underlyingly cardinal vowels (i.e. vowels that are more peripheral in the acoustic vowel space) are ‘altered’ to be corresponding mid vowels, when they are long and stressed.

14.6 Acoustic studies of lexical stress In recent years there has been an increase in the number of acoustic and experimental phonetic analyses of stress seeking to verify and/or quantify the impressionistically reported correlates of stress detailed in the previous section: duration, loudness, pitch, and vowel quality. The acoustic measures investigated are the physical correlates of the auditory cues: duration, intensity (considered overall, or in a particular frequency band), fundamental frequency (f0), and formant frequency (usually the first (F1) and second (F2) formants). There exist very few experimental acoustic phonetic studies of stress or correlates of metrical prominence in Australian languages which thoroughly explore if and how those above-mentioned acoustic measures distinguish levels of stress. However, this is slowly being remedied. In those studies that do exist, the ways stress and prosodic structure are encoded are found to be different from other languages of the world. As Pentland notes (2004: 4; see also Pentland and Laughren 2005), major differences in the prosodic systems of Australian languages compared with other languages of the world are not surprising given the other differences found in their phonetics (see e.g. Butcher 2006; Tabain, Breen, Butcher, Jukes, and Beare 2016; Tabain and Butcher 1999; 2015). A chief difference is that consonants appear to play an important role in the encoding of prosodic prominence in some Australian languages (see Fletcher and Butcher 2014 for further discussion). See also recent work by Babinski and Bowern (e.g. 2019) for a cross-linguistic investigation. We consider acoustic correlates of stress realized on vowels in Sections 14.6.1–14.6.2. Post-tonic consonant lengthening and other effects of prominence on consonants are discussed in Section 14.6.3. The discussion in the following sections revolves predominantly around phonetic evidence for primary stress. Tabain et al.’s (2014) study of Pitjantjatjara arguably represents the most comprehensive acoustic phonetic analysis of stress in an Australian language and investigates effects of primary and secondary stress on f0, duration, Root Mean Square (RMS) amplitude, spectral tilt, and the first, second, and third formants of vowels. The authors find no evidence for phonetically prominent word-internal syllables, that is, secondary stressed syllables. As mentioned in Section 14.5, this does not mean that a syllable is not metrically stressed

(i.e. abstractly the head of a metrical foot), but it does pose the question of what the linguist listener has cued into to perceive secondary stress in analyses of Pitjantjatjara, and possibly Australian languages more generally.

14.6.1 Fundamental frequency F0, the physical correlate of pitch, is the foremost acoustic measure to be considered in experimental studies of Australian languages. Note also that pitch was called upon most frequently in the impressionistic auditory descriptions of stress in Australian languages surveyed in Section 14.5.1. Fletcher and Butcher (2014) suggest that intonational tonal events may have been analysed by linguists as stress. This is not an issue that is restricted to the Australian context. Gordon (2014) highlights that the prominences attributed to different levels of prosodic constituency have not been disambiguated in many studies of lexical stress, including in experimental acoustic studies. The conflation of word-level prominence (i.e. lexical stress) and phrase-level prominence (i.e. intonational events such as pitch accents) can occur because only words uttered in isolation are investigated, or because words occur in a focal position within the utterance (see for discussion M. E. Beckman 1986; van der Hulst 2014). However, the existence of an intonationally conditioned pitch event may still be an important cue to metrical structure, despite it not being a cue to stress specifically; the association of a pitch accent may imply that the docking syllable is stressed, by nature of it being eligible to be promoted to pitch accent status (Gordon 2014). Tabain et al. (2014) investigated f0 in Pitjantjatjara (as well as duration, RMS amplitude, spectral tilt, and vowel formant structure). As is most commonly reported for languages in the Pama-Nyungan family, primary stress occurs at the left edge of words in Pitjantjatjara while secondary stresses have previously been reported to occur following a pattern of alternating trochaic rhythm, with word-final syllables in words of odd-numbered syllables being extrametrical. Working with an intonationally annotated database of narratives (Pitjantjatjara version of the Three Billy Goats story and the South Wind and the Sun passage), the authors examined the f0 values of main stressed syllables compared with unstressed and secondary stressed syllables as well as where pitch peaks occurred. The authors found that a pitch peak occurred on the first syllable of words. This confirmed their hypothesis that f0 had a delimitative function for word boundaries and also marked main word stress (see also Tabain and Fletcher 2012). For Bininj Kunwok, Bishop (2002a, b) reported that pitch is the only consistent correlate of stress. Bishop interprets this as showing that intonational accents are attracted to

155

kathleen jepson and thomas ennever metrically strong syllables, but there is no phonetic stress. Thus, it is only through considering this higher level of prosody and prominence, in which abstract metrical prominences and phonetic intonational prominences align, that there is phonetic evidence of metrical structure. Further, Bishop found for Bininj Kunwok that a pitch accent was associated with the leftmost and rightmost feet of nominals and verbs, and thus has a delimitive function, marking the boundaries of these words. Jepson (2019) also found that consistent correlates of primary stress were difficult to establish for Djambarrpuyŋu, with f0 being the only consistent cue. However, this was possibly due to the association of a post-lexical tonal event such as a pitch accent, as all target words in the analysis were in focus.

14.6.2 Vowel formants and duration In acoustic analyses, other phonetic effects of stress that are observable on vowels are limited. Vowel duration has been investigated in Arrernte (Tabain 2016), Djambarrpuyŋu (Jepson 2019), Kayardild (Round 2012), Pitjantjara (Tabain et al. 2014), and Warlpiri (Pentland 2004). Vowels did not have longer duration in primary or secondary stressed syllables in Warlpiri, nor in Djambarrpuyŋu. However, vowels in primary stressed syllables in Pitjantjatjara were longer than vowels in weak word-medial syllables and secondary stressed (i.e. strong word-medial) syllables (Tabain et al. 2014). Tabain’s (2016) study of Arrernte vowels showed that vowel categories behave differently in terms of durational enhancement, with the open vowel /a/ showing lengthening effects associated with location within a primary stressed syllable, whereas the two close vowels and schwa did not. There are only a small number of languages for which there are systematic studies of potential relationships between more extreme vowel formants—or overall expansion of the vowel space—and metrical prominence. Graetzer’s (2012) study of Gupapuyŋu, Arrernte, Burarra, and Warlpiri showed that vowel formants were not consistently affected by being in a metrically prominent position. Prominent vowels in Burarra are, for only some speakers, found to be more dispersed in the vowel space than their nonprominent counterparts. Gupapuyŋu vowels in prosodically weak syllables showed more formant variability for only some vowels (Graetzer 2012: 185–90). For the closely related language Djambarrpuyŋu, Jepson (2019) did not find differences in formant frequencies for primary or secondary stressed vowels (i.e. any metrically prominent positions). Formant frequency differences were also not found to covary with prominence in Pitjantjatjara (Tabain et al. 2014), nor Bininj Kunwok (Bishop 2002a; Fletcher, Stoakes, Loakes,

156

and Butcher 2007; cf. Fletcher and Butcher 2014, see also Fletcher and Butcher 2003 on the effect of accentual prominence on vowel formants in Bininj Kunwok, Dalabon, and Kayardild).

14.6.3 Consonant related phenomena Due to the paucity of effects of stress (i.e. metrical strength) on vowels, research has partially refocussed on the effects of prosodic prominence on consonants in Australian languages. See for example, Butcher and Harrington (2003a, b), Fletcher, Stoakes, Loakes, and Singer (2015), Pentland (2004), Tabain and Butcher (2015), Tabain et al. (2016), and Jepson et al. (2021). It has been claimed that Australian languages operate somewhat differently in terms of prosodically conditioned consonant lengthening and strengthening compared with other languages of the world (see Butcher 2006). As discussed in Sections 14.4.1–14.4.2, wordinitial and syllabic onset positions are prone to phonetic weakening in some Australian languages. The position of prosodic strength is instead suggested to be the post-tonic position—the consonant after the vowel that bears a phrasal pitch accent—irrespective of whether the consonant is affiliated with the tonic syllable or not (Butcher 2006). That is not to say that onsets of stressed syllables never show an effect of stress; they have been investigated in a small handful of languages and have also been found to show signs of strengthening (e.g. Pentland 2004; Tabain 2016; Tabain et al. 2016; Tabain and Butcher 2015). Both positions are discussed in this section. See also discussion of posttonic consonants in Section 14.3 with respect to bimoracity constraints. Consonants in stressed syllable onset position have been found to show an effect of metrical strength through enhancing paradigmatic contrasts in some Australian languages. Pitjantjatjara (Tabain and Butcher 2015) provides an example of how phonotactics and phonetics differ in the evidence they provide for stress. Recall that the position of primary stress in Pitjantjatjara is the initial syllable of the word. The contrast between retroflex and alveolar apical consonants, however, is neutralized in the word-initial position which coincides with the onset to the primary stressed syllable (such a neutralization is observed in many Australian languages with an apical contrast, see Butcher 1995). Tabain and Butcher (2015) showed that, despite this loss of a place contrast (suggesting that that position is not strong; see e.g. Hyman 2014), the neutralized consonant in wordinitial/stressed syllable onset position has higher spectral tilt and centre of gravity (reflecting possibly a more open mouth posture and louder articulation) than either of the non-neutralized counterparts elsewhere within the word.

lexical stress These spectral energy changes for the apical consonants were accompanied by the lowering of spectral tilt and centre of gravity (i.e. ‘darkening’) for the velar stop /k/. Thus, the contrast between the apical and dorsal stops was found to be enhanced in stressed position. These types of enhancements are also reflected in burst duration in Pitjantjara (Tabain et al. 2016), as has been reported for Arrernte and Warlpiri. In particular, stop burst duration of palatal /c/ and velar /k/ in stressed position is longer than when not stressed, a pattern which is not observed for the coronal consonants /t̪, t, ʈ/. Tabain et al. (2016) suggest that burst duration may be asymmetrically augmented to enhance the distinction between the anterior and posterior consonants in primary stressed positions. In terms of segment duration and lexical stress, consonant lengthening is investigated with respect to both consonants in the onset position of stressed syllables, as well as the consonants in the post-tonic position. Consonants in the former position have been found to be lengthened in Tabain’s (2016) study of prosody in Arrernte. Post-tonic consonant lengthening has been investigated acoustically in a small handful of acoustic studies, noted at the beginning of this section, the results of which are discussed in the following paragraphs. In addition, the phenomena has been mentioned in other linguistic work on, for example: ArabanaWangkangurru (Hercus 1994), Djapu (Morphy 1983), Djinang (Waters 1979), Ndjébbana (McKay 2000), Nhanda (Blevins 2001a), Paakantyi (Hercus 1994), Wik-Mungkan (McConnel 1945; Sayers 1976a), and Yintyingka (Verstraete and Rigsby 2015). For Kayardild, Evans (1995c), mentions a slightly different situation in which consonants are lengthened when they occur between a main stressed and secondary stressed vowel. Evans suggests that this could be a strategy for ensuring contiguous stresses are in non-contiguous morae, as mentioned in Section 14.3.3 with respect to stress clash. However, Round (2009: 323) found that post-tonic lengthening occurred only when words were in utterance initial position. Pentland’s (2004) analysis of Warlpiri data found that both word- and utterance-initial position influenced consonant duration. However, stops and nasals were affected differently. Specifically, word-initial stops were long when the word was in utterance-medial position compared with stops that occurred word-medially following stressed vowels in utterance-initial position and also stops word-medially following unstressed vowels. Nasals on the other hand were longest when in word-initial utterance-initial position compared with nasals in all other word and utterance positions that Pentland measured. Nasals did not lengthen in wordinitial utterance-medial position. Pentland did not compare word-initial utterance-initial stops. Considering the post-tonic position in Warlpiri, stops following primary stressed vowels are lengthened in words

that are in utterance-initial position (Pentland 2004). Stops did not show an effect of post-tonic position when words were utterance-medial. Further, nasals did not show an effect due to being in the post-tonic position in Pentland’s study, though are found to be lengthened in post-tonic position in heterorganic clusters in Warlpiri in an electropalatography study by Fletcher, Loakes, and Butcher (2008). Warlpiri data also showed articulatory strengthening of /j/ in post-tonic position such that there was a greater degree of tongue raising when the target words containing /j/ were in a focussed context and marked by a post-lexical pitch event (Butcher and Harrington 2003a). Somewhat differently to Warlpiri, where post-tonic lengthening of consonants is restricted to stops, sonorants in Mawng have a longer mean duration after an accentually prominent vowel—82 ms compared with 68 ms after a non-tonic vowel (Fletcher et al. 2015). By contrast there is minimal lengthening of accented (i.e. tonic) vs. unaccented vowels. In particular, the authors emphasize that through lengthening the sonorant, the sonority of the whole vowel+sonorant sequence is boosted, and further, the longer sonorant segment enhances the rising pitch movement associated with accentual prominence. They conclude that the post-tonic sonorant in Mawng forms part of the entire accentual gesture (Fletcher et al. 2015). Accentual gesture here refers to a range of articulatory gestures related to accentuation, including bigger and faster opening and closing features, as well as more extreme tongue gestures (see e.g. Fletcher 2010 on speech timing and rhythm). A similar pattern is observed in studies of Bininj Kunwok, in which post-tonic nasals in the first position of a heterorganic consonant cluster are longer than nasals in the second position, intervocalic, and word-initial positions (Fletcher and Butcher 2014; Fletcher et al. 2010). In a recent study of Djambarrpuyŋu, Jepson et al. (2021) found that post-tonic consonants in disyllabic words were longer than consonants in post-tonic position in longer words and other wordmedial positions. Fortis stops were found to show the effect more strongly than nasals. This ‘lengthening and strengthening’ of post-tonic consonants appears to serve dual purposes of signalling prosodic prominence on the one hand and maintaining and possibly enhancing phonemic consonant contrasts on the other (Butcher 2006; Fletcher and Butcher 2014). This tendency for preserving or enhancing the place information of consonants at the expense of the quality of neighbouring vocalic segments, for example, has been termed the ‘Place of Articulation Imperative’ by Butcher (2006). As mentioned earlier in this section, most Australian languages neutralize coronal contrasts in word-initial position (Baker 2014; Butcher 2006; Dixon 2002)—the full range of contrasts is often restricted to intervocalic or inter-continuant position. The strengthening and lengthening of these intervocalic

157

kathleen jepson and thomas ennever or inter-continuant consonants is also associated with the cueing of phrasal prominence although further research is greatly needed to see whether these patterns hold across other Australian languages for all classes of consonants.

14.7 Chapter summary Aspects of lexical stress in Australian languages have contributed a great deal to discussion of formal metrical theory in the past thirty years. Our understanding of the range of principles of stress assignment possible in the world’s languages has benefitted from considering phenomena found in languages of the Australian continent. However, if we may venture a prospective outlook for future studies of lexical stress in Australian languages, we see a great value in pursuing two avenues of future research. First, there is arguably much to be gained by re-examination of the primary data and the acoustic properties underlying even ‘well-cited’ stress patterns. Studies in the vein of Bowern et al. (2013) and Babinski and Bowern (2019) bring a much-needed rigour to earlier impressionistic studies and will refine our understanding not only of the metrical systems themselves but also of the acoustic properties that underlie them. This in turn will allow for a more complete assessment of the compelling claims surrounding the primacy of morphological/word boundaries in the assignment of metrical prominences throughout the phonological word (Baker 2014). Secondly, there is an acute need for additional acoustic studies that attempt to elucidate the relationship between phonetic

158

stress and intonational prominence (accent) (see Gordon 2014; Gordon and Roettger 2017; Roettger and Gordon 2017). As illustrated by Bishop (2002a), findings relating to exactly how metrical prominences may or may not be realized at different intonational levels are only possible when examined within the context of the complete prosodic system. While there is strong phonological and phonetic evidence (both impressionistic and acoustic) for the location of primary stress in a number of languages of Australia, it must be explicitly stated that despite numerous phonological descriptions of stress patterns in Australian languages positing secondary stress, the phonological and phonetic reality of secondary stress has been called into question in recent acoustic studies. As mentioned in Section 14.6, Tabain et al. (2014), for instance, find no acoustic evidence in Pitjantjatjara; nor does Jepson (2019) for Djamparrpuyŋu (Yolŋu). Previous analyses of secondary stress have similarly been revoked for Kalkatungu (Blake 1979a), Murrinhpatha (Mansfield 2017: 5), and Yidiny (Bowern et al. 2013). The explanation for why multiple analysts have heard secondary, rhythmic stresses in these languages has been explained by the effects of ‘stress ghosting’—essentially hearing stress where one expects to, based on one’s L1 phonology—most frequently English, in the Australian context (Round 2009; Tabain et al. 2014). Languages which have only ever been argued to have a single primary stress include Kugu Nganhcara (Paman, PN) (I. Smith and Johnson 2000: 384) and Burarra (Maningrida) (Glasgow 1981b). The extent to which this pattern is actually reflected more broadly across Australia awaits further acoustic investigation as well as native speaker judgements.

chapter 15

Intonation Janet Fletcher

15.1 Introduction Intonation is a general term that is used to describe meaningful variation in voice pitch that can perform many post-lexical functions including the encoding of pragmatic meaning and information structure, speech acts, grouping words into phrases and discourse chunks as well as signalling a range of paralinguistic meanings (Cruttenden 1997; Ladd 2008; Himmelmann and Ladd 2008; Gussenhoven 2004; Beckman and Venditti 2011). Intonation is one of the four meaning-bearing elements of language, alongside syntax, lexicon, and morphology. All languages have intonation, yet the intonational systems of under-resourced languages have been poorly studied relative to well-resourced languages like English, Swedish, German, Dutch, Swedish, French, and Japanese, for example. The majority of Australian languages lack full intonational descriptions, although a number of grammars from the 1970s onwards provide some impressionistic detail of typical intonational tunes used in spoken narrative or interactive discourse, e.g. Alawa (Sharpe 1972: 34–9), Alyawarra (Yallop 1977; Nungubbuyu (Heath 1984), Gooniyandi (McGregor 1990), and Gunwinygu (Carroll 1995) to name a few (see King 1994; Bishop 2002a; Simard 2010; Ross 2011; Jepson 2019 for detailed discussion of earlier treatments of intonation in Australian languages). More recent grammatical descriptions include more detailed descriptions of intonation patterns, e.g. for Bardi (Bowern 2012a), Jaminjung (Schultze-Berndt and Simard 2012), and Kayardild (Round 2009), and it is also a key feature of a number of substantial studies of information structure, discourse, and talk and interaction in languages like Bininj Kunwok (Garde 2013), Mawng (Singer 2006b), Murrinhpatha (Blythe 2015), Warlpiri (Simpson and Mushin 2008), and Garrwa (Gardner and Mushin 2015). It also features in a number of focussed studies of phonology and morphology in selected languages (e.g. Baker 2008b, Ngalagkan; Mansfield 2019b, Murrinhpatha). There have been several dissertations in recent years that provide excellent overviews of intonation in Australian languages (e.g. Bishop 2002a; King 1998; Simard 2010; Jepson 2019). In the

sections that follow, the major insights from these studies are discussed along with more recent approaches to intonation in selected Australian languages.

15.2 Prosodic structure and intonational categories in Australian languages It is widely agreed that intonation performs a basic delimitative or ‘chunking’ function across all languages, including the Indigenous languages of Australia. In other words, there are pitch movements that mark out the edges of a phrasal constituent. Some languages also have pitch movements that align to metrically prominent syllables or a foot that performs a prominence-enhancing strategy within the domain of an intonational constituent. Within the prevailing Autosegmental-Metrical (henceforth AM) framework (after Ladd 2008), these pitch events are modelled as underlying tone targets with f0 as the major phonetic exponent and they combine sequentially to give rise to a tune to signal a particular meaning or function. Languages like English, Japanese, Swedish, and French all have edge tones that delimit the edges of a phrase, particularly the right edge. These phrase edges are often accompanied by other junctural phenomena, namely lengthening, glottalization, and pauses. Tones at phrase edges are usually referred to as boundary tones, and they mark out stretches of speech into constituents called intonational phrases. In addition, the left edge of a constituent may be marked by a sharp rise in pitch. In languages like Korean ( Jun 1998), French ( Jun and Fougeron 2000), Ambonese Malay (Maskikit-Essed and Gussenhoven 2016), and Nafsan (Fletcher, Billington, and Thieberger 2019), left and right edge boundary tones are the main indicators of intonational constituency. The tonal events that are linked to a metrically strong syllable or foot are called pitch accents (see Gordon and Roettger 2017 for a useful survey of various languages). Tonal events also occur within a particular pitch range for a particular speaker. Pitch range in AM intonation models is usually thought of as the ‘grid’ or graph paper on

Janet Fletcher, Intonation. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Janet Fletcher (2023). DOI: 10.1093/oso/9780198824978.003.0015

159

janet fletcher which pitch accent, phrase, and boundary tones are scaled (e.g. Pierrehumbert and Hirschberg 1990; Ladd 2008). The extent of the grid is usually defined by the highest observed point in the f0 contour (the f0 topline) and the lowest f0 value observed for a particular speaker across utterances (the baseline). The work of King (1998) on Dyirbal is the first theoretical description of the intonational system of an Australian language within the AM approach. In fact, the majority of phonetic and phonological research on the intonation systems of Australian languages since the 1990s has been couched within an AM-type model with the exception of Simard’s (2010) analysis of Jaminjung that uses the PENTA framework (Xu 2005), which models f0 on a syllable-by-syllable basis. In the intonational descriptions of Australian languages to date, it is assumed that they have both pitch accents (i.e. prominence-lending intonational features) and boundary tones (edge marking intonational features), as in: Dyirbal (King 1998); Bininj Kunwok (Fletcher and Evans 1990; Bishop 2002a); Iwaidja (Birch 2002; Mailhammer and Caudal 2019); Bardi (Bowern 2012a); Mawng (Hellmuth, Ku¨gler, and Singer 2007; Fletcher, Stoakes, Singer, and Loakes 2016); Dalabon (Ross 2011; Fletcher 2014); Kayardild (Fletcher, Evans, and Round 2002; Ross, Fletcher, and Nordlinger 2016; Round 2009); Warlpiri (King 1999); Djambarrpuyŋu ( Jepson 2019); Jaminjung (Simard 2010); Yukulta (Bonnin 2014). For Pitjantjatjara, Tabain et al. (2012) propose tone targets that delimit the edge of major constituents, as well as signalling prominence. Given that pitch accents in Australian languages tend to be either at or near the beginning, or at or near the end of intonational constituents, these languages appear to reflect so called combined head-edge marking prosody (after Jun 2014a). In some models, more than one level of intonational constituency is assumed with accentual phrases (e.g. Dalabon, Fletcher 2014; Bardi, Bowern 2012a), or phonological phrases (Bishop 2002a) proposed as minor intonational constituents above the prosodic word to capture constituents that have smaller degrees of phonetic juncture than intonational phrases. Simard (2010) and Round (2009) also propose a level of constituency above the intonational phrase i.e. prosodic sentences in Jaminjung, and prosodic utterances in Kayardild respectively. As a general rule in the languages surveyed so far, there are relatively few distinctive pitch accent shapes compared to well-resourced Germanic languages like German or Dutch (e.g. see Gussenhoven 2004, and also Fletcher and Butcher 2014 for an overview of a basic tonal inventory for Dalabon). Conversely, a more complex inventory is proposed for Kayardild by Round (2009). The dominant pitch accent tends to be a simple H* tone target (realized as a high pitch target in a speaker’s pitch range) or L+H* (rising pitch movement) with the H tone phonologically associated with a metrically prominent syllable in a foot (e.g. Bininj Kunwok,

160

Mawng, Warlpiri), or with the initial syllable of a word (e.g. Pitjantjatjara). Some analysts have also proposed a H*+L sharply falling pitch accent that contrasts with a simple high H* accent (e.g. Jepson 2019 for Djambarrpuyŋu). These combine with either left edge high or low tones (e.g. %H or %L), or right edge low tones (L%), high tones (H%), low rising tones (LH%), or sustained high tones (e.g. H::H%) to give rise to a range of tunes (see Section 15.3 for an illustration of some of these tunes). Given the relative sparseness of pitch accent tones in these languages and the reduction of tonal register to effectively two levels, L (low) and H (high), additional diacritics are often used in some intonational descriptions to indicate local pitch upstep within an intonational phrase, e.g. ∧ H (e.g. Dalabon, Fletcher 2014; Yukulta, Bonnin 2014), and local pitch downstep, e.g. !H (e.g. Bininj Kunwok, Bishop 2002a; Bishop and Fletcher 2005; Bardi, Bowern 2012a; Yukulta, Bonnin 2014; Djambarrpuyŋu, Jepson 2019; Mawng, Fletcher et al. 2016). Additional pitch register tones have been used by researchers to indicate phrase-level pitch range adjustments including HiF0 to indicate a pitch topline within intonational phrases (e.g. Fletcher and Evans 2002 for varieties of Bininj Kunwok; Ross 2011, for Dalabon), and Bowern (2012a) uses %R to indicate an upwards pitch range reset at the beginning of an accentual phrase in Bardi. They also propose a putative M- right edge tone to indicate a flat contour that is followed by a pitch range reset for a following intonational phrase (Bowern, McDonough, and Kelliher 2012: 347). Final_Lo is also occasionally used to annotate an extra low pitch boundary tone that comes at the end of a sequence of intonational phrases (e.g. Fletcher 2014, Dalabon). For more detail about specific intonational transcription conventions within an AM framework, the reader should examine the various chapters in Jun (2014b), and specifically King (1998) for Dyirbal, Chapter 1.7 in Bishop (2002a) for Bininj Kunwok, Tabain et al. (2012) for Pitjantjatjara, Ross (2011: Chapter 3) and Fletcher (2014) for Dalabon, Round (2009: Chapter 6) for Kayardild, Bowern (2012a) for Bardi, and Jepson (2019: Chapter 7) for Djambarrpuyŋu. The following sections include some illustrations of these intonation features in particular functional contexts.

15.3 Sentence modality and grammatical structure In those languages with full intonational descriptions or for which there are impressionistic observations, a typically occurring pattern for declaratives is the so-called ‘flat hat’ pattern usually consisting of a relatively high pitch onset at the beginning of the intonational phrase followed by falling intonation at the right edge of the intonational phrase. In this respect, the intonation of declaratives in Australian

intonation languages follows a typical pattern observed in many of the world’s languages (e.g. Hirst and Di Cristo 1998; Ladd 2008). This tune can be minimally represented as a sequence of a H* pitch accent followed by a L% boundary tone. Further representations can be sequences of H* H* L% or H* !H* L% tone targets when there are one or more pitch accents (e.g. see Bowern 2012a; Bowern, McDonough, and Kelliher 2012 for Bardi, Fletcher 2014; Ross 2011 for Dalabon and Kayardild). In some languages speakers also use a sequence of rising L+H* pitch accents which is not dissimilar from the ‘pointed hat’ pattern observed in many of the world’s languages. The pitch accents generally occur on content words and the final L% boundary tone aligns with the final voiced segment of an intonational phrase. In a corpus study of narratives in Dalabon, Pitjantjatjara, and varieties of Bininj Kunwok, Fletcher and Butcher (2014) report that the dominant declarative tune across the three languages is a flat hat with falling intonation at the right edge (H L), ranging from just over 51% of all tunes for BGW to 67% for Dalabon, and 84% for Pitjantjatjara. King (1998) also reports a high proportion of falling declarative tunes in Dyirbal narratives (74%). Figure 15.1 shows typical declarative ‘pointy hat’ pattern for Mawng from interactive discourse.

500 300

Pitch (Hz)

200 100 70 50 Iwu-wunya-ng L+H*

this is often represented by a sequence of H* H% tone targets. This tune accounts for between 20–32% of tunes in the Dalabon and Bininj Kunwok corpus reported in Fletcher and Butcher (2014). The high or mid-level tune is also widely observed in many other languages including Alawa (Sharpe 1972), Nunggubuyu (Heath 1984), Kunwinjku (Carroll 1995), Jaminjung (Simard 2010), Bardi (Bowern 2012a), Iwaidja (Birch 2002; Mailhammer and Caudal 2019), Garrwa (Mushin 2012a), and Kayardild (Ross 2011; Round 2009). In all of the above languages, a striking sustained highlevel plateau-like intonation pattern is observed in particular kinds of narrative discourse. An example is shown below from the Kundedjnjenghmi variety of Bininj Kunwok in Figure 15.2. This is variously called the stylized high level intonation contour (e.g. Bishop and Fletcher 2005), narrative high monotone (e.g. Sharpe 1972), iconic lengthening (e.g. Simard 2010), or ‘linear lengthening intonation’—LLI (Mailhammer and Caudal 2019), and typically involves prolonged lengthening of the vowel in a final syllable of an intonational phrase. Mailhammer and Caudal (2019: 53) suggests that LLI conveys a quantificational or quantification-related semantic content. This kind of extreme lengthening is therefore generally associated with interpretations such as ‘prolonged or continuous actions’, listing, and so forth. Bishop (2002a: 82) uses the boundary tone configuration H::H% to distinguish normal high plateau-like tunes (e.g. H* H%) from these stylized lengthened high-level tunes in Bininj Kunwok. Simard (2013: 74–5) notes that in complex sentences in Jaminjung, iconic lengthening may be observed in the first IU (i.e. intonational phrase) representing some kind of quantificational content (duration of an event or geographical distance) and is often immediately followed or bound by a

jarripang L+H*

L%

500 1.277

0

300 200

Figure 15.1 An example of a declarative intonation contour produced by a female speaker of Mawng1 for the utterance Iwuwunyang jarripang ‘They were boiling up the trepang’.

Pitch(Hz)

Time (s)

100 70 50 ba-djal-wam ba- djal-

The other dominant tune in narrative discourse reported for many languages is a high flat or mid-pitch plateau and usually signals ‘continuation’ or incompleteness in narrative discourse. In AM descriptions of Australian languages, 1 My thanks to Ruth Singer and Salome Harris who provided the ELAN files for conversational discourse between speakers of Mawng, recorded on Goulburn Island in February 2012.

wa::m

H*

H::

0

H% 2.539

Time (s)

Figure 15.2 An example of the stylized high level intonation contour (or linear lengthening intonation) in the Kundedjnjenghmi variety of Bininj Kunwok for the utterance badjalwam ‘he went along…’.

161

janet fletcher second IU with sharply falling intonation expressing some kind of event-delimiting function (see also Schultze-Berndt 2002). The interpretation here is that the overall combination LLI plus sharply falling contours expresses ‘both their syntactic integration and their semantic relationship with the main clause’ (Simard 2013: 74). Distinctive intonational features are also associated with interrogatives in Australian languages.2 In some of the languages investigated so far, a Wh-question word like ‘where’, ‘what’, or ‘who’ is often utterance-initial and realized in elevated pitch register relative to surrounding speech material for the interlocutor (e.g. Dalabon, Fletcher 2014; Bininj Kunwok, Bishop 2002a; Jaminjung, Simard 2010; Djambarrpuyŋu, Jepson 2019). The rest of the utterance shows a downtrending pitch pattern after the initial question word. An example is shown in Figure 15.3 for Mawng where expanded pitch register is shown using the ∧ H* pitch accent which is realized on kiki, ‘what’. This expansion of the f0 topline (measured at the pitch peak associated with the question word) can be of the order of around six semitones compared to initial pitch accents in neutral declarative utterances produced by the same speaker. The other major pitch accent shape on question words is an expanded range rising accents e.g. L+∧ H* as in Dalabon (Fletcher 2014) and Bininj Kunwok (Bishop and Fletcher 2005). Regardless of the annotation conventions, this often steeply downtrending contour after the question word is a fairly typical overall pattern that is observed for content questions that are directed towards elucidation of entities or processes (e.g. after Blythe 2020). While the overall phrasal pitch pattern is similar to standard declarative intonation in many 350

Pitch (Hz)

200 100 70 50 Kiki yirrik arrkpanamin ta arukaruk? ^H*

!H*

0

!H*

L% 1.238

Time (s)

Figure 15.3 An example of an f0 contour illustrating a Wh-content question in Mawng: Kiki yirrik arrkpanamin ta arukaruk? ‘What are we doing in the afternoon?’ with an expanded range pitch accent ∧ H* on the initial Wh-word kiki. 2

See also Bowern, Chapter 36, this volume.

162

of the languages surveyed to date, it has been noted that these question utterances are largely distinguished by expanded pitch register rather than by tune itself. In Jaminjung (Simard 2010), polar questions are often realized in a more expanded pitch register than content questions. Tags tend to show weak rising or falling patterns relative to content or polar questions. Similarly, Jepson (2019) observes an expanded range rise-fall pattern at the end of yes/no questions in Djambarrpuyŋu. By contrast, Bowern (2012a) notes that content questions in Bardi usually terminate with rising intonation. This pattern is also observed in Dyirbal interrogative intonation, particularly for polar questions, but most are consistently realized with a suspension of phrasal pitch downtrends resulting in a more plateaulike contour (King 1998). Many analysts also comment on the similarity of overall falling tune among declaratives, interrogatives, and imperatives, with the main contrast being one of overall expanded pitch register for imperatives compared to interrogatives and declaratives (e.g. Simard 2010; Bowern 2012a).

15.4 Alignment with grammatical constituents As in many of the world’s languages, intonational phrases or intonation units in Australian languages frequently align with clause boundaries (e.g. see Nordlinger 1998a for Wambaya; Mushin 2012a for Garrwa; Bowern 2012a for Bardi; Croft 2007 for Wardaman; Ross, Fletcher and Nordlinger 2016 for Dalabon; McGregor 1990 for Gooniyandi; Simard 2010 for Jaminjung, among others). Given the highly agglutinative nature of many Australian languages, the number of grammatical words per intonational phase in Gunwinyguan languages like Kuninjku and Dalabon can vary between 1.9 to 2.4 in narrative discourse (Bishop and Fletcher 2005; Ross 2011). Analyses of prosodic/grammatical word alignment in Dalabon have also shown that grammatical words can be broken into two fully-formed intonational phrases suggesting a mismatch between prosodic structure and grammatical wordhood (Evans et al. 2008). In a related vein, Baker and Bundgaard-Nielsen (2016) show that in Wubuy, speakers can produce major phonetic junctures within grammatical words corresponding to the edges of nouns and inflected verbs, aligning them with an intonational phrase boundary. Conversely, prosodic integration is an important factor in the grammar/intonation relationship in many of these languages. In Gunwinyguan languages, it is not unusual to find two complex verbal elements bound together in a single intonation phrase to illustrate grammatical subordination

intonation or semantic cohesiveness (e.g. Evans 2003a; Bishop 2002a,b, for Bininj Kunwok). In their study of Dalabon, Ross et al. (2016) examine multi-verb intonation units (IUs) and note two major tunes consisting of either a high flat plateau or down drifting contour terminating with a L% low edge tone that often shows additional final lowering. They argue that the use of a single intonational constituent for complex units is relatively common in these multi-verb IUs whereas it has been proposed by others (e.g. Croft 2007; Jun 2014b among others), that complex constituents are more likely to be produced in separate major prosodic constituents. In a related vein, Simard (2010) proposes an even higher level intonational constituent in Jaminjung, a prosodic sentence, that is a constituent larger than an intonational phrase or an intonation unit and is defined largely by semantic function as well as prosodic criteria. She specifically states that the prosodic sentence consists of ‘sequences of IUs having a semantic relation with each other, a relation which is apparent in the structure of the clauses or non-clausal constituents corresponding to its IUs’ (Simard 2010: 69). They can be up to 13 or 14 syllables in length and are more likely to be followed by much longer pauses (around 1650 ms) than IUs which tend to be followed by shorter pauses (e.g. 600 ms). Prosodic sentences also show a greater magnitude of final lowering of the final pitch target at the right edge compared to IUs.

15.5 Discourse and information structure It is widely assumed that the tonal shape of an intonational phrase can depend on word order; however the non-configurational nature of many Australian languages including Warlpiri (Hale 1983), Bininj Kunwok (Evans 2003a), Wardaman (Merlan 1994), and the Nyulnyulan languages (Bowern 2012a) is widely attested (see Nordlinger, Chapter 4, this volume, and Brody, Chapter 35, this volume). It is important therefore to consider what consequences this has for intonation in Australian languages. As Mushin (2005b: 253) states, ‘while ordering of constituents in language use may not carry any grammatical weight …, it does contribute to the interpretation of discourse as coherent and meaningful’. Many researchers have therefore considered the relationship between prosody and information structure (IS) in Australian languages (e.g. Bishop 2002a; Singer 2006b; Hellmuth et al. 2007; Simpson and Mushin 2008; Simard 2010; Fletcher et al. 2016; Jepson 2019). Three major components of IS tend to be taken into account in relation to intonation and prosody—focus, givenness, and topic (after Krifka 2008). Ku¨gler and Calhoun (2019) outline different types of prosodic encoding of these information structure categories. Briefly, after Krifka (2008), focus implies selection

among alternatives, givenness refers to information status of constituent i.e. whether it forms part of the common ground, and topic typically refers to a constituent that the rest of the sentence is effectively about i.e. an ‘aboutness’ topic. Topics can also be contrastive. Languages vary in the extent to which they use different kinds of intonational features to signal these different IS categories, including pitch register manipulation, special types of pitch accents, and prosodic phrasing. With the exception of Simard (2010) for Jaminjung and recent analyses of Djambarrpuyŋu ( Jepson 2019) that consider a wide range of IS categories, most intonation studies of Australian languages have examined focus realization, reflecting a widespread tendency in information structure studies of prosody across languages (Baumann and Ku¨gler 2015; Ku¨gler and Calhoun, 2019). What is apparent from the quantitative investigations conducted so far is that distinctive rising pitch accent shape that is found on words in narrow focus constructions is less important than in languages like German, English, or Portuguese (see an overview in Ladd 2008; Ku¨gler and Calhoun, 2019). Hellmuth et al. (2007) present compelling experimental evidence to show that a focal pitch accent in Mawng is largely a hyperarticulated upstepped version of a ‘regular’ L+H* pitch accent and does not constitute a distinctive pitch accent category. Later work manipulating word order and information structure confirms this for Mawng (Fletcher et al. 2016). Jepson (2019) observes two different pitch accent shapes H* and a falling pitch accent H*+L in two corrective focus contexts in Djambarrpuyŋu, with speaker-specific preferences for one over the other accent-types and no consistent focus-related pattern across the board. Most studies have found clear pitch register cues that mark narrow or contrastive focus in various languages (e.g. Jaminjung, Simard 2010; Bininj Kunwok varieties, Bishop 2002a; Kayardild, Round 2010a; Mawng, Fletcher et al. 2016; Bardi, Bowern 2012a) with the focused constituent or word realized in elevated register and/or widened pitch span, and reduced post-focal pitch register for non-focal mater ial. Word order also plays a significant role in most of these studies with a narrow or corrective focussed constituent e.g. an NP or nominal, often prosodically left-dislocated or ‘fronted’ in the discourse. Indeed, it has been widely observed in the grammatical literature that initial position in a clause is a site of discourse prominence (e.g. Bowern 2012a; Simpson and Mushin 2008; Singer 2006b). This usually has intonational consequences as pitch span (i.e. the width of the pitch ‘grid’) is generally widest at the beginning of an utterance. Nevertheless it appears that even greater widening of pitch span, raising of pitch register i.e. producing a higher pitch peak on an accented syllable, and inserting a prosodic phrase boundary are common strategies in Mawng

163

janet fletcher to realize focal prominence (Fletcher et al. 2016). Similar strategies are observed in Bininj Kunwok (Bishop 2002a), Dalabon (Fletcher 2014), and a range of other Australian languages. Position of the focused element in an utterance has a clear effect on scaling of pitch accents across an intonational phrase or prosodic sentence, with final narrow or corrective focused tokens not necessarily having the highest scaled pitch accent in an utterance in Mawng for example (e.g. Fletcher et al. 2014). This is also observed in Djambarrpuyŋu ( Jepson 2019), where a more complex relationship is found among pitch register manipulation, focus, and topic marking. While downstepping (i.e. lowering of the H pitch target, usually represented as !H* in AM treatments) of the second or third pitch accent in an intonational phrase is relatively common in Australian languages and produces a similar perceptual effect to total ‘de-accenting’, some speakers of Djambarrpuyŋu actually de-accent non-focal material that is towards the right edge of an intonational constituent. This is a highly gradient phenomena and has not generally been observed in other Australian languages for which we have full intonational descriptions. For almost all of the languages described so far, pitch accents tend to be maintained in post-focal position. In the case of Mawng and Bardi, nominals receive a mandatory pitch accent (e.g. Singer 2006b; Bowern 2012a) if they appear at all in the discourse. In Yidiny, early and later pitch accents within an intonational phrase do not differ significantly in prominence (Bowern, personal communication). Another feature that is often referred to in the literature are the so-called ‘afterthought’ constructions. These typically occur after the main clause and are produced in their own full intonational phrase, usually after a pause (e.g. Bardi, Bowern 2012a; Garrwa, Mushin 2005b). In cases observed for Mawng and Dalabon, these tend to be realized with reduced pitch span compared to focussed clause-initial constituents. Nevertheless they are usually associated with an upwards pitch range reset, disrupting any pitch declination trends from preceding material. Pitch register modifications are also an important index of more global aspects of discourse segmentation and different types of talk and interaction in Australian languages, as in many other languages of the world. Fletcher and Evans (2000) compared global and local pitch topline trends in a corpus of Bininj Kunwok narratives and found that pitch reset at the start of an intonational phrase often signals the beginning of a major new discourse segment, and final lowering (an extra low boundary tone) often signals the end of the segment. These trends also tend to be associated with the start and end of a chain of successive intonational constituents. Similar pitch register patterns have been observed in varieties of Bininj Kunwok (Bishop 2002a), Dyirbal (King

164

1998), Dalabon (Ross 2011), and Jaminjung (Simard 2010). An upwards pitch register shift is also often a key indicator of ‘reported’ speech in a wide range of languages including Gooniyandi (McGregor 1990), Kayardild (Round 2009), Murrinhpatha (Blythe 2009), and Gunwinyguan languages including Bininj Kunwok varieties and Dalabon (Evans et al. 1999; Garde 2013), usually without the accompanying overt frames like ‘she said’, ‘he said’. There is an abrupt upwards resetting of pitch range from the surrounding talk, often with accompanying voice quality modifications. Raised pitch register and exaggerated pitch contours are also typically associated with ‘baby talk’ (specifically the language style used to imitate the speech habits of very little children3 ) in languages including Warlpiri (Laughren 1984), Bardi (Bowern 2012a: 63), Arandic (Turpin et al. 2014), Murrinhpatha (Forshaw 2016), and Gurindji Kriol ( Jones and Meakins 2013), along with a range of other phonological modifications. Jones and Meakins (2013) describe the intonation of janyarrp as involving a sharply falling intonation contour after the first couple of syllables of an utterance which the language consultants in their study refer to as ‘soriwei of talking’, i.e. a particular tone of voice and mode of speaking to express grief or some kind of longing ( Jones and Meakins 2013: 182). In a similar vein, Ponsonnet (2018a) describes the ‘compassionate contour’ in Dalabon which involves a sharp fall in pitch register across the first morpheme or the first two syllables of an intonational phrase followed by a long mid-level plateau accompanied by other modifications in voice quality. A similar type of contour is also found in other Gunwinyguan languages including Kunwinjku, and it is also found in Bardi. These are areas which require a great deal more investigation in other Australian languages.

15.6 Conclusions In sum, while the intonation systems of Australian languages may lack the array of pitch accent types or elaborate tunes found in Germanic languages like English, they use other intonational devices such as phrasing, pitch range, and glottal adjustments in addition to contour to give rise to a variety of linguistic and paralinguistic affects. This review is by no means exhaustive as researchers have referred to the tone of voice in Australian languages for decades. What is apparent however is that more detailed investigation is needed for those languages including Kriol that are still being acquired by children, particularly as languages continue to be lost at an alarming rate. 3

See Davidson, Chapter 60, this volume.

chapter 16

Sound change Barry Alpher

16.1 Introduction Australian-language phonemic inventories, right across the continent, show a considerable degree of uniformity, often remarked upon.1 Over wide stretches, subsets of languages, contiguous and non-contiguous, have identical inventories. On this basis one might expect that there has been little or no sound change throughout history, but this expectation is not borne out: the place has been seething with soundchanges, though likely not to the extent recorded in, say, the Indo-European-speaking area.2 There have been plenty of changes that do not alter phonemic inventories, as well as innovations of sound types and categories. It does appear that sound changes involving large differences in point (as opposed to manner) of articulation of consonants are vanishingly rare in Australia, apart from *k > c > /th/ (dorsovelar to alveopalatal to dental, if my hypothesis is correct; see Section 16.9).3 Here, I survey sound changes attested in and across Australian-languages, doing this via cross-cutting categories: by kind of process (lenition, fortition, initialdropping, cluster simplification, and so on4 ) and by product (effects on phoneme inventories), with cross-references where appropriate. In no case do I tally ALL the sound changes attested in the history of any given language, and 1 Recent examples include Baker (2014: 141) and Dixon (2002: 547). See also Bowern (2017b) regarding the pitfalls of notions like ‘common in’, ‘widespread in’, and ‘almost absent from’ applied to features of Indigenous languages in Australia and, for that matter, the world at large. 2 This discrepancy, if it exists, has not to my knowledge been documented quantitatively. But Austin’s (1981b: 300, 307) observations of absence of phonological change are noteworthy: ‘Bayungu and Dhalandji reflect ProtoKanyara forms exactly while Burduna has undergone some striking phonological developments’, and ‘Reconstruction of Proto-Mantharta shows that all daughters except Dhargari reflect the proto forms unchanged’. See further Miceli and Round (2022) and Bowern (2002) for discussion. 3 Compare pIE-to-Latin *dh > /f/ (fumus ‘smoke’, Sanskrit dhu:ma ‘smoke’), Latin-to-Romanian /kt/ > /pt/ (octo: > opt ‘eight’), and the merger of ProtoPolynesian *t with /k/ in Hawaiian. 4 Examples of sound-shifts that are not sound-changes under these criteria include Atnyamathanha initial [ß], an allophonic realization of /p/; and Kala Lagaw Ya /s, z/ from and replacing *c,*j, as in /kosar/ ‘two’ < *kucarra and /siw/ ‘liver’/ < *cipa (< *kipa). These pattern with the stops: consider the loanwords /seynz/ ‘change’ and /zon/ ‘John’.

it is clearly both inappropriate and impossible in an article of this kind to do so for the entire continent. I use the term sound change5 as it is defined and subtypologized by Blevins (2004: 32–9) on the basis of the phonetics of production and the perception, expectations, and construals of the hearer, and effecting a change (phonologization, in a loose sense6 ) in the system of contrasting sounds: phonemic split, merger, loss, or generation. Using data from language families with published historical reconstructions of protolanguages where available, and using citations from neighbouring languages that most closely resemble reconstructed forms where not, I take a neogrammarian7 approach to regularity of sound change, calling attention to phenomena that are problematic for this approach, and with this stipulation that the domain for the use of the terms ‘initial’ and ‘final’ is (unless otherwise specified) the word. Certain sound-change outcomes have an areal nature, often, but by no means always, in contiguous or nearcontiguous languages. The areal developments treated here are not the results of direct diffusion of loanwords,8 but rather are genuine sound-changes, likely promoted by the practice of marriage with owner-residents of neighbouring territories: members of different clans who in some areas (for example, Cape York Peninsula and North-east Arnhem Land) speak different languages (see Sutton 1991). Heath (1978a: 31) terms this ‘indirect diffusion, whereby the pronunciation patterns of one language trigger convergent historical developments in the other’. Examples in 5 ‘Sound change’ in this technical sense is distinct from shifts in sounds at the pre-phonological level. Where there is a potential misunderstanding, as in the term ‘chain shift’ applied to a series of true sound changes, I alert readers to the discrepancy. 6 Kiparsky (2015) gives a narrower definition (loss of conditioning environment), which does not encompass all of the phenomena I am calling phonologization; see my discussion of Kugu Nganhcara [nhc] (Section 16.7) with its footnote. 7 As defined and discussed by M. Hale (2003: 344–6): regular under strictly phonetically defined conditions, grounded in the individual, very local, and carefully distinguished from change propagated by diffusion, however regular. Note that Hale is using ‘neogrammarian’ in the phrase ‘neogrammarian sound change’ as a limiting adjective: in his frame of reference, there are other kinds of ‘sound change’. His model of the sound-change process (2003: 345–6) is roughly the same as Blevins’s. 8 See Heath (1978a: 63).

Barry Alpher, Sound change. In: The Oxford Guide to Australian Languages. Edited by Claire Bowern, Oxford University Press. © Barry Alpher (2023). DOI: 10.1093/oso/9780198824978.003.0016

165

barry alpher the treatment below include increases of consonant and of vowel inventories (Sections 16.2–4; a likely strong example at Section 16.2.2.3 is the innovation of distinctive prenasalized stops in Mpalitjanh and Yinwum).9 Other mechanisms, not mutually exclusive with the above or among one another, include pre-existing tendencies not realized before the phylogenetic separation of a single language into two,10 and a common substrate language (see further Miceli and Bowern, Chapter 5 and Angelo, Chapter 56, this volume). I distinguish my approach from Dixon’s (2002: 558–658), which covers much of the same ground and at much greater length. Dixon (2002: 102, 126–7) takes a typologyand-diffusion-prior approach (see commentary by Alpher 2006: 796), working from ‘canonical systems’ (Dixon 2002: 547–8) of phonemic inventories and ‘basic forms’ (102) of words, rather than from protolanguages and reconstructions (in a rigorous sense). Regarding questions of soundchange regularity in Dixon’s (2002) work, see Alpher (2006: 796–7).

16.2 Changes in consonant inventories: manners 16.2.1 Innovation of contrastive obstruent manners In the Australian continent as a whole the commonest stop inventory is a single series, undifferentiated as to manner (see also Round, Chapter 10 this volume).11 However, in a fair number of languages, mostly but not exclusively northern, there are manner contrasts among stops, variously described as fortis/lenis (tense/lax), long/short, 9

See also Bowern (2018b), especially slide 35. Sapir’s (1921: 184–92) linguistic drift; his example is vowel umlaut in Old English and Old High German. 11 The following conventions are used in this chapter, in addition to the conventions that are detailed in the volume preface. Lower-case ‘p’ prefixed to language-family names is to be read ‘proto’, as with pPNy for protoPama-Nyungan. Curly brackets {} enclose basic forms in an ordered-rules approach. Italics are used for transcriptions reconstituted from old sources. Regarding the laminals c, j, ñ, ʎ vs. th, dh, nh, lh: I make the working assumption (not uncontroversial) that proto-Pama-Nyungan did not contrast an alveopalatal series with a dental series, and I use the symbols *c, *j, *ñ, *ʎ to represent the reconstructed sounds; I postulate further that each of these had an alveopalatal allophone after a heterorganic consonant and before *i and a dental allophone elsewhere, as in the Warburton Ranges dialect of the Western Desert Language (Douglas 1964: 14; Alpher 2004a: 113). This means that the creation of the dental/alveopalatal contrast in Pama-Nyungan is a matter of one or more sound changes from the sounds labelled *c etc. and is apt to cause notational confusion when one of the results, also labelled /c/, has come into contrast with the other result, /th/. A bit of extra awareness is called for in the reading of the last paragraph of Section 16.7 below. 10

166

voiceless/voiced, aspirated/unaspirated.12 In many but by no means all Australian languages where fortis and lenis have demonstrably become contrastive for stops, fortition of stops (/p/ vs. /b/, etc.) has taken place after a short V1 (contrastive with a long V1 ) and a non-nasal C1 . Contrastive fortis nasals (/pm/ or /bm/ vs. /m/) are formed under the same conditions. Both sets become contrastive if vowel length is neutralized and (with regard to the nasal tenseness contrast in some but not all initial-dropping languages) if C1 is lost.13 There is a diversity of paths to stop contrasts often classed under the ‘voiced/voiceless’ or ‘lenis/fortis’ feature. I am thinking particularly of the conditioning of the genuinely ‘voiced’ series, which in the data I have surveyed is a matter of simplification from prestopped nasals (Rimanggudinhma, Lamalama, and Flinders Islands; Section 16.2.1.7) and of loss of the nasal in nasal+stop clusters (Yandruwandha; Section 16.2.1.4), as opposed to most of the remainder of instances, in which nasality figures as a conditioning factor only in the C1 position and the result is arguably a lenis/fortis distinction (see Section 16.6.2 on Arrernte and Uw-Oykangand). It is likely of importance with regard to further developments in CYP languages (see especially Lamalama, Rimanggudinhma, and Flinders Islands data at Section 16.2.1.7) that the stops of prestopped nasals in CYP languages are voiced, whereas those in Arandic languages are voiceless (Section 16.6.2). Note that in Tharrgari, loss of nasals in *nasal+stop has also worked in the opposite direction, producing fortition of the stop (see Section 16.2.1.6).

16.2.1.1 Contrastive stop series possibly due to vowel-length merger Warumungu is a C1 -retaining language which, following loss of the V1 length distinction, now has contrastive stop series.14 Stops labelled voiceless (orthographic doubled; Simpson and Heath 1982: 6) after a non-nasal C1 and a short V1 , as with /pukka/15 ‘rotten’ < *puka and /wankka/ ‘alive’ 12 Fletcher and Butcher (2014: 101–2): ‘A group of languages spoken largely in the northern part of Australia also have a more complex stop series which is described as a fortis-lenis contrast or a short-long contrast’; they mention also a voicing-onset time contrast (aspirated/unaspirated) (105) and observe that ‘intervovalic long stops are produced with higher articulatory intra-oral pressure’ (fortis/lenis) (105). There is no mention of true voicing contrasts. 13 A phonetic fortis/lenis difference in stops and nasals is attested in some Wik-subfamily languages (non-initial-dropping), for example Pakanh: [’a:ku] ‘ground’ < *Ta:kurr and [nhi:na] ‘sit’ < *ñi:na; vs. [kac:in] ‘yamstick’ < *kacin, [kuʔ:a] ‘dog’ < *kuta, [kuñc:i] ‘penis’ < *kulñci, [kam:u] ‘blood’ < *kamu, and [kan:i] ‘up’ < *kani. (In view of [nhap:in] ‘egg’ < *ñapin, it is only the length of V1 and not the nasality of C1 in that is responsible for the laxness of C2 in Pakanh [nhi:na] ‘sit’.) 14 The vowel-length distinction has now been restored, with long-V1 forms of no discernable etymological depth. 15 Many modern-era published analyses use fully phonemic transcriptions in practical-orthography notation. I use these transcriptions here, as

sound change < Western *wanka, contrast with voiced stops after a nasal C1 as in /mangka/ ‘hole in ground’ < *mangka and after a long V1 as in /lurnka/ ‘kingfisher’ < putative *lu:rnka. With the last example, compare the Western Desert and Warlpiri forms /lu:rnpa/ with a long vowel followed by /pa/ to avoid a consonant falling in final position.16 The two series are in contrast only word-medially. There is no fortis/lenis distinction for the nasals. Concerning the pronunciation of these sounds, Simpson and Heath (1982: 6–7) say ‘“Voiceless” stops are voiceless and usually aspirated. “Voiced” stops are unaspirated and often voiced. “Voiced” stops are never long, while “voiceless” stops are long depending on their position in the word’.

16.2.1.2 Stop manner contrasts possibly originating from loanwords Margany-Gunya (Breen 1981a: 283–4), alone in the Maric subfamily (PNy), has developed word internal contrastive voiceless (tense) and voiced (lax) stop series. The voicedstop words with deep etymologies, intervocalic and in clusters (I have found no voiceless stops in clusters), so strongly predominate statistically (list frequency) that in my opinion the voiceless-stop words are suspect as borrowings. Barrett (2005: 157) lists as a phonological change ‘the irregular development of a second [voiceless] stop series’. Yolngu (PN, Arnhem Land) contrastive lax (voiced) stops originate apparently exclusively in loans from surrounding non-Pama-Nyungan languages (Bowern and Alpher 2005, slides 7 and 8; we disagree here with Heath 1978a: 35 and Hamilton 1993: 17, 23–4).

16.2.1.3 Manner contrast in non-apical stops In Wik-Muminh (K. Hale 1976c), the non-apical stops *p, *th, *c, *k continue as voiced stops /b dh j g/ after a long V1 and as voiceless /p th c k/ otherwise, with the length distinction in V1 lost.17

16.2.1.4 Manner contrast in apical stops only The Karnic languages, for example Diyari (Austin 1981a: 16–19), contrast voiced apical stops (/d, rd/) with voiceless counterparts (/t, rt/). Austin (1990a: 178–9) reconstructs the voiced apical stops, together with the contrasting voiceless ones, to pKarnic, where they were presumably a for Warumungu and a number of other languages treated below, throwing the occasional IPA character where necessary for clarity. 16 I take my cue in this analysis from David Nash (pers. comm.). The attestation of factors conditioning lenis stops is skimpy, and there are residual forms across all categories. 17 In the dialect KuguMumminh (my fieldwork) the nasals have also split into distinctive lenis and fortis (long) series.

pre-Karnic innovation. He does not speculate as to their origin, but as /d, rd/ share phonetic traits with /rr, R/ it is worth considering whether /d, rd/ originated from them rather than from voiceless stops (see Breen 1997: 78–9). Voiced stops in the non-apical positions, however, are innovations that did originate from voiceless ones and are now in contrast with them in some of the other Karnic languages: Yandruwandha (Breen 2015: 10–14, who does describe the contrast as ‘voiced’ vs. ‘voiceless’), Yawarrawarrka, and Wangkumara. Austin (1990a: 179–80) accounts for their origin as follows (emphasis in original): ‘… reconstruction suggests that voiced stops (other than apicals) arose in these languages following a nasal (homorganic or non-homorganic) as the second member of a consonant cluster when the word-initial consonant in the proto-language was a nasal’. An example of /g/ < *k is /ngarn.ga/ ‘beard’ < pPNy *ngarnka; contrast (for /k/ vs. /g/, where *C1 was not a nasal) /darnka-/ ‘to find’ from Proto-Central Karnic *tarnka-.

16.2.1.5 A three-way obstruent manner contrast The Northern Paman language Mbiywom (K. Hale 1976a: 20– 2) contrasts voiceless stops /p th t c k/ with voiced stops /b dh d j g/ and also with spirants /β ð ɣ/. The contrasts are here illustrated with reflexes of pPaman *c and *y: /tha-/ ‘to bite’ < *paca-, /aða-/ ‘to die’ < *Ca:ñca-, /ija ~ iji/ ‘another’ < *wi:ya, /ci-/ ‘mother’s father’ < *ngaci, /ica-/ ‘to put’ < *ngi:ca-.18

16.2.1.6 Contrastive fortis stops conditioned by adjacent sonorants Two non-contiguous languages of Western Australia have acquired a tense/lax stop contrast via assimilation of sonorants in manner to following stops. One is Tharrgari (Mantharda subfamily of Nyungic; Austin 1981b: 308–9; 1992b), with for example /puga/ ‘bad, wrong’ < *puka, /ngartga/ ‘beard’ < *ngarnka, and /pugga-/ (orthographic puka-) ‘to dig’ < *pungka-. The fortis stops from homorganic nasalstop clusters, orthographically p, t, rt, c, k, contrast with lenis counterparts b, d, rd, j, g. The change *nasal > stop is a fortition; see Section 16.10.3. The other is Nhanda (Blevins 2001a: 39–41); reconstructions are regional words unless otherwise specified), with two sources of fortis stops. One is fortition (see Section 16.10.3) of non-nasal sonorants to nonsonorants (stops), as with /wurtka/ ‘ear’ < *kurlka and /atpi/ ‘tie’ < pPNy *karrpi-, followed by total assimilation to the second stop in obstruent clusters (in progress and with fluctuating forms in the contemporary language, as with /witku/ ~ 18 The conditions are not comprehensively statable because recorded data are scant (Hale 1976d: 20).

167

barry alpher /wiku/ ‘belly’, with /k/ in the latter variant contrasting with the /g/ of /wigi/ ‘saliva, spit’). A second and analytically problematic source of fortis stops is a preceding long vowel, as in /a:ci/ ‘little boy’ < *ga:jung; in other examples cited, the contemporary form has a short vowel presumed to have been originally long.

16.2.1.7 Contrastive voiced stops originating from prestopped nasals The Paman (Lamalamic subfamily; see Verstraete 2018a and Chapter 72, this volume) language Rimanggudinhma (one of the languages designated Kuku Warra; [Bruce Rigsby, pers. comm.]), of the Princess Charlotte Bay region (Sommer fieldnotes 1972–1974 and Rigsby 1972) is an initial-dropping language with contrasting voiced and voiceless stop series. I surmise that the voiced series developed from prestopped nasals19 at C2 (*bm, *dnh, *dn, *gng, themselves from earlier m, *nh, *n, *ng), which simplified to the voiced stop component. Attested in a very short wordlist are /b, dh, g/, as in (1) /ba/ ‘person, man’ (Ergative /bal/) < *pama, /bi|ð/ ‘grandmother’ < *kami (contrast /pa|ð/ ‘sister’ < *yapa), /dharr/ ‘crocodile’ < *kañarra (contrast /tha-/ ‘to bite’ < *paca-), /da/ ‘they (Pl)’ < *cana (contrast /uta/ ‘dog’ < *kuta), /garr/ ‘devil’ < *wangarr, and /ugol/ ‘mosquito’ < *kungul (contrast /ukon/ ‘grass’ < *pukan) Like Rimanggudinhma is Flinders Islands Language (Sutton n.d.: Section 2 Phonology), with a distinctive voiced stop series,20 as in (2) /woga/ ‘sun’ < *punga, /adhul/ ‘who’ < *wañu-l, /oda/ ‘excrement’ < *kuna

‘the transcriptions above are incompletely phonemicized’), there are prenasalized voiced stops (/mb, nhdh, ngg/), as in (5). (5) /mba/ ‘person’ < *pama, /nhdharr/ ‘crocodile’ < *kañarra, /nggul/ ‘mosquito’ and /nggaR/ ‘devil, white man’ < *kungul and *wangarr, respectively; contrast /dharr/ ‘dream’ < *picarra and /karr/ ‘bluetongue lizard’ < *pangkarra These contrast with both plain voiced stops and voiceless stops, as in (6) /burun/ ‘heart’ vs. /puna/ ‘by and by’ vs. /mba/ ‘person’ (7) /dhun/ ‘elbow’ vs. /thu/ ‘blood’ vs. /ndwa/ ‘excrement’ The prenasalized stops likely developed from plain voiced stops, as attested in Rimanggudinhma and Flinders Islands, with no preceding or following nasal context, and reflect a general tendency in south-west Pacific languages to use a preceding nasal to reinforce voicedness in stops;22 an example from Yir-Yoront that is doubtless a loanword via Kriol is /thampáyk/ ([mp] ~ [mb]) ‘tobacco’. The prenasalization of Lamalama /mb, nhdh, ngg/ can be taken as evidence that what distinguishes Rimanggudinhma /b, dh, g/ from /p, th, k/ is indeed the feature of voicing (rather than laxness). In other languages of this general area, prenasalized stops developed along a different path (conditioned by a nasal *C1 ) and have no association with stop voicing. See Section 16.2.2.3 below and, with regard to all claims of prenasalization as a stop manner-of-articulation, Round (Chapter 11, this volume).

These contrast with voiceless stops, nasals, and nasal-stop clusters (here illustrated for the lamino-dentals /th/, /dh/, /nh/, /nhdh/ and for the apicals /t, d, n, nd/) as in

16.2.2 Other obstruent series: spirants and prenasalized stops

(3) /wutha:la/ ‘over the top’, /thapil/ ‘sing’, /wadhi/ ‘the other’, /wanhdhe:wa/ ‘today’, /nhangal/ ‘cold’

16.2.2.1 Spirants

21

(4) /to|rr/ ‘dog’ < *kuta, /yada/ ‘they (Pl)’ < *cana, /ngunil/ ‘grind’, /yundu/ ‘you (Sg)’ < *ñuntu Corresponding to Rimanggudinhma /b, dh, g/ etc. in another Lamalamic language, Lamalama (see Verstraete 2018a and Rigsby 1972, who cautions in a note that 19

These are nasals that have undergone fortition; see Section 16.6.2 regarding similar fortition in Arrernte. 20 Sutton is clear that: ‘The “voiced” and “voiceless” series are both lenis and unaspirated and roughly of equal duration, with glottal tone apparently being the only distinguishing factor’. It is not clear from the data available what the conditioning factors are. 21 The vertical stroke (‘pipe’) here and below separates material not part of the original etymon, presumably a morphological addition at a later date.

168

Lenition of the stops *p, *th, *k to (voiced) spirants /β, ð, ɣ/ (continuing in some of the languages with alternates in /w, y, w/, with which they merge) took place in the Northern Paman languages Mpalitjanh-Luthigh, Linngithigh-Alngith, Awngthim-Ntra’ngith, and Uradhi in C2 after a preceding long vowel or in VC3 V.23 Examples from Uradhi (K. Hale 22 As for example in Fijian [ndoketa] for English doctor. The lands of Lamalama are in the Princess Charlotte Bay region. The Bay was frequented, 100 and more years ago, by pearling luggers whose crews included many South-west Pacific islanders. I surmise that this [b] > [mb] (etc.) change was a contact-induced phenomenon. 23 These contexts also condition mergers, as with stops to glides in Yir-Yoront *p > /w/, *th and *c > /y/, *t > /R/, and *k (> *ɣ) > zero; see Alpher (2004a: 101 and fn. 11) and also in Walmajarri (Alpher 2006: 797).

sound change 1976c and Crowley 1983) and Mpalitjanh (K. Hale 1976d), respectively, are /waða/ and /aða/ ‘crow’ < *wa:ca (vs. /watha-/ and /tha/ ‘to bite’ < *paca-), /utaɣa/ ‘dog’ < *kutaka, and /miñaɣu/ and /inhaw/ ‘for meat’ < *miña-ku.

16.2.2.2 Spirants originating from nasal-stop clusters In other Paman languages, the spirants /β, ð, ɣ/ originate from a different source, nasal-stop clusters, after long vowels or in more intricate conditioning contexts: Aritinngithigh /aβi/ ‘ashes’ < *Ni:mpi, /aða-/ ‘to die’ < *Ca:ñca-, and /uɣu-m/ ‘long’ < *Cu:ngku; nasal-stop clusters following short vowels remain as such: /mpayrr/ ~ /mpirr/ ‘up’ < *kampiy, /ngkwu/ ‘knee’ < *pungku. Like Aritinngithigh in this regard but with variations of scope and conditioning are Uradhi and Awngthim.

16.2.2.3 Prenasalized stops In Yinwum (K. Hale 1976d: 11), intervocalic stops at C2 were prenasalized if *C1 was nasal: n

n

(8) / pi/ ‘one’ < *ñipi, / ci-/ ‘mother’s father’ < *ngaci, /n kurr/ ‘mother’s older brother’ < *mukuR Nasal-stop clusters following nasal C1 lost the nasal segment but retained the stop as prenasalized: (9) /in kwe-/ ‘to smell’ < *ñu:ngkaAll of the above contrast with nasal-stop clusters in which C1 was non-nasal: (10) /mpu/ ‘urine’ < *kumpu, /nti-/ ‘spear w multipronged spear’ < *yinta-, /ngkwa-/ ‘to cry’ < *Rungka Developing closely similarly (but not quite identically) to Yinwum was its near-neighbour Mpalitjanh. With nasal *C1 followed by short *u or *i and a nasal-stop cluster at C2 , both nasals and the intervening vowel were lost: (11) /itu/ ‘you (Sg)’ < *ñintu With an initial nasal, short *a or any long vowel at V1 , and a nasal-stop cluster at C2 , C1 (but not V1 ) was lost, as was the segmental nasal in the cluster, but the stop survived as prenasalized: (12) /an pu/ ‘tooth’ < *ngampu, /an ka/ ‘mouth’ < *ñangka Otherwise, C1 (but not V1 ) was lost and the nasal in a nasal– stop cluster remained: (13) /ampi/ ‘up’ < *kampiy, ungku ‘knee’ < *pungku The contrast between prenasalized stop and nasal-stop cluster is realized phonetically as, respectively, a nasal segment homorganic with a following voiceless stop ([ampu] ‘tooth’,

[angka] ‘mouth’) vs. a nasal segment followed by a voiced stop ([ambi] ‘up’, [onggo] ‘knee’).24 Other CYP languages for which prenasalized stop series have been postulated include Lamalama (see Section 16.2.1.7 above), Mpakwithi (Crowley 1981: 153), and Awu Alaya (Kuku-Thaypan; Rigsby 2011). In the latter two languages, no contrasting nasal+stop clusters are attested, and in AwuAlaya heterorganic nasal+stop clusters are common. For these languages, the prenasalized-stop analysis is justified on grounds of parsimony of phonotactic statement alone (Crowley 1981: 155; Rigsby 2011: chapter ‘Phonology’, especially pp. 13–14) and is questionable.

16.3 Lenition conflating stops and liquids with corresponding glides 16.3.1 Lenition of initial stop Lenition of initial stops was sporadic (Alpher 1976) in Kaanytyu–Umpila–Kuuku-Ya’u, Umpithamu, Flinders Island, and Uradhi, all Paman. Examples include Umpila /yuma/ ‘fire’ < pPaman *cuma (conflating initial /y/ < *c with pre-existing initial /y/ in /yaʔa/ ‘older sister’ < pPNy *yapa) vs. /thulʔi/ ‘belly, stomach’ < pPaman *culpi. Verstraete (2018b) is currently studying these phenomena in detail. Initial consonant lenition was regular, conditioned by following segments, or unconditioned, in a number of instances. In Nhanda, initial *k > w preceding *u where C2 was not palatal or velar, as in /wumbu/ ‘urine’ < *kumpu and /wuna/ ‘faeces’ < *kuna (Blevins 2001a: 39), falling together with the /w/ in /wuR7aRa/ ‘black, dark’ and /wuR7a/ ‘stone’. In Iwaidja and Garig/Ilgar (Warrkbi branch, Iwaidjic subfamily of Iwaidjan), initial *p > /w/, *c > /y/, *k > *c (> y, optionally > zero) in *ki, and *k > /w/ in *ku (Evans 1997b: 245); in Mawng (248–9), initial *p > /w/, *c > /y/, *k > /y/ in *ki and /w/ in *ku, and *w > zero. However, in Nungali (Mirndi family), lenition of initial stops appears to have taken place without conditioning by context: initial *k>w, *p>w, *j>y (Harvey 2008b: 17–18).

16.3.2 Lenition of medial stops A number of languages attest lenition of medial stops. One is Jawony (GW family; Harvey 2003c: 223), in which medial intervocalic lenis stops lenited to their corresponding (pre-existing) glides: pGW *pirti > piRi ‘beeswax’, *picip > 24 Hale did not specify the phonetic difference in his published article, but it is clear from his Mpalitjanh fieldnotes (Hale 1960: 1).

169

barry alpher /piyip/ ‘to squeeze’, *pukurr > /puwurr/ ‘dream’, *kartap > /kaRap/ ‘spider’, *melpe > /melwe/ ‘mud’. Nunggubuyu (Wubuy) (Section 16.3.4 below for details) has done much the same, but context-free and with subsequent (or concomitant) degemination of the fortis stops. In Mawng (Iwaidjan), medial *p, *rt, *c, *k (respectively) merged with pre-existing /w, R, y, ɣ/ (Evans 1997b: 249). Bardi (Nyulnyulan family; Bowern 2007b: 7–8 and fn. 4) lenited *p to /w/ between vowels following a stressed syllable, as in /ji:wa/ ‘boomerang’ < *ji:ba and /jarrgawiñ/ ‘frog sp.’ < *jarrkabiñ and lost *p between identical vowels, fusing the newly adjacent identical vowels into a single short vowel. Intervocalic *k and *b were lost between identical vowels, as in /i:ndu/ ‘curlew’ < *wiinduku, /burru/ ‘kangaroo’ < *burruku, and /gurlil/ ‘turtle’ < *kurlibil. Yir-Yoront (Pama-Nyungan), after a preceding long V1 and in forms that are unstressed relative to what immediately precedes or follows, lenited stops to glides at corresponding positions: *p > /w/, *th and *c > /y/, *rt > /R/, and *k > zero (Alpher 1991: 12; 2004a: 101). Examples: /woy/ ‘song, ceremony’ < *wu:ci, /puymál/ ‘right-hand’ with /puy/ < /puth/ < *puñca ‘arm’, both with /y/ falling together with the pre-existing /y/ of /may/ ‘vegetable food’ < *mayi.25 In Gamilaraay and Yuwaaliyaay (a branch of Central New South Wales; Austin 1997a: 26, 29–30),26 intervocalic *k > y if adjacent to *i, as in Gamilaraay and Yuwaaliyaay /bayin/ ‘sore’ < *bagin and /miyay/ ‘girl’ < *migay, falling together with the /y/ of /giyal/ ‘afraid’ < *giya; but if the surrounding vowels are exclusively *a or *u, then *k > w, as in Gamilaraay /dhawu|n/ ‘ground, country’ < pPNy *Ta:ku, falling together with the /w/ of Yuwaaliyaay /guwa:/ ‘rain’ < *guwang.

16.3.3 Lenition of liquids to glides Lenition of intervocalic *rr to /R/ has occurred in Nyungar (Douglas 1976), as in /waRa/ ‘bad’ < pPNy *warra and /koR/ ‘behind’ < pPNy *kurri, merging with the /R/ of /maR(a)/ ‘hand’ < pPNy*maRa, though realization of the rhotic appears to have been variable (cf. Bowern, Smith-Ali, et al., Chapter 74, this volume). And lenition of intervocalic *rr to /y/ (not homorganic) has taken place in several Wik languages: Wik-Muminh /thayan/ ‘hard’ < PP *carran, merging with /y/ as in /ngaya/ ‘I’ < *ngayu and /mayi/ ‘vegetable food’ < *mayi (K. Hale 1976c). 25 Another lenition-product in YYoront is /lh/, as in /yulh/ ‘earth, soil’ (KThaayorre /(Ra:k-) yo:ñc/ ‘earth, soil’, Local *yu:ñca); the contexts that differentiate this /lh/ from /y/ as lenition products are not entirely clear to me. 26 Reconstructions are Proto-Central New South Wales unless otherwise specified.

170

16.3.4 Chain shifts Iwaidja (Iwaidjan family) underwent a series of sound changes, ‘the great Iwaidjan consonant shift’ (Evans 2009a: 162–3). The medial long stops *p:, *th:, *t:, *rt:, *c:, *k: (reconstructed on the basis of cognates in GW) lenited to short stops; the short stops *p, *th, *t, *rt, *c, and *k lenited, respectively, to /w, lf [a flapped lateral; see 16.11.12], rr (~ zero), R, y, and ɣ (y/i_i and w /u_u)/ in the context *V(L)_V, where L is a liquid (Evans 1997b: 246; 1998: 131; 2009a: 162 and Table 2); and the pre-existing glides *y and *w disappeared. The question of directionality (‘push chain’, ‘drag chain’, or simultaneity) is an open one. A chain shift also took place in Nunggubuyu (Wubuy) (GW family), in which, however, it was not limited to medial stops. Here the stops of the pGW plain series lenited to approximants, with *p, *rt, *th, *c, *k > (respectively) /w, R, lh, y, w/, all except /lh/ (which is not attested elsewhere in GW) merging with pre-existing glides; see also van Egmond, Chapter 68, this volume. The geminate (fortis) stops subsequently lenited to singletons (lenis). (Heath 1978a: 37–41; Harvey 2003c: 211).

16.4 Changes in vowel inventories Three-vowel systems have been postulated for ProtoWorrorran (McGregor and Rumsey 2009: 34), Proto-Mirndi (Harvey 2008b: 16–22), Proto-Iwaidjan (Mailhammer and Harvey 2018: Section 5.7), Proto-Nyulnyulan (with three ± length; Bowern 2007b) and pPNy (three ± length; O’Grady 1998: 217; Alpher 2004a: 107–9). Proto-Gunwinyguan (Harvey 2003c: 221) is reconstructed with five vowels and Proto-Tangkic with four (Section 16.4.1). A number of PNy languages in CYP (Sections 16.4.1–16.4.3,), Victoria, and the South-west (Section 16.4.2) have developed inventories of four or five (or more) vowels. Most of the larger systems in CYP, although many are in contiguous languages, have developed along separate pathways.

16.4.1 A couple of four-vowel systems In Ntra’ngith (Northern Paman27 ) *a in V2 > e before final *l, which is subsequently lost, hence /ke/ ‘foot’ < *ñukal; contrast /nga/ ‘sun’ < *punga, /la-/ ‘they Du’ < *pula. The contemporary system is /i e u a/. 27 In Northen Paman languages, all length distinctions in V1 were neutralized, subsequent to effects on C2 and V2 (K. Hale 1976a: 8–22).

sound change Lardil, a branch on its own of the Tangkic family, has a four-vowel system (/i e u a/) ± length (reconstructed to Proto-Tangkic). The /e/ corresponds to /a/ in the other branches of the family, and Round and Evans (nd: Section 2, Section 5.5.1) reconstruct *e, *e: in contrast with *a, *a: and with *i, *i: in stressed syllables (the first syllables of roots and stressed suffixes) for pTangkic. Examples (Tables 23 and 18): Ganggalida /rlarla(tha)/, Lardil /ReRa/ ‘drop it’ < pTangkic */rlerla/; contrast Ganggalida /rtirralta/ and Lardil /rtirral/ ‘slippery’ < *pTangkic *rtirral, and Ganggalida /thartara/ ~ /tharta:/ ‘lower arm’ and Lardil /tharta/ ‘shoulder’ < pPtangkic *tharta. In subsequent syllables, Lardil developed /e/ regularly from *i, and less regularly by other routes as well. Contrastive *e and *e: likely originated in pre-Tangkic.

16.4.2 Development of some five-vowel systems Linnngithigh (Northern Paman; K. Hale 1976d: 13) gained /æ/ and /o/ for a contemporary system of /i u æ o a/; the conditioning is not fully statable given the data at hand, but an illustrative emergent contrast is /mæ/ ‘fire’ < *cuma, with /æ/ < *a as conditioned by the laminal C1 , vs. /ma/ ‘man’ < *pama. In Wik Muminh (PNy, Wik subfamily, Kugu Nganhcara dialect cluster; see Smith and Johnson 2000 and K. Hale 1976e: 50–60), vowel-length contrasts were neutralized; *a(:) in V1 > /e/ if C1 was *y (/yengan/ ‘hair’ < *yangan) or if C2 was a cluster beginning in *l (/kek/ ‘spear < *kalka, /theba/ ‘wind’ < *ca:lpa); with these contrast /kaci/ ‘far’ < *kaci and /paba/ ‘breast’ < *pa:pa; and *u(:) > /o/ if C2 was a cluster beginning in *l, as in /woñje/ ‘gather together (intransitive)’ < *wu:lñci- (contrast /muñji/ ‘swim/ < *mu:ñci- and /wañci/ (KuguMumminh /wan.ci/) ‘sick’ < *wan.ci). Yir-Yoront, Yirrk-Thangalkl (Yirrk-Mel), and Kok-Kapér (a phylogenetic subgroup within South-west Paman; Alpher 1973) share among other things a five-vowel system. The distinctions in V1 evolved from a three-vowel system in part as indicated below; the changes in V1 illustrated in (14–18) are identical for the three languages and no doubt took place in their common ancestor. Subsequent changes to V1 and V2 , loss of V2 , and neutralization of vowel length took place independently in each of the three languages. The following examples (originally disyllables) are from Yir-Yoront (here and below, the numbering of the examples does not indicate chronological ordering): (14) *C1 a C2 V > /C1 aC2 / as in /pam/ ‘person’ < *pama, /kam/ ‘blood’ < *kamu, /kal/ ‘dew’ < *kali (15) *C1 iC2 a > /C1 iC2 / as in /pin/ ‘ear’ < *pina; *C1 uC2 a > /C1 uC2 / as in /thum/ ‘fire’ < *cuma

(16) *C1 iC2 i and *C1 iC2 u > /C1 eC2 / as in /theR/ ‘edge, boundary’ < *kiRi (Yidiny /giRi/) and /pel/ ‘hip’ < *pirlu (Yidiny /bilu/) (17) *C1 uC2 u and *C1 uC2 i > / C1 oC2 / as in /poʔ/ ‘knee’ < *pungku and /kol/ ‘anger’ < *kuli (18) *C1 uC2 a > C1 uC2 as in /puth/ ‘arm’ < *puñca There are other paths to the mid vowels, for example *a > /e/ preceding a following consonant cluster beginning with *l as in /ngelʔeR/ ‘tongue’ < areal *ngalkaRi, but in view of /kalʔ/ ‘spear’ < *kalka, the conditioning factors are not fully statable on the basis of data at hand. In some varieties of Nyungar (Nyungic, PNy; Douglas 1976), very briefly stated, *a > e in *C1 aC2 i (/kep/ ‘water’ < *kapi) but remained in *C1 aC2 a (/maR/ ‘hand’ < *maRa); *u > o in *C1 uC2 i (/koR/ ‘back to [a place]’ < *kurri and /kont/ ‘camping place’ < *kunti) but remained in *C1 uC2 u (/kul/ ‘louse’ < *kulu); and *i remained in *C1 iC2 u (/miR/ ‘spearthrower’ < *mirru). In the above (Ntra’ngith, Wik Muminh, Yir-Yoront) note the recurrence of *l (syllable-closing: final, or when first in a consonant cluster) conditioning preceding *a to [e]. In the Wik languages Wik Me’nh (K. Hale 1976e: 53) and Wik-Ngathan (Sutton 1995b: 107), *u preceding *lp lowered and fronted to [o¨], as in (respectively) /tho¨yp/ and /tho¨lp/ ‘stomach, belly’ < *culpi.28 In a couple of non-PNy families, GW and Ngarinyin– Worrorra, the historical depth of five vowel-systems, with particular respect to the mid vowels /e/ and /o/, is currently problematic. For pGW, the contrastive mid-vowel series /e, o/ has been reconstructed to the pGW stage (Harvey 2003c: 221). Harvey does not probe into pre-GW developments, but there are data suggesting that a development from a three-vowel system like that of pPNy took place and could be modelled as presented above for some CYP languages. For starters, consider /o/ originating as suggested by some Dalabon data: (19) /norr/ ‘you’, /nom/ ‘to smell’, /mo-/ ‘bone’, pPNy respectively *ñurra, *ñu:ma-, *muku and /e/, as suggested by some pGunw data (20) *ceñ, Dharawal (PNy, NSW) /jañ/ ‘fish’, possibly originating as a *CiCi sequence. With these contrast pGW *tharr (‘thigh’), *thak(k)u (‘left hand’), respectively pPNy *carra and *caku, and pGW *thiw (‘liver’), pPNy *cipa (< *kipa) 28 Regarding the origins of front-rounded vowels from *i in V2 , see also Section 16.5.1 below.

171

barry alpher As for Ngarinyan and Worrorra, McGregor and Rumsey (2009: 20–2) display the five-vowel system common to them on separate branches of the deepest split within the Worrorran family. They hesitate to commit to this as the system for Proto-Worrorran, however, citing correspondences of the sequences /ayi/ and /awu/ in the three-vowel language Unggumi (Worrorran branch). These have cognates with, respectively, /e/ and /o/ in other Worrorran languages that presumably have subsequently acquired loanwords with /ayi/ and /awu/ to produce synchronic contrasts with /e/ and /o/.

16.4.3 Separate development of structurally identical six-vowel systems Awu Alaya (Kuku Thaypan) and Ogunyjan, of the AlayaAthima subgroup (Alpher 2017), illustrate how closely related languages have evolved identical six-vowel systems with a stressed non-low central vowel from a three-vowel system via different pathways: Awu Alaya developed a six-vowel system (with loss of the length distinction and of vowel-quality distinctions at V1 ) as follows (Rigsby 1976: 72–6)29 : (21) *C1 aC2 a > /C2 e/ as in /me/ ‘man’ < *pama (22) *C1 aC2 a > /C2 a/ as in /tha/ ‘rotten’ < *kaca) (23) *C1 a:C2 a > /C2 e/ as in /me-/ ‘to spear’ < *ta:ma(24) *C1 a:C2 a > /aC2 a/ as in /anha/ ‘yamstick’ < *ka:ña) (25) *C1 uC2 a > /C2 o/ as in /no-/ ‘lie down’ < *wuna(26) *C1 uC2 a > /C2 wo/ as in /nggworr/ ‘north’ < *kungkarr (27) *C1 u:C2 a > /C2 o/ as in /wo/ ‘sand’ < *yu:ka (28) *C1 uC2 u > /C2 3/ as in /k3/ ‘bone, spine’ < *muku (29) *C1 uC2 u > /C2 u/ as in /ku/ ‘tree’ < *yuku (30) *C1 u:C2 u > /C2 u/ as in /ku/ ‘speech’ < *ku:ku (31) *C1 a:C2 i > / C2 i/ as in /ði-/ ‘cry’ < *pa:ciAdditionally, KThaypan *u and *i (and in some cases *a as well) in *V1 , before disappearing, were copied after *C2 (for details see Rigsby 1976: 74, and see note 35 in Section 16.5.3 below): /kw3n/ ‘grass < *pukan, ßye|ng ‘father’ < *pi:pa, and 29

There are different phonetic outcomes for different lexical items under conditioning not fully statable given the data at hand. The nature of C1 is apparently not a factor in the development of subsequent vowels in Kuku-Thaypan.

172

see also (26) above. The Mbabaram (Dixon 1991a: 356–9) sixvowel system developed along pathways closely similar to those followed by Kuku-Thaypan, with discrepancies in a few lexical items. Mbabaram and Kuku-Thaypan lands are not geographically distant from each other but they are not contiguous. Ogunyjan (Alaya-Athima), after loss of the length distinction but retention of V1 , contrasts five unstressed (in V1 : /i, e, a, o, u/) and (at least) six stressed vowels (/i, e, a, o, u, and mid-to-high-central 3/) in V2 . Pairing with /a/ as high to low and alternating with it in some contexts, /3/, originally < *a, is the optional final vowel (phonetic only) in a disyllable, now standing also where a vowel other than *a was lost, hence /abm/ ‘person’ < *pama, citation forms [abm, abm3, abmv] ‘person’, and /ek/ ‘tree’ < *yuku, citations [ek, ek3, ekv]. The likely original pronunciation of /3/ was as schwa ([ə]) from /a/ in unstressed syllables,30 which was later stressed when stressshift took place. Stress shifted to the second syllable only, in open disyllables ([abm3] ‘person’ < *pama) and in closed disyllables (/ebm3l/ ‘foot’ < *camal); in trisyllables *V2 was stressed in both *C1 V1 C2 V2 C3 and *C1 V1 C2 V2 C3 V3 , and /3/ (< schwa < *a) became distinct from /a/ when final V3 was lost, hence /angkáR/ ‘bluetongue lizard’ < *pangkarra vs. /angk3r/ ‘flesh’ < *pangkaR, along with for example the ergative-instrumental forms of ‘person’ /abmál/ < *pama-lu (contrast the V2 of /ebm3l/ ‘foot’ < *camal), of /ar/ ‘hand’ (/aráb/ < *mara-bV), and of /ek/ ‘tree, stick’ (/ekó-b/ < *yuku-bV). Other vowel-quality changes were as follows (note that some of the outcomes differ according to whether V2 was final or there was a C3 or C3 V). Because the initial laminal (*c, *ñ, *y: ‘L C1 ’) and non-laminal (‘-L C1 ’) consonants had differential effects on the quality of following vowels, it is clear that initial-dropping followed these changes. With initial laminal consonant: (32) *L C1 iC2 a > /L C1 iC2 a/ (no change), as in /inka-/ ‘to sit’ < *ñi:nka-31 (33) *L C1 uC2 u > /L C1 eC2 o/, as in /ek, ekob/ ‘tree’ (Abs, Erg) < *yuku (34) *L C1 uC2 a > /L C1 iC2 a/, as in /ibm, ibmab/ ‘fire’ (Abs, Erg) < *cuma, /irr/ ‘you (Pl)’ < *ñurra, /ipal/ ‘you (Du)’ < *ñupala (35) *L C1 aC2 u > /L C1 eC2 o/, as in /ekom/ ‘left-hand’ < *caku+mV 30 With regard to the distinction between schwa (reduction vowel, always the lowest possible stress level; I use ‘v’ in transcription) and the stressed mid-to-high central vowel (I use ‘3’), see Lass (1986) but also remarks below on Arrernte at Section 16.6.2 and its note 37. 31 Irregular nonPast inkyel; all other tense-forms in inka-.

sound change (36) *L C1 aC2 a > /L C1 eC2 a/, as in /edn/ ‘they (Pl)’ < *cana, /ebm3l/ ‘foot’ (< *camal)

16.4.4 Loss and reinstatement of vowel-length contrasts

With initial non-laminal consonant:

The vowel-length contrast has been lost in a great many PNy languages. A few instances are mentioned in passing above. However, some of the languages in which the length contrast was lost have regained it. Here are two dis*-L C1 uC2 i > /-L C1 oC2 e/, as in /onhthe-/ ‘to go parate examples. In Warlpiri, subsequent to vowel-length swimming’ < *mu:ñcineutralization, monosyllables (CVC) were phonetically long. *-L C1 uC2 u > /-L C1 oC2 3/, as in /olp3/ (Erg /olpoβ/ ‘old Then /-pa/ was introduced after a word-final consonant (as happened in a number of languages of the area, makman’) < *wulpu, /omb3/ ‘urine’ < *kumpu ing every word vowel-final), phonologizing the change, *-L C1 uC2 a > /-L C1 uC2 o/, as in /udn/ ‘excrement’ < hence *ngu:rr ‘throat’ > /ngu:rrpa/, contrasting in vowel *kuna, /ulon/ ‘possum’ < *kulan, /ukon/ ‘grass’ < length with pre-existing /ngurrpa/ ‘ignorant’ (K. Hale, via *pukan, /ubmon/ ‘thigh’ < *kuman D. Nash pers. comm.). But in Yir-Yoront, in which the Indigenous vowel-length distinction was lost fairly recently, *-L C1 aC2 i > /-L C1 eC2 i/, as in /etin/ ‘skin’ < *patin, English monosyllables in final /r/ + stop were heard and /eli-/ ‘to go’ < *kalientered the language with long vowels, hence /ya:t/ ‘yard’ and /ma:k/ ‘mark’, contrasting in vowel length with pre*-L C1 aC2 u > /-L C1 aC2 o, as in /aghoR/ ‘land’ < *Ra:kurr, existing /ngart/ ‘fish’ < *ngata, /mart/ ‘little’ < *manta, /ma /amoR/ ‘armpit’ < *nga:murr ʔ/ ‘bottom’ *mangka: a distinction introduced without a *-L C1 aC2 a > /-L C1 aC2 a/ (no change), as in /alk, alkanh/ sound-change. ‘spear’ (Abs, Instr), /angkaR/ ‘bluetongue lizard’ < *pangkarra, /angk3R/ ‘flesh’ < *pangkaR

(37) *-L C1 iC2 a > /-L C1 iC2 a/ (no change), as in /inh, inhal/ ‘animal, meat’ (Abs, Erg) < *miña (38) (39) (40)

(41) (42) (43)

As of (32) and (37) vs. (35) and (36), /e/ is distinct from /i/ in V1 . As of (38) vs. (41), /e/ is distinct from /i/ in V2 . As of (41) vs. (42), /e/ is distinct from /a/ in V1 . As of (37) vs. (38), /e/ is distinct from /a/ in V2 .32 And /a, o, u/ became distinct in V2 : *u in V1 C spread (see Section 16.5.3) its labialization to C2 when V2 was *3 (< *a), with *3 becoming /o/ (mid-open to mid-close rounded [U]) in the past tense-forms of verbs fitting the description, as with /uka-/ ‘to give’ (KThaypan /(a)kwo-/), Past /ukon/ ‘gave (Past)’ < *ukw 3n < *wukan. V2 in the past tense-forms merged with V2 in all tense-forms of /itoR-/ ‘to wait’ (with original /o/), Past and Participial /itoRən/. It is possible that the process was blocked by a laminal C2 , as in /uca-/ ‘return, go back’, Past /uc3n/, and /uñja-/ ‘eat, smoke’ < pPaman *kuñca-, Past /uñj3n/, participial /uñjan/33 . The vowel /o/ in /ukon/ ‘give’ (Past) is distinct from /u/ (high back rounded) as in /aRun/ ‘yesterday’ and from /a/ as in /ukan/ ‘give’ (Participial). 32 Some of the pathways to a five-vowel system followed by the neighbouring languages Uw-Oykangand and Uw-Olgol are like those followed by Ogunyjan, and some are not. See also remarks below on Kuku-Thaypan and Mbabaram. 33 Both /uñja-/ ‘to eat’ (other members of the subfamily have a reflex of *paca- for ‘to eat’) and /uca-/ ‘return’ are likely loanwords postdating labialization of V2 .

16.5 Vowel assimilation across consonants 16.5.1 Regressive

Vowel assimilation from the end of the word towards the beginning took place in Wik-Ngathan (Sutton 1995b: 44, 107), Wik-Me’nh, and Wik-Ngathrr (K. Hale 1976b: 53): *u in *V1 was fronted and rounded to /o¨/ when *V2 was *i, with subsequent loss of *V2 , as in (all three languages) /tho¨l/ ‘spearthrower’ and /mo¨:nhth(v)-/ ‘swim, immerse self ’ < *culi and *mu:ñci-, respectively. In Yir-Yoront, vowel assimilation was also regressive: following loss of V2 in original disyllables (*thul ‘spearthrower’ < *thul), a stressed vowel assimilated completely to a following *i or *u in new disyllables (< trisyllables), as in /thililh/ ‘spearthrower’ (Instr; Nominative /thol/) < *tholi-ñcu < *culi-ñcu), in /kililh/ ‘dew’ (Erg; Nom /kal/) < *kali-ngku (with merger of V1 *u and *a as /i/), and in /kuʔu/ ‘louse’ (Erg; Nom /kaʔ/) < *kaku-ngku. Contrast the vowels of /peʔerr/ ‘small’, /kaʔaR/ ‘moon’, /koʔoR/ ‘juice of beeswax’.

16.5.2 Progressive In Kuku-Yalanji (Hershberger 1964: 73–4; Hershberger and Hershberger 1986: 237; Patz 2002: 31–33, 45–8), vowel

173

barry alpher assimilation has been progressive, working towards the end of the word. The vowel /a/ of a case-ending such as Ergative assimilates to a preceding /u/ in a stem, hence /bama/ ‘person’, /kami/ ‘grandfather’, and /duñu/ ‘husband’, respectively, have Ergatives (in underlying {a}34 ) /bamangka/ < *bama-ngka and /kamingka/ < *kami-ngka, but /duñungku/ ‘husband (Erg)’ (with the final /u/ merging with that of pre-existing /kuludu/ ‘dove’). I surmise that monomorphemic forms like /kurrubal/ ‘freshwater cod’, /wulbuman/ ‘old woman’, and /jarruka/ ‘scrub fowl’ were not exceptions to this assimilation as a sound-change but rather were not in the language at the time: /jarruka/ ‘scrub fowl’ continues an earlier *carruka but likely is a recent loan from a neighbouring language; I assume the same for /kurrubal/; /wulbuman/ is a loan from English. Warlpiri vowel assimilation is also progressive, as in /watingki/ ‘man (Erg)’ < {wati+ngku/} and /malikirli/ ‘dog (Erg)’ < {maliki+rlu}. The result is judged phonologized (opaque) since one cannot tell from the sound alone whether V3 /i/ in /watingki/ is an assimilation product from V2 /i/ or is original as in /maliki/ ‘dog (Nom)’.

16.5.3 Progressive assimilation preceding initial-dropping Copying of V1 past C2 has taken place in a number of PNy languages, preceding initial-dropping (Section 16.6).35 Here are examples from Awngthim (Northern Paman; K. Hale 1976d: 15–16): Short V1 was copied after C2 if not identical to V2 and then lost, as in /ntr ya/ ‘arm’ < *pinta, /lwan/ ‘possum’ < *kulan, /mwa/ ‘fire’ < *cuma, /lwið/ ‘tooth’ < *mulirr, /thay/ ‘mother’s father’ < *ngaci, /mpaw/ ‘tooth’ < *ngampu. Where short V1 and V2 were identical, 34 The pPaman and pPNy form of this suffix was *-ngku, but the nearneighbouring languages Kuku-Thaypan (Rigsby) and Aghu-Tharrnggala (Jolly 1989) have /-ngga/, among other alternants. It seems to be a regional variant, and I treat it as a reconstructed form here. The ergative suffix cited here is one of the alternants of the ‘potent’ (Patz 2002: 46–7) or ‘living things’ ergative (Hershberger and Hershberger 1986: 286). 35 This change was first brought to general recognition by K. Hale, who in a published preliminary description (Hale 1964: 260–1), termed it metathesis. It was phonologized when V1 was subsequently lost, a change typically accompanying stress-conditioned initial (C1 ) dropping (see Section 16.6.2). Regarding the phonetics of this phenomenon, Fletcher and Butcher (2014: 107, citing Graetzer 2012), remark that ‘[t]here is also evidence of vowelto-vowel coarticulation, with close vowels exerting coarticulatory pressure on trans-consonantal open vowels.’ Note that in some of the languages reflecting this process, *a in V1 has also (like *i and *u) spread past C2 , as in Awngthim as shown above and in Awu Alaya /thaw/ ‘my’ < *ngacu, /yaw/ ‘I’ < *ngayu, and /maw|n/ ‘armpit’ < *nga:murr (Rigsby 1976: 75–6); however, generalization from the high (close) vowels (i, u) to the low (a) is a matter of dropping a feature specification—a simplification of a rule or process. Furthermore, spread of an initial vowel across a consonant followed by another high vowel, including an identical one, has occurred in some languages, as in Mbiywom (Hale 1976a: 22) /nggwu/ ‘knee’ < *pungku (contrast /ngku(lu)/ ‘to cry’ < *Rungka-, /ngko(ðRo)/ ‘black’ < *cungku).

174

V1 was simply lost, as in /ngku/ ‘knee’ < *pungku. A long non-identical V1 was copied after C2 like a short one but survived as schwa in V1 , as in /vnay/ ‘what’ < *nga:ni; a long V1 identical in quality to V2 survived as schwa: /vRi/ ‘nose’ < *ñi:rri. Other languages reflecting changes of this type include some other Northern Paman languages (Mpalitjanh, Luthigh, Yinwum, Ngkoth, Aritinngithigh, and Mbiywom; Hale 1976d: 8–12, 16–22), together with Kuku-Thaypan (Awu-Alaya) and Ogunyjan (Section 16.4.3); Kurrjar (Southwest Paman), Lamalama, Unbuygamu, Kalkatungu, Nganjaywana (Section 16.8), and Arrernte (Section 16.6.2, but see Koch 1997b and 2001 for a different analysis).

16.6 Initial-dropping Initial-dropping, in the Australian context, is the loss of word-initial consonants in an originally uniformly consonant-initial and first-syllable-stressed language. The account below follows Blevins (2001b: 481–92), adding to it and differing from it in some particulars. For an extensive (near-exhaustive) list and typology of initial-dropping languages, readers should consult the Blevins article. I augment and modify this list in some regards here.

16.6.1 Segmentally conditioned C-loss The loss of word-initial consonants ‘with inherently weak perceptual clues’ (Blevins 2001b: 482–4) but not of other consonants has taken place in a number of languages. Blevins (2001b: 483–4) lists (some of) these susceptible consonants as g, th, p, y (with suggested phonetic paths to loss), together with the names of languages that attest their loss. A language not on Blevins’s list is Yandjibara (Breen 1990b) of central-western Queensland (it is not contiguous with any other initial-dropping language), which has lost initial *k (to be classed with Blevins’s g) and no other consonant, with complete regularity and with stress retention on the first syllable. Compare Yandjibara with its initialretaining neighbour Dharawala (see Table 16.1). Initial *p was regularly lost, without stress-shift, in Nhanda (Blevins 2001a: 37), as with /arndi-/ ‘smell, sniff ’ < pPNy *parnti-. Bardi (Nyulnyulan family; Bowern 2007b) lost initial *w and *y, as in /a:mba/ ‘man’ < *wa:mba, /i:ndu/ ‘curlew’ < *wi:nduku, and /angan/ ‘close by’ < *yangan(a). Another class of less-than-total initial consonant loss (Blevins 2001a: 38, of Nhanda) is that of words most apt to occur utterance-initially, most notably ‘kinship terms (used of vocatives), “dog” (used to call a dog), as well as

sound change Table 16.1 Loss of initial *k in Yandjibara. gloss

Yandjibara

Dharawala

water

amu

kamu

blood

uma

kuma

eaglehawk

udhala

kudhala

catfish

akuRu

kakuRu

faeces

una

kuna

[but] fire

buRi

buRi

terms of endearment (“sweetheart”), and deictics … [which are] typically phrase-initial in their vocative/deictic functions, where they occur as single-word utterances’. To these I would add, for Nhanda (Blevins 2001a: 135, citing the O’Grady–Hale corpus), the ‘yes/no’ question particle aRi, usually clause initial and apparently continuing pPN *kaRi (‘not’). However, the content-question words ngana ‘who’, nhaa ‘what’, wandha ‘where’ retain their initial consonants but predominantly occur sentence-initially (126–7). Initial-dropping in a number of Paman languages of north-eastern Cape York Peninsula and the Princess Charlotte Bay area is apparently sporadic (see Alpher 1976). Verstraete (2018b) is currently working on a detailed study of dropping and lenition of initial consonants in this area.

16.6.2 Stress-conditioned C-loss (Blevins 2001b: 484–5) Prototypical cases of stress-conditioned initial consonant loss involve C1 loss preceded by stress-shift from the first syllable to the second, together with or followed by transfer of first-syllable vowel qualities to the second syllable, loss of the vowel of the first syllable or neutralization of the first-syllable length distinction, and in some cases effects on consonants after the first. Here are some illustrative examples. Harvey and Mailhammer (2017: 473, Table 1) propose reconstructions of Proto-Australian noun-class prefixes (five classes).36 These are relatively unstressed relative to what follows and as such are subject to initial-dropping and lenition (Harvey and Mailhammer 2017: 477). Details differ among the very many daughter languages. One example is 36 These (Harvey and Mailhammer 2017: 473, Table 2) are dependentmarking prefixes to adjectives and demonstratives within noun phrases.

the Class III (animal) prefix /a-/ in Wubuy (Nunggubuyu) (Harvey and Mailhammer 2017: 495), derived by initialdropping from *Ra-, as reconstructed for Warndarrang and Ngandi (Harvey and Mailhammer 2017: 479, citing Heath 1978a and 1980b). I suggest that it is possible that *Rais itself a lenition-product from *ta-, the Proto-Australian Class III prefix reconstructed by Harvey and Mailhammer (2017: Table 1). Eastern/Central Arrernte (Arandic),37,38 lost all initial consonants and most V1 Among those retaining V2 (changed to /a/) are words in which it is known to have been originally long: (44) /tne-/ ‘to stand’ < pPN *cana, with fortition of the nasal *n that is normal after a non-nasal *C1 and a short *V1 (45) /ane-/ ‘to sit’ < pPN *ñi:na- with lenis nasal conditioned by nasal C1 and long V1 , itself replaced by /a/ (46) /arne/ ‘stick’ < *ci:rni (< *ki:rni), with lenis nasal conditioned by long *V1 , itself replaced by /a/. The Western Arrernte cognate is /irne/ (47) /akwe/ ‘left-handed’ < pPN *caku and /mpwe/ ‘urine’ < pPN *kumpu, with labialization of *k between *u1 and *u2 Nasals in Arrernte are prestopped if V1 was short—compare (44) with (45 and 46)—but the conditioning was likely more complicated; once again see the work of Koch cited above. Linngithigh lenited stops and nasal-stop clusters after long *V1 and retained the quality of long *V1 , shortened. Compare here Uw-Oykangand (South-west Pama), with regular stress-shift and *C1 loss but with *V1 retained (with loss of the length distinction) and distinct in pronunciation (Sommer 1969; Hamilton n.d.; and my fieldwork). See Table 16.2. Apparently irregular initial dropping in the North Queensland language Kalkatungu is worth a brief and speculative treatment here. Despite initial dropping, some 37 I transcribe Arandic in the standard orthography, as used for example in Henderson and Dobson (1994). Orthographic a represents the single low vowel; ‘e’ represents the short non-low vowel. Called ‘schwa’, it is schwa-like in its variable realization depending on neighbouring consonants (akngwelye [akngúlyv] ‘dog’, atyelpe [acílpv] ‘native cat’), though its default position is a mid-to-high central as in atherre [ath3rrv] ‘two’. But it is unlike schwa (see note 30) in that, when it is the first vowel following a consonant (as in the above examples), it is stressed. Arandic historical phonology is rather an intricate business (for example, the presence or absence of initial /a/ is hard to predict, and there have been a great many loans from and to adjacent non-initial-dropping languages); readers are referred to the work of Koch (1997b; 2001) for more detailed explications. 38 The earliest recognition in the linguistic literature on Australian languages of the phenomenon of initial dropping was with regard to Arandic.

175

barry alpher Table 16.2 Among many languages that lost C1 , Linngithigh lost short V1 and Uw-Oykangand retained it.

mother’s younger brother

ala-

ala-

*ka:la

or contrast—for the free pronouns, and (ii) ergative casemarking for free pronouns vs. nominative–accusative for bound pronouns (as in Kalkatungu), in light of the observation that ergative marking, as well as free-pronoun status, can represent discourse prominence. The extent to which (i) is true of Kalkatungu is not clear to me from the source.39 In Kurrjar (with its sister-language Kuthant), a few pronouns are exceptions to phonologically (and subsequent analogically) based initial-dropping following on stress shift to V2 when followed by C3 (Black 1980: 206–7 and Edwards and Black 1998; see also Blevins 2001b: 485). Kurrtjar has

father’s older sibling

inha|ghay

iña‘auntie’

*pi:ña

(50) /nga:l/ ‘we (Du)’, /nga:n/ ‘we (Pl)’, and /dhan|aß/ ‘they (Pl)’ < *ngali, *ngana, and *cana

swim, bathe

oði|mthi–

uñci-

*mu:ñci-

Linngithigh

Uwprotoform Oykangand

mother’s mother, woman’s daughter’s child

mi-

abmi-

*kami

mother’s father

thi-

aji-

*ngaci

but, with long V1 :

Kalkatungu pronouns (Blake 1979a: 31–2) with etymologies dating to pPNy have survived with their initial consonants intact. I hypothesize that Kalkatungu underwent regular initial dropping of C1 a very long time ago, with a great many apparent exceptions accumulating subsequently, including some pronouns. Examples (developments at *C2 are regular) include (48) /nhutu/ ‘you (Pl)’, /puyu/ ‘they (Du)’, and /thina/ ‘they (Pl)’ < *ñurra, *pula, and *cana The corresponding contemporary cross-referencing bound subject pronouns (Blake 1979a: 35–8), postposed to verbs (obligatorily with some tenses), are respectively (49) /-nhurr/ ‘you (Pl)’, /-(mu)yu/ ‘they (Du)’, and /-na/ ‘they (Pl)’ The last two (Blake 1979a: 136–45) result from truncation of the first syllable of the corresponding full forms; I am surmising that they were at the relevant stage of the language’s history identical to the full- (proto-) form pronouns though functioning as clitics. Contrast the regularly initial-dropped /mpaya/ ‘you (Du)’ < *ñumpala, clitic form /-na/. I suggest that the non-initial-dropped forms are reinstated as full pronouns from cliticized forms, which were not word-initial. Relevant phenomena (cross-linguistically, here abstracted from Mushin and Simpson (2008: 588–90)) can include (i) the function of bound pronouns, in languages that have both bound and free, as the maintenance of continuing reference, vs. discourse prominence—focus

176

Contrast the initial-dropped forms below: (51) /a:nt/ ‘you (Sg)’ < *ñuntu and /œ:rr/ ‘you (Pl)’ < *ñurra However, in the absence of clitic pronouns in Kurrjar as recorded in the 20th century, there is nothing upon which to base even a speculative scenario such as advanced for Kalkatungu above.

16.7 Alterations of consonant clusters Clusters of two consonants (one or more of nasal+stop, liquid+stop, glide+stop, and liquid+nasal) occur in all of the Indigenous languages of the continent. Clusters of as many as three40 consonants (liquid+nasal+stop) are attested and in many cases are historically traceable. In Warluwarra and Bularnu (PNy, Ngarna subfamily) and in Yir-Yoront–Yirrk-Thangalkl (Paman), a nasal preceding a homorganic stop was lost: Warluwarra /guca-/ ‘to drink’ < *kuñca, Yirrk-Thangalkl and Yir-Yoront (YM and YY) /puth/ ‘arm’ < *puñca, /yirrp/ ‘rain, cloud’ < *yirrmpa, and, respectively, /yakáR/ and /yaʔaR/ ‘shin’ < *yangkaRa, /walk/ and /walʔ/ ‘belly’ < *walngka. In YM–YY, a heterorganic nasal remained, as in YM /kunk/, YY /kunʔ/ ‘alive’ < *kunka. 39 An instance of the creation of a free pronoun from a bound form in modern Irish (with an inflectional verbal ending as the bound form) is reported in Bybee et al. (1994: 13–14), whose concern is with change in the opposite direction (grammaticization) and who sees change in this direction as an extreme rarity. An instance in which the change bound-to-free has been postulated as in-process is Warumngu (Mushin and Simpson 2008: 585–7). For a syntactic theoretization (movement of a pronoun to a position higher in the tree), see Van Gelderen (2004: 62–9). 40 Four, if glides in Flinders Island language (Sutton 1975)—lmbw, lmby, lnggw, lnggy, rrnggy—are counted.

sound change The Yolngu languages, by contrast, lost (likely at the pre-Yolngu stage) the homorganic stop in these clusters, as in Djambarrpuyngu and Gupapuyngu /yangaRa/ ‘lower leg’ from *yangkaRa, with the heterorganic stop remaining, as in Gupapuyngu /gunja|lk/ ‘pandanus’ < *kunca+, and the homorganic clusters re-entering via borrowing, as in /wanggañ/ ‘one’, /guñjurlu/ ‘spotted gecko lizard’. Like Yolngu in this regard was the Coastal dialect of the Sydney language (Nash 2010). Examples of correspondences (Coastal to Inland dialects; transcription is reconstituted from old sources): C gama-, I gamba- ‘to shout, call’; C bangu, I banggu; ‘flying squirrel’; C gunya, I gunydya ‘house or hut’. Attestation of heterorganic clusters and their history is difficult to interpret (Nash 2010: 167–8, Section 10.2.5, and Table 10.6). Tri-clusters are attested as such in many but not all of the languages of CYP; in PNy languages elsewhere at least one of the three was lost; attested are *lmp, *lnt, *lñc, *lngk, *rrmp, and *rrngk:

(55) *lngk remains as /lngk/ in Kuku-Yalanji /walngka/ ‘swamp’ < *walngka, and elsewhere in CYP, but losing the nasal in PNy languages to the west and south, as in Atnyamathanha (Thura-Yura subfamily) /warlga/ ‘everted kangaroo pouch bag’, Panyjima and other Ngayarda languages /warlka/ ‘womb’, and likely Gupapuyngu and other Yolngu languages /wa:lk/ ‘umbilical cord, placenta, afterbirth’. Like *lngk is *rrngk, in putative pPNy *wi(:)rrngka-, continuing with /rrngk/ in Jirrbal (respect register) /wirrngga/ ‘scrape, scratch’ and in other CYP languages but losing the nasal in Arrernte /irrke/44 ‘scratch self ’ and in Djapu (and other Yolngu) /wirrk-/ combining with the inflecting verb /–dhu/ in /wirrkthu-/ ‘to scratch’.

In Yir-Yoront, the sequences *lkvrr and *nkvrr resolved to /lt/ and /nt, respectively, but survive in Yirrk-Thangalkl /tolkvrr/ ‘light in weight’. Yir-Yoront *lkvrr lost the schwa and conflated into the close-transition sequence *lkrr, which became /lt/ as in /lolt/ ‘light in weight’, merging (52) *lmp > as in *kalmpa remains as /lmp/ in Kokwith pre-existing /lt/ as in /yolt/ ‘empty-handed’ < pPNy Kapér /kalmpáw/ ‘first, already’ and Flinders Island *yurltu and with /lt/ from *R.t as in /lalt/ ‘emerge, go up’ < /elmba|l/ ‘quickly’, but > /lp/ in YYoront –YMel *TaRti—compare the Past tense-form /laRll/. That this hapand /mp/ in other Paman languages (Wik-Mungknh pened is confirmed by the Locative case-form /tholt/ of /kamp/ ‘fast’) and elsewhere in Australia (Yandruwandha /kampa|rri/ ‘at the front, first’).; *rrmp as /tholʔorr/ ‘distant, long’ (see note 50 to Section 16.8 rein Kok-Kaper /yirrmp/’rain, cloud’, YYoront /yirrp/. garding Yir-Yoront and see Section 16.11.2 regarding the disparate origin of phonetically somewhat similar flapped (53) *lnt as in *palnta remains as /lnt/ in Kuuk-Ya’u laterals in Iwaidja). As for *nkvrr, it survives in /thont/ /palnta/ ‘upper arm’ and Yir-Yoront (with regular ‘together as a group’, attested only in /thontwen/ ‘join toloss of the *n) /palt/ ‘shoulder, where it joins the gether as a group’ (with /wen/ ‘become’) and continuing neck’ but has lost the *l in Kuku-Thaypan /nda/ *thonkrr (via the usual syllable and consonant reductions ‘shoulder’. This etymon is the only *lnt one enin verbs, which are unstressed relative to the preceding countered and apparently does not continue in element) /ñc/ elsewhere in CYP (Guugu (South-west Paman) and Kuuk-Yak (an isolate within PaYimidhirr /ngañca/ ‘initiation ceremony’, Kok-Kaper man), the apical *n and *t of *nc and *tc have assimilated /ngañc (yok-)/ ‘poisonous tree [species]’, Kukuthe laminality of *c to produce [nhc], [thc] (and [dhj]) as in Thaypan /añc/ ‘taboo, a secret/sacred thing’) and [wanhci] ‘sore’ < pPNy *wanci and [thathce] ‘goanna’ < (ulelsewhere in Australia (Yindjibarndi /ngañca|li/ timately) *tharrci.45 On structuralist grounds (clusters [np, 43 ‘prohibited food’). nt, nk] occur and [nhp, nht, nhk] do not), [nhc] in these three languages is best analysed as /nc/, and (as clusters [thp, 41 The term poison in CYP can mean ‘taboo, ritually avoided’ (of places, tht, thk] do not occur but [tp, db] do) [thc] as /tc/, giving people, food). 42 Also in Kuku-Yalanji, in which however the etymon *ngalñca does not continue. 43 With the loss of *l in *lñc, the previously allophonic relationship of [(nh)th] and [(ñ)c] becomes phonologized: compare, Guugu-Yimidhirr /wanhdha/ and Yindjiparnti /wanhtha/ ‘where’, both < *wañca, with, respectively, GYim /ngañja/ and /ngañcali/ as detailed above. See Alpher (2004a: 112–14).

44 Harold Koch (pers. comm.) connects this to Warlpiri /pirrka-/ ‘to scrape’ and not to *wi:rrngka. 45 For Kugu Nganhcara see Smith and Johnson (2000: 378–81); it is to them that this phonetic observation is to be credited. I have directly observed (2005 fieldwork) [nhc] in KThaayorre and KYak and both [nc] and [tc] in KMumminh.

177

barry alpher /wanci/ and /thatci/. No phonologization has taken place; and *nc and *tc survive as [nc] and [tc] in other languages of this dialect cluster, for example KuguMumminh, as in [wanci] ‘sickness’ < *wanci and [thatci] ‘goanna’ < *thatci. At the stage of [nhc] there has been no sound change, but this cluster holds explanatory value for subsequent sound changes: it is clear that the *nc > [nhc] assimilation also took place but was then phonologized46 (i) in Yir-Yoront– Yirrk-Thangalkl, where the assimilation was later completed forward to give /nhth/, filling a void left by the loss of nasals before homorganic stops, and (ii) in Wik-Mungknh, Wik-Me’nh, and Wik-Muminh, where the assimilation was later completed backwards to give /ñc/, hence merging *nc with *ñc. A cluster reduction took place in Ngaanyatjarra (Warburton Ranges, Western Desert dialect cluster, Pama-Nyungan), the subphonemic loss of the [rr] in /rrc/. It is of interest here because it has taken place during a period (1964–2003) covered by written linguistic records. On structuralist grounds (because /rr/ occurs before the other stops, and /l/ occurs before all of the stops), the cluster is still /rrc/, and the effect of the heterorganic [rr] on the following /c/, conditioning its pronunciation to lamino-alveopalatal [c] rather than lamino-dental [th] (Douglas 1964: 13–14) persists, hence /purrcu/ [pucu] ‘itch’ vs. /pucurrpa/ [puthurrpa] ‘numb’ (Glass and Hackett 2003: 3–4).47 If the analysis of (now silent) /rr/ as a persisting phoneme is not correct, then the result of its loss must be seen as the phonologization of the /th/ vs. /c/ contrast.

16.8 Loss of unstressed vowels, including rhythmic reduction Final vowels in di- and trisyllables were lost in Nyungar, as in /maʔ/ ‘hand’ < *maRa and /mical/ ‘rain’ < *micaRa, and in Wik-Ngathan, Wik-Mungknh, and Wik Me’nh, as in Wik Ngathan /maʔ/ ‘hand’ < *maRa. In these Wik languages, everything from C3 on was lost unless C3 was *n: Wik Ngathan /thalp/ ‘tongue’ < *calparr, /kenh/ ‘freshwater crocodile’ < *kanharr, but /nhepvn/ ‘egg’ < *nhapin. Final V2 was also lost in the South-west Pama languages,48 46 These developments are not instances of phonologization in the classic sense (loss of conditioning environment) discussed by Kiparsky (2015: Section 1). As well, [nh] here does not seem to be a ‘quasi-phoneme’, in Kiparsky’s system, since this term is applied to a conspicuous product of conditioning (as in Old High German umlaut) which persists after the conditioning factor is lost. 47 This failure of following /u/ to condition preceding /c/ to [th] here is an illustration of the ‘non-phonologization problem’ (Kiparsky 2015: Section 1 and Section 3). 48 With the exception of Uw-Oykangand’s sister-dialect Uw-Olgol, in which the quality of final V2 is preserved in free alternants: [abm] ~ [abmá] ‘person’, [ug] ~ [ugú] ‘tree’, [alh] ~ [alhú] ‘fire’.

178

as in Uw-Oykangand /abm/ ‘person’ < *pama, /uk/ ‘tree’ < *yuku, /alh/ ‘fire’ < pPNy *yaʎu, and in Ogunyjan (see Section 16.4.3). Some other examples, briefly, are as follows: (56) In Jaminjung, Ngaliwurru, and Nungali (Mirndi family, Yirram branch; Harvey 2008b: 137), V3 was lost after V2 followed by a stop, as in /warrij/ ‘freshwater crocodile’ < *warrija; cf. Wambaya /warriji/.49 (57) In Mbabaram (Dixon 1991b) V3 (but not V2 ) was lost: /nggaR ‘whole leg’ < *yangkaRa (but /be/ ‘liver’ < *cipa < *kipa). (58) Loss of V3 in Yir-Yoront and Yirrk-Thangalkl (rhythmic) and in Ogunyjan left as a trace the nonreduction of V2 as in pPNy *pangkarra, continuing as YYoront50 /paʔarr/ (not /paʔvrr/) and Ogunyjan /angkaR/ (not /angk3R/; see Section 16.4.3) ‘frill lizard’. In Guugu-Yimithirr (Haviland 1979b) and Yidiny,51,52 V3 was lost, entailing compensatory lengthening of V2 , as in /ganha:rr/ ‘crocodile’ (both languages) < *kañarra. A historical account of Yidiny truncation is problematic from the standpoint of regularist principles, since it appears to have involved a process, stated as ‘Rule 2’ in Dixon’s synchronic analysis (1977a: 15 and 1977b: 58 [the rule] and 71 [the claim that it was a diachronic event as well as a synchronic rule]), whereby at the end of a word with an odd number of syllables the entire final syllable (*C3 V3 ) is lost up to, but not including, a preceding consonant that is itself preceded by a vowel, and which remains as the coda of the preceding syllable, hence for example in /gaña:rr/53 ‘crocodile’ {gañarrA}54 ið, also a laminalization, as in /lwið/ ‘tooth’ < *pPaman *mulirr. However, it is also possible to interpret the Victorian data as having the directionality tj/th > rt > R, i.e. an apicalization, in which case the CYP comparison would be with the names of two adjacent sisterdialects of Uradhi from northernmost CYP (Crowley 1983: 60 Evans believes that this change took place in pre-Pama-Nyungan and is phylogenetic-group-defining. I believe on the basis in part of Harvey (2003c: 250, 257) that the relevant initial consonants contrasted as *t vs. *c in pPNy, as evidenced by pPNy *caku ‘left’, pGW *cakku ‘left hand’ vs. pPNy t/Ra:ku|rr ‘place (etc.)’, pGW *Rak and *Rerr (both ‘camp’) and that laminalization took place after the split of PNy into subgroups and individual languages. This is however not the place to argue these claims; the point is that laminalization did take place, at whatever phylogenetic level(s).

179

barry alpher 311) Yaðaykenu (with /ð/ < *th or *c as per Section 16.2.2.1) and Yaraytjana (with /rr/ < *ð, where (‘r’ = /rr/). Merger of *ki with *ci took place in most branches of Pama-Nyungan,61,62 with *ci surfacing as /thi/ in languages where [th] was the normal reflex of *c in this context.63 Examples:

these cases, and hence there has been no sound change. Related to this, however, is the question of the phonological interpretation of the Jingulu (Mirndi family) sound(s) [ky ] and/or [y k]. Pensalfini (2003: 26–33) considers these to be realization(s) of one or more of the clusters /kj/, /jk/, /ky/, and /yk/ and postulates a single unit phoneme, /ky /. Under Pensalfini’s analysis,67 there has been no sound-change; un(61) Proto-Eastern pNy *ki(:)npal > Gamilaraay and der Chadwick’s there has, with /ky / distinct from both /c/ Yuwaalaray (NSW) /gi:nbal/ ‘scales (of fish)’, YYoront and /k/. *thinpl ‘fin, including tailfin, of fish’, and Umpithamu Proto-pNy *yk continues unchanged in some languages, /yinpal/ ‘tail of dugong’. for example in Kuku-Yalanji, with /buyku/ ‘paperbark’, (62) pPNy *ki:rni > Yuwaalaray /giniy/ ‘tree, stick’, Diyari /ngayku/ ‘my’ and in Nyawaygi with /ngaygu/ ‘my’. But /kini/ ‘penis’ (GNOG), YYoront /then/ ‘penis’, KKaper *yk > /c, j/ in numerous other languages, as in Bidyara (Breen 1973) /buju/ ‘bag’ < *puyku; in Bidyara the normal /then/ ‘tail’, WArrernte /irn(e)/ ‘tree, stick’. reflex of pPNy *c is /dh/ (/midhard/ ‘frost, cold weather’ < pPNy *micaRa and /widhi|ñ/ ‘a sore’ < *wi:ci), contrast(63) Eastern *ki:n > Sydney /jin/ ‘woman’, Yugambeh ing with /j/ < *yk. Note also Margany–Gunya (Breen 1981a) /kin/ ‘young woman, girl’.64 /gaca/ vs. Gugu-Badhun (Sutton 1973) /gayga/, both ‘rotten’. See Alpher et al. (2008: 16–17 and notes 5 and 6) for (64) pPNy *kipa > Dharumbal /giba/, Girramay more examples. and Warrgamay /giba/, GYimithirr /dhiba/, An example of /ci/ < *ki from Warriyangga (PNy, ManKYalanji /jiba/, YYoront /thip/ ‘liver’.65,66 tharda subgroup; Austin 1992b) is probably very recent and (65) KYalanji /warrki/ ‘can’t come close because involves what is likely a loanword: /maliji/ ‘man’, putative of relationship’ < pPNy *warrki (continuing source Western Desert (as in Warburton Ranges) /maliki/ elsewhere via *warrci with reflexes meaning ‘stranger’; a connection with Warlpiri /maliki/ ‘dog’ is se‘bad’, etc.) mantically justifiable. From Proto-Iwaidjan (Evans 1997b: 245, 248), initial *ki > In a number of languages, the phoneme /k/ is fronted when /(y)i/ in Iwaidja, as in /imacak/ ‘bird’s wing’ < pIwaidjan *kipreceding /i/, bringing the articulation of /ki/ very close to macak, and similarly in Mawng, as in /yili/ ‘fear’ (compare (but nonetheless distinct from) that of /ci/, making it difMayali /kele/ ‘fear’). ficult at times for non-native-speakers to decide whether they are hearing /c/ or /k/; the same holds for /ngi/ vs. /ñi/. There has been no merger between /k/ and /c/ in 61 In hypothesizing the change *ki > *ci (Alpher 2004a: 109), I suggested that the scarcity of etyma in *ki (vs. *ka and *ku) supported the hypothesis. However, a list count of words in initial /g/ (/k/) in Nyawaygi (Dixon 1983: 504–11), with 72 /gu(:)/, 67 /ga(:)/, and 16 /gi(:)/—Nyawaygi, retains *ki (and *yk)—nonetheless exhibits a paucity of initial /ki/. The frequency of /i/ at V2 with initial /b/ and /j/ is likewise the lowest among the three vowels. The argument from list scarcity apparently fails. 62 Harold Koch (pers. comm.) holds that I have the directionality of this change reversed. 63 Erich Round (pers. comm.) raises the interesting question of why there are no synchronic alternations k ~ c in any language anywhere in Australia. For PNy languages, attestation might take the form of a suffixed /kV/ where V is /i/ alternating with /a/ or /u/. Warlpiri presents this situation (but no /c/), with Dative (underlying {-ku}) case-forms: /ngapaku/ ‘for water’, /warluku/ ‘for firewood’, and (with progressive assimilation, but no /c/) /watiki/ ‘for a man’. The change *ki > /ci/, in languages that attest it, likely took place a very long time ago, and there are a great many counterexamples requiring further analysis, both diachronic and synchronic. 64 *kiyan is also possible here. 65 There are problems with the identification of loans among languages in the Qld. Rainforest area, together with the inconvenient fact of pGW (Arnhem Land) *ciw (‘liver’). 66 These etyma together with those from *yk cited below in this section are consistent with the suggestion that there was only *c in pPNy, and not two distinct laminals (*th and *c).

180

16.10 Consonant fortitions not creating new manner series 16.10.1 Fortition of initial C In Yidiny (Y; Dixon 1991b) and Kuku Yalanji (KY; Hershberger and Hershberger 1986), initial *yu > /ju/, as in Y /juRu/ ‘elbow’ < *yuRu, Y /juRi/ ‘sharp’ < *yuRi (evidently a variant of *yiRi), KY /juka|R/ ‘sand’ < *yu:ka, and Y and KY /juki/ ‘tree’ < *yuki (evidently a variant of *yuku), with /j/ from this source merging with the /j/ in Y /jaku|y/ and KY /jaku|ji/ ‘left hand’ < pPNy *caku. There are also instances of initial *ya > *ca in KY, as in /jarra/ ‘more, farther’, < pPNy *yarra, but it is unclear whether this shift is regular. The shift initial *y > /j/ has also occurred in Yugambeh (Sharpe 1998), though here there is apparently no regularity: 67 Included in Pensalfini (2003: Table 2.1, permissible consonant clusters, p.33) are /jk/ and /kj/, but not /ky/ and /yk/.

sound change pPNy *y > j /#_ as in /jalu/ ‘fire, smoke’ < *yaʎu; problematic forms include /yalañ/ ‘tongue’ and perhaps also /jalañ/ ‘throat’ < *calañ and /yan(a)-/ ‘go, walk’ < *yana. So also with Walmajarri (ASEDA 0167) /jangkarla/ ‘lower leg between shin and ankle’ and Karajarri (ASEDA 0069) /jankala/ ‘calf of leg’, whose direct descent from pPNy *yangkaRa is questionable in view of forms like Walmajarri /yapa/ ‘baby, child’ < *yapa and Karajarri /ya|ña/ ‘go, walk’ < *ya-, and the geographic proximity of the Nyulnyulan languages, whose cognate forms are reconstructed as pNyulnyulan *jangkala (‘calf ’).

16.10.2 Fortition of intervocalic C In Warluwarra, *rr > t in /ngutu/ ‘nose’ < pPNy *ngurru (Yanyuwa /ngurru/). It seems likely that (if general at all) this took place only in the context *u_u, in view of /kuti/ ‘swan’ < *kuti (widespread) and other examples with no clear etymological origin, like /kuturla/ ‘the other side’, /kutu/ ‘shut’. With contemporary intervocalic /rr/ are /ngurra/ ‘string’ and /kurrupa/ ‘paperbark, tea tree’, presumed to have entered the language after fortition had taken place. In Uradhi (all dialects) intervocalic *rr and *R > /t/ after short *V1 as in /yutu/ ‘elbow’ < pPaman *yurru (K. Hale 1976b: 43; Crowley 1983: 418) and /mata/ ‘hand’ < pPNy *maRa, merging with pre-existing /t/ in /utaɣa/ ‘dog’ < *kutaka, /ataßa/ ‘river’ < (per sources) *waRapa, and /putu/ ‘boat’ (< English) and now in contrast with the /rr/ of /murruku/ ‘horse’.68 Kalkatungu (Blake 1979a: 134) intervocalic *rr > /t/ as in /nhutu/ ‘you (Sg)’ < pPNy *ñurra, /ati/ ‘meat’ (compare Yalarnnga /warri/), /iti-/ ‘man (Erg stem)’ (compare Yalarnnga /irri/) ‘man’. Contrast /rr/ in /curru/ ‘reed’ and note merger with /t/ as in /utingarr/ ‘emu’ and /titil ~ titirri/ ‘centipede’.

16.10.3 Fortition of a C adjacent to a C Pakanh *rr > /t/ before *c in /thatpi/ ‘goanna’ < Local *tharrpi. A similar change (cited as synchronicmorphophonemic; Teo 2007: 27, 31) takes place in Maung, with underlying {rrk} becoming /d/ before /j/, as in /adjayik/ ‘our (Incl) bones’, with /arrki/ ‘our’ and /jayik/ ‘bones’; phonologization (does /dj/ occur independently of this process?) is difficult to assess here. Similar phenomena take place in Iwaidja (Teo 2007: 22), but here involving the never-segmentalized morphoneme {K} (see below in this 68 As given in Crowley (1983: 409, 418), although it looks as if it were a loanword for ‘motorcar’.

section),69 in Mangarrayi (Merlan 1989: 209), and in Larakia (Capell 1984: 60). Uradhi preconsonantal *l following a short vowel > /t/, as in /lutpi/ ‘belly’ < *culpi (Crowley 1983: 332; K. Hale 1976b: 43–4). *V1 length later neutralized but [l] and [t] do not contrast (Crowley 1983: 321) in this position ([tp] vs. [lb, lg]; Hale’s notation); apparent exceptions ulgin ‘snot’ vs. ulkin ‘saliva’ (Crowley 1983: 411) are from different dialects. Whether preconsonantal [t] < *l in [tp] has phonologized in Pakanh (this section) needs further research. In Burduna (Mantharda; Austin 1981c: 302–3) and Tharrgari (Kanyara; Austin 1981c: 309), nasals preceding nonhomorganic stops have become stops, as in Burduna /ngartka/ ‘beard’ < *ngarnka and Tharrgari /watka/ ‘alive’ < *wanka. In Tharrgari, this fortition occurred also before homorganic stops, creating contrasting fortis and lenis stop series (Section 16.2.1.6). Like Tharrgari in this regard was Kamu (Eastern Daly; Harvey 2003a: 175–7), in which, Harvey claims, the detransivizing suffix /cci/ corresponds (via fortition) to the pGW reciprocal suffix *ñci and is likely derived from a protoform *ñci. From this pGW protoform Harvey derives, via fortition in Ngalakgan and Rembarrnga, the detransitivizing suffix /cci/70 , which (Merlan 1983: 115) is the Ngalakgan reflexive-reciprocal allomorph with all auxiliaries and all simple roots except /pu/ ‘hit’ (with /-yci /) and /wu/ ‘give’ (with /-ycci/); these latter make problems (correspondence irregularities) for Harvey’s historical scenario. Nhanda liquids preceding stops underwent fortition to stops (Blevins 2001a: 39–40), as in /atpi/‘to tie’ < *pPNy *karrpi (see Section 16.2.1.6 above for more detail). In Iwaidja (Evans 2009a: 163), assimilatory fortition of a nasal to a stop (here ‘S’) and assimilation to its manner takes/has taken place in the contexts /V_S/ and /S_V/, as in /wunakpiñ/ ‘became an owner’ (with /piñ/ ‘became (Past)’ < *miñ). It is not clear whether this change has become phonologized. A similar change in Mawng whereby {Kl, Km, Kng} > /d, b, k/, where {K} ‘is an underlying morphophoneme that is never realized as a segment (but hardens the following seqment) and continues, generalized, the old Miscellaneous gender prefix *aK’ (Teo 2007: 10, 11, 27, and notes 9 and 10) has become phonologized because of loss of conditioning environment; examples include /adijbularr/

69 The changes presented as illustrative of fortitions here and below are listed in the source (Teo 2007: Tables 11 and 12) in two morphophonemic charts, each a matrix of upwards of 70 cells, of prefix-final and prepound(verb-) initial consonants. 70 Note that Harvey’s analysis of the fortis/lenis contrast in stop systems of north central Australia is as geminate (pp, cc, etc.) vs. plain (p, c, etc.), whereas Merlan in her work treats these as singleton fortis (p, c) vs. lenis (b, j). In order to reduce confusion here, I have restated Merlan’s transcriptions according to Harvey’s convention, under which the change *ñc > /cc/ can be interpreted as a fortition like that in Tharrgari.

181

barry alpher ‘young plants’ < {aK-lijbularr}, /abawurr/ ‘armlike projection of yam’ < {aK-mawurr}, and /akijalk/ ‘yam, grain, fruit’ < {aK-ngijalk}. Other languages exhibiting this fortition include Wardaman (Merlan 1994: 36), Amurdak (Handelsman 1991: 42), and Larakia (Capell 1984: 60); Gurr-goni prefixes have nasals that become stops when preceding stops (Green 1995).

16.10.4 Fortition of a word-final C In Bininy Gun-Wok (Harvey 2003c: 222, 260; reconstructions are pGW), final *rr > /t/, as in /-pit/ ‘hand’ < *-pirr, /-tat/ ‘thigh’ < *-tharr, and /yat/ ‘yabby’ < *yarr. Harvey (2003c: 222) lists exceptions and cites an earlier and superseded reconstruction, *carr, for ‘thigh’.

16.11 Creation of individual sounds not previously in the inventory 16.11.1 Apical stops with trilled release Some CYP and Karnic languages attest apical stops with a trilled rhotic [rr] release: [tr , dr ]. In Yinwum (NPaman; K. Hale 1976a: 10–12), *t > /tr / in *an__a, as in /ntr a-/ ‘put, leave’ < pPaman *wanta-; cf. /nti/ ‘spear with multi-pronged spear’ < pPaman *yinta-. In some of the other NPaman languages the same thing happened, but independently of the flanking vocalic context, as in Linngithigh, Alngith, Awngthim, Ntra’ngith, and Ngkoth /ntr a/ ‘put’ < *wanta-, and (from *pinta) Linngithigh-Alngith and Ntra’ngith /ntr æ/ and Awngthim /ntr ya/ ‘arm’. Like these is Anguthimri (Crowley 1981: 157–8), with [ndr ] /dr / < *nt when *C1 was labial, as in [ndr ya] (/dr ya/) ‘arm’ < *pinta and [ndr a] (/dr a/) ‘to leave it’ < *wanta-. In Ngkoth, contrast the /t/ in /twa/ < *kuta; contrast /t/ in Linngithigh, which appears in stems confined to Northern Paman. In Kok-Kapér (South-west Paman), initial [tr ] alternates freely with [rr] (trilled) in a number of words (such as [tr ikipı´rr ~ rrikipı´rr] ‘little’, [tr vpén ~ rrvpén] ‘antbed’) and with [t] in some others (such as [tr úku- ~ túku- ~ rruku-] ‘enter’). After /n/ in a stressed first syllable both [tr ], and [t] occur, as in [kúntr ~ kúnt] ‘hit (Past)’, /tr antr / ‘cheeky’, but after /n/ in a stressed syllable after the first, only [t] occurs, hence /pathént/ ‘ate’. I have not recorded sets of words that consistently distinguish among initial [t, tr, rr] and between postvocalic [nt, ntr] in a monosyllable; Paul Black finds the same (pers. comm., and Edwards and Black 1998: 2–3, which see for orthographic practice). Initial /l/ as in /lángkvrr/ ‘here’ is distinct from [t, tr , rr]. Intervocalic /t/, /rr/, and

182

/l/ are distinct, as in /watárr/ ‘good’, /parráy/ ‘emu’, and /kalál/ ‘went’. Like Kok-Kapér in this regard is Kayardild (Round 2009: 49–51), with the apical stop /t/ having an optional subphonemic trilled release when preceded by a sonorant (/n, l, R/) as in the sequence /untu/ [untu ~ untr u]. Most of the Karnic languages (Queensland, South Australia, and NSW) feature one or several trill-released apical consonants. The origins of /dr / and /rdr / in Yandruwandha (Breen 1974: 13–14; 2004b) are regular,71 as follows (see Austin 1988d: 228–34 for a detailed treatment): (66) Initial /rdr / from an apical stop (/rd/ in other Karnic languages such as Diyari, and from some apical non-nasal72 in various PNy languages), as in /rdr aka/ ‘pierce’ and /rdr ama/ ‘cut’ (respectively from widespread Eastern *taka- and from pPNy *ta:ma- ‘spear’/‘dig’/‘cut’). Contrast also initial /rd/ (orthographic d) as in /rdaka/ ‘make (as a boomerang)’. There is no contrasting initial /t/ or /d/. (67) Intervocalically, as in /kudr i/ ‘swan’