Phonological Explorations: Empirical, Theoretical and Diachronic Issues 9783110295160

324 57 3MB

English Pages x, 355 pages : [368] Year 2012

Report DMCA / Copyright


Polecaj historie

Phonological Explorations: Empirical, Theoretical and Diachronic Issues

Table of contents :
Notes on contributors
Allomorphy and the architecture of grammar
From prof to provo: some observations on Dutch clippings
Recursion in phonology?
The Grimm-Verner push chain and Contrast Preservation Theory
Segmental structure and vowel shifts
The distribution of vowels in English and trochaic proper government
A propos of the Dutch vowel system 21 years on, 22 years on
A minimal framework for vowel harmony
Greater than noise: frequency effects in Bantu height harmony
The phonological representation of the Limburgian tonal accents
Quantity or durational enhancement of tone: the case of Maastricht Limburgian high vowels
Using local constraint conjunction to discover constraints: the case of Mandarin Chinese
Implications of Harmonic Serialism for lexical tone association
A constraint-based explanation of the McGurk effect
Liquids in a case of unfolding early L1 Dutch: from null realizations through free variation through probabilistically bound variation to lexical contrast
The Tibetan numerals segmentation problem and how virtual learners solve it

Citation preview

Linguistische Arbeiten


Herausgegeben von Klaus von Heusinger, Gereon Müller, Ingo Plag, Beatrice Primus, Elisabeth Stark und Richard Wiese

Bert Botma, Roland Noske (Eds.)

Phonological Explorations Empirical, Theoretical and Diachronic Issues

De Gruyter

ISBN 978-3-11-029516-0 e-ISBN 978-3-11-029517-7 ISSN 0344-6727 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at © 2012 Walter de Gruyter GmbH & Co. KG, Berlin/Boston Gesamtherstellung: Hubert & Co. GmbH & Co. KG, Göttingen ∞ Gedruckt auf säurefreiem Papier Printed in Germany

Contents Notes on contributors................................................................................... VII Introduction......................................................................................................1 Geert Booij Allomorphy and the architecture of grammar ..................................................9 Camiel Hamans From prof to provo: some observations on Dutch clippings ..........................25 Irene Vogel Recursion in phonology? ...............................................................................41 Roland Noske The Grimm-Verner push chain and Contrast Preservation Theory................63 Janet Grijzenhout Segmental structure and vowel shifts.............................................................87 Krisztina Polgárdi The distribution of vowels in English and trochaic proper government ......111 Bert Botma & Marc van Oostendorp A propos of the Dutch vowel system 21 years on, 22 years on ...................135 Harry van der Hulst A minimal framework for vowel harmony ..................................................155 Diana Archangeli, Jeff Mielke & Douglas Pulleyblank Greater than noise: frequency effects in Bantu height harmony ..................191 Ben Hermans The phonological representation of the Limburgian tonal accents ..............223 Carlos Gussenhoven Quantity or durational enhancement of tone: the case of Maastricht Limburgian high vowels ..............................................................................241 Jeroen M. van de Weijer Using local constraint conjunction to discover constraints: the case of Mandarin Chinese ........................................................................................255

VI John J. McCarthy, Kevin Mullin & Brian W. Smith Implications of Harmonic Serialism for lexical tone association.................265 Paul Boersma A constraint-based explanation of the McGurk effect .................................299 Frans Hinskens Liquids in a case of unfolding early L1 Dutch: from null realizations through free variation through probabilistically bound variation to lexical contrast .............................................................................................313 Diana Apoussidou The Tibetan numerals segmentation problem and how virtual learners solve it..........................................................................................................333

Notes on contributors Diana Apoussidou Diana Apoussidou works as a ‘Referent’ at the Stiftung für Hochschulzulassung in Dortmund, Germany. She completed her PhD at the University of Amsterdam in 2006 and held postdocs at the University of Amsterdam, the University of Massachusetts Amherst, and the University of Utrecht. Her research focuses on computational approaches to phonology and issues of learnability. Diana Archangeli Diana Archangeli is a professor of Linguistics at the University of Arizona. She completed her PhD at Massachusetts Institute of Technology in 1984. Her main research interests include understanding language sound systems from the abstract phonological level to the concrete details of articulation, with special emphasis on vowel harmony. Paul Boersma Paul Boersma is professor of Phonetic Sciences at the University of Amsterdam. He received a PhD from the University of Amsterdam in 1998. His research focuses on modelling and simulating the acquisition, evolution and typology of the production and comprehension of phonology and phonetics. He is also the designer and main author of Praat, the world’s most used computer program for the analysis and manipulation of speech. Geert Booij Geert Booij is a professor of Linguistics at the University of Leiden. He took his PhD at the University of Amsterdam in 1977. He was previously professor of General Linguistics and dean of the Faculty of Arts at the Free University of Amsterdam. His research interests include phonology, morphology, and the architecture of grammar. Bert Botma Bert Botma is a lecturer of English Language and Linguistics at the University of Leiden and holds a postdoc at the same university. He completed his PhD at the University of Amsterdam in 2004. His main research interest is in segmental phonology. Janet Grijzenhout Janet Grijzenhout is professor of English Linguistics and director of the Baby Speech Laboratory (BSL) at the University of Konstanz. She obtained her


Notes on Contributers

PhD from Utrecht University in 1995 and her habilitation from the HeinrichHeine-Universität, Düsseldorf in 2001. Her main research interests include all areas of phonology, morphology, English linguistics, comparative linguistics, historical linguistics, first and second language acquisition, and infant speech perception and production. Carlos Gussenhoven Carlos Gussenhoven was professor of General and Experimental Phonology at the Radboud University Nijmegen and professor of Linguistics at Queen Mary, University of London until his retirement in 2011. He took his PhD at the University of Nijmegen in 1984. His research has been on the intonation and tone structure of languages, including British English, Standard Nigerian English, and standard and non-standard varieties of Dutch. Camiel Hamans Camiel Hamans studied Dutch Language and Literature at the University of Amsterdam. After teaching at the Universities of Leiden and PoznaĔ he left academic life to pursue a career in journalism and politics, which led to his appointment as Secretary General of the Dutch Social Democrats in the European Parliament. He has continued to publish on topics relating to the phonology-morphology interface. Ben Hermans Ben Hermans is a Senior Researcher at the Meertens Instituut of the Royal Netherlands Academy of Arts and Sciences. He obtained his PhD from the Free University of Amsterdam in 1994. He is interested in the synchronic and diachronic phonology of Limburgian dialects. Frans Hinskens Frans Hinskens is a Senior Researcher at the Meertens Instituut of the Royal Netherlands Academy of Arts and Sciences and professor of Language Variation and Language Change at VU University Amsterdam. He completed his PhD at the University of Nijmegen in 1993. Between 1998 and 2002 he was chair of the Department of Dutch at the University of Leipzig. Harry van der Hulst Harry van der Hulst is a professor of Linguistics at the University of Connecticut. He obtained his PhD from the University of Leiden in 1984. His main research interests are feature systems and segmental structure, syllable structure, word accent systems, vowel harmony and sign language phonology.

Notes on Contributers


John J. McCarthy John J. McCarthy is Distinguished University Professor, Vice-Provost for Graduate Education, and Dean of the Graduate School at the University of Massachusetts Amherst. He completed his PhD at Massachusetts Institute of Technology in 1979. His research interests lie in the areas of phonology, morphology, and their interface. Jeff Mielke Jeff Mielke is an associate professor of Linguistics at the University of Ottawa. He completed his PhD in 2004 at the Ohio State University and held a postdoc at the University of Arizona. He uses laboratory and computational techniques to study language sound patterns at the typological level and at the level of the individual. Kevin Mullin Kevin Mullin is a doctoral student in the Department of Linguistics at the University of Massachusetts Amherst. He does experimental research on speech perception and formal theoretical phonology in Serial Harmonic Grammar and Harmonic Serialism. Roland Noske Roland Noske is an associate professor at the Department of Foreign Languages and Cultures (Dutch section) at the University of Lille and a member of the Joint Research Group (CNRS) STL. He completed his PhD at the University of Tilburg in 1992 and his habilitation at Denis Diderot University (Paris 7) in 2001. His research interests include Dutch linguistics, French linguistics, syllable structure, typology, and historical linguistics. Marc van Oostendorp Marc van Oostendorp is a Senior Researcher at the Meertens Instituut of the Royal Netherlands Academy of Arts and Sciences and professor of Phonological Microvariation at the University of Leiden. He obtained his PhD from the University of Tilburg in 1995. His research interests include the phonology of varieties of Dutch, language policies, and the interface between phonology and morphosyntax. Krisztina Polgárdi Krisztina Polgárdi is a Senior Research Fellow at the Research Institute for Linguistics at the Hungarian Academy of Sciences. She took her PhD at the University of Leiden in 1998. Her research focuses on syllable structure issues in Hungarian, Turkish, Dutch and English.


Notes on Contributers

Douglas Pulleyblank Douglas Pulleyblank is Head of the Department of Linguistics at the University of British Columbia. He completed a PhD in Linguistics at the Massachusetts Institute of Technology in 1983. His main research interests are phonology and the interface between phonology and phonetics. He has a long-standing interest in Nigerian languages, particularly Yoruba, as well as African languages in general. Brian W. Smith Brian W. Smith is a doctoral student in the Department of Linguistics at the University of Massachusetts Amherst. His research deals with theoretical approaches to phonological variation and the phonology-morphology interface, especially in the framework of Harmonic Grammar. Irene Vogel Irene Vogel is Full Professor at the University of Delaware. She received her PhD from Stanford University. Her research focuses on the various interfaces of phonology with other components of grammar. Jeroen van de Weijer Jeroen van de Weijer is Full Professor of English Linguistics at Shanghai International Studies University. He took his PhD at the University of Leiden in 1994. His research focuses on combining models of theoretical phonology with psycholinguistics.

Bert Botma & Roland Noske

Introduction This collection of phonological papers grew out of the editors’ realization, back in 2010, that Norval Smith would soon reach the age of 65, when he would be due to retire from his post at the University of Amsterdam. Our intention was to present him with this book on the day of his retirement (September 30, 2011). As it turns out, the project took almost exactly a year longer than we had anticipated. The contributors to this volume include former colleagues, collaborators and students of Norval’s. Their contributions explore a broad range of topics. We believe that this reflects not only Norval’s own research interests (for which ‘broad’ is perhaps an understatement), but that it also offers a good representation of current phonological thinking, from different empirical, theoretical and diachronic angles. In “Allomorphy and the architecture of grammar”, Geert Booij considers a number of patterns of Dutch allomorphy. He observes that only a small part of these can be accounted for in terms of a common underlying form, with the surface forms being derived by phonological rules. For example, while it has been claimed that the relation between words like rode [roٝdԥ] and rooie [roٝjԥ] ‘red (inflected)’ is the result of a process of d > j weakening (see e.g. Smith 1973), such a view is in fact problematic. Not only is d > j weakening lexically restricted, but words with j may also have idiosyncratic meanings. For instance, rooie has the additional, unpredictable meaning of ‘communist’, suggesting that the word is stored. On the basis of such observations, Booij concludes that most allomorphs have to be listed in the lexicon, as words or as constituents of words. Booij goes on to show that this view of allomorphy has important consequences for the architecture of grammar. In “From prof to provo”, Camiel Hamans explores the interaction between phonology and morphology from the perspective of Dutch clippings. Hamans observes two distinct patterns: monosyllabic clippings (e.g. prof < professor) and a more recent pattern of disyllabic clippings ending in -o (e.g. provo < provocateur). A further development is that this -o is now also added to full forms, as in suffo ‘silly person’ (compare English sicko). Hamans examines two previous analyses of Dutch clipping, Van de Vijver (1997) and Hinskens (2001), and a more recent proposal for English in Lappe (2007). He observes that the analyses of Van de Vijver and Hinskens can be applied to the change from a monosyllabic to a disyllabic clipping template (with the change itself involving constraint reranking), while Lappe’s approach can be applied to the extension of -o to full forms.


Bert Botma & Roland Noske

Is recursion restricted to the syntactic component of the grammar, or is it also a property of the phonological component? This question has recently met with renewed interest (see esp. Van der Hulst 2010). In “Recursion in phonology?”, Irene Vogel focuses on an area where recursion has often been claimed to be relevant, viz. in structures involving Phonological Words. Vogel evaluates these claims and shows that the recursive structures that have been proposed introduce problems with regard to both phonological structure (e.g., they violate the Strict Layer Hypothesis) and linguistic structure in general (recursion in Phonological Words exhibits different properties than recursion in syntax). Vogel goes on to consider two alternative recursion-free proposals: a string-based analysis, and an analysis which retains the prosodic hierarchy. The latter account assumes an additional ‘Composite Group’, a constituent between the Phonological Word and the Phonological Phrase. The Composite Group permits skipping of prosodic levels (subject to several restrictions), and therefore avoids the problems with the Strict Layer Hypothesis. Vogel evaluates this analysis against a range of crosslinguistic data and psycholinguistic experiments. The next two contributions are concerned with chain shifts in the history of Germanic. In “The Grimm-Verner push chain and Contrast Preservation Theory”, Roland Noske shows that the adoption of the Glottalic Theory of the Proto-Indo-European obstruent inventory makes it possible to analyze Grimm’s Law and Verner’s Law as a single, bifurcating chain shift. This analysis resolves a number of problems associated with the traditional Neogrammarian view, in which Verner’s Law applied at a later time than Grimm’s Law. The assumption of a bifurcating chain shift explains why the undergoers of Verner’s Law are properly included in those of Grimm’s Law. It also replaces the peculiar chronological development that is traditionally assumed for Germanic (voiceless plosive > *voiceless fricative > *voiced fricative > voiced plosive) by a scenario in which voiced plosives derive directly from voiceless ones, without unattested intermediate forms. In the second part of his paper, Noske shows that the proposed scenario receives a natural account in Contrast Preservation Theory (àubowicz 2003, 2012; see also Montrueil 2006). In “Segmental structure and vowel shifts”, Janet Grijzenhout accounts for a number of historical developments in Germanic vowel systems, including vocalic chain shifts. In Grijzenhout’s model, vowels are specified in terms of the elements |I|, |U|, |A|, |@|, and combinations of these. Such combinations involve head/dependency relations between elements, such that one element is the head (and so phonetically more prominent) and the other the dependent (and so phonetically less prominent). Grijzenhout’s approach offers a straightforward interpretation of a number of diachronic changes. Most importantly, since elements are privative, they offer an inherent evaluation



metric to measure the complexity of segments. This allows a natural account of such processes as unrounding of front vowels. For example, the change /yٝ/ > /iٝ/, as has occurred in e.g. Old Icelandic, involves the loss of the |U| element, viz. |I,U| > |I|, yielding a less marked structure. Grijzenhout provides a dependency-based analysis of this change and a host of other vowel changes from Proto-Indo-European to Proto-Germanic, and from English and Icelandic. Turning to synchronic issues, in “The distribution of vowels in English and trochaic proper government”, Krisztina Polgárdi presents an account of the distribution of RP English vowels (short tense, long tense, short lax, and long lax) in terms of a loose CV framework, a representational approach that builds on earlier work in Government Phonology, in particular Lowenstamm (1996). An important ingredient of Polgárdi’s analysis is that stressed rhymes must properly govern an empty nucleus to their right. Polgárdi shows that this assumption helps to explain the complementary distribution of short and long vowels: the former do not occur in open syllables (e.g. *[bræ], where the short lax vowel cannot properly govern) while the latter cannot occur in closed syllables (e.g. *[viٝktԥ], where the second part of the long vowel cannot properly govern the following empty nucleus). Polgárdi assumes that apparently open syllables with short lax vowels (e.g. the first syllable in city) are closed by the first part of a ‘virtual geminate’. This allows the generalization that short lax vowels must be followed by a consonant cluster. Support for this comes from Welsh English, where virtual geminates are realized as long phonetically, e.g. city [‫چ‬sʏtٝʏ] and hook [hԂkٝ]. In “A propos of the Dutch vowel system 21 years on, 22 years on”, Bert Botma and Marc van Oostendorp take a fresh look at the phonological specification of the two sets of Dutch monophthongs (viz. /aٝ, eٝ, øٝ, oٝ, i, y, u/ vs. /:, Ų, ʏ, Þ, ֦/). In doing so, they build on insights from Smith et al. (1989), who were themselves inspired by the work of Rudolf de Rijk (cf. De Rijk 1967). Botma and Van Oostendorp identify three approaches to the Dutch vowel system: one in which the contrast is in terms of length, one in which the contrast is in terms of quality (between ‘tense’ and ‘lax’ vowels), and one in which the contrast is between ‘strongly cut’ and ‘weakly cut’ syllables. After reviewing the evidence, Botma and Van Oostendorp argue in favour of the last option, a view which originated in the work of Sievers and Trubetzkoy. Botma and Van Oostendorp point out that an advantage of a syllabic approach is that the contrast between the two sets of vowels need no longer be made in terms of a feature [tense] or [lax], for which phonetic support has not been forthcoming. Harry van der Hulst’s “A minimal framework for vowel harmony” is couched in the framework of Radical cv Phonology, a minimalist offshoot of the dependency-based approach to segmental structure. After outlining the


Bert Botma & Roland Noske

general background of the framework, Van der Hulst applies RcvP to a number of processes of vowel harmony. His main proposal is that vowels in a harmonic domain contain the harmonic element as a variable element. Variable elements surface only when they are ‘licensed’, either positionally (e.g. because they are in word-initial position) or laterally (i.e. because of a preceding or following instance of the same element). This approach has a number of interesting consequences. One concerns the status of ‘transparent’ vowels. Consider Finnish, a language with progressive palatal harmony. Van der Hulst formalizes the Finnish harmony pattern in terms of variable |I|, which, when present, is positionally licensed in the first vowel of a word and laterally in every following vowel (e.g. tyhmæ-stæ ‘stupid-ILL’). Finnish /i e/ have traditionally been described as ‘transparent’ (e.g. Van der Hulst & Smith 1986), since they occur in both front and back words, and do not impose their frontness on a following back vowel (e.g. tuoli-lla ‘chairADESS’, where /i/ is followed by a back vowel in the suffix). In Van der Hulst’s analysis, back words do not contain any instance of variable |I| (the [i] in tuolilla is argued to derive from underlying /ɰ/), so that here no harmony takes place. In front words, /i e/ simply participate in the harmony, as their specification for variable |I| is licensed either positionally or laterally (e.g. velje-llæ ‘brother-ADESS’). Hence, in Van der Hulst’s account ‘transparent’ vowels do not in fact exist. A quite different perspective on vowel harmony is presented in Diana Archangeli, Jeff Mielke and Douglas Pulleyblank’s “Greater than noise: frequency effects in Bantu height harmony”. The authors contrast two fundamentally different theories of phonological competence, viz. Universal Grammar (UG) and Emergent Grammar (EG). These two theories make different predictions about how phonological patterns are reflected in a language’s frequency data. Under UG, phonological patterns are expected to be (close to) categorical, predicting a tight statistical adherence to postulated patterns. Under EG, phonological patterns are expected to have more exceptions, predicting a looser statistical adherence. In addition, unlike UG, EG predicts that phonological patterns may be gradually extended to broader classes, with the relevant generalizations acting as ‘data attractors’. Archangeli et al. test these predictions by comparing the frequencies of vowel co-occurrence patterns in 6 Bantu languages with a five-vowel system (/i, e, a, o, u/) and height harmony, against 6 control languages with a five-vowel system and no harmony. The canonical height harmony pattern is displayed by Ciyao. In this language, harmony is primarily a property of verbal stems (in root-suffix combinations). The pattern is asymmetric, in that high front vowels surface as high after high and low root vowels and as mid after mid vowels, while high back vowels surface as high after high and low root vowels and after front mid /e/, but lower to mid back after /o/. Thus, the



sequences e…i, o…i and o…u (but not e…u) are expected to be virtually absent under a UG account, but merely under-represented under an EG account. The results of Archangeli et al.’s study are consistent with EG. The data indicate that height harmony in verbs is not categorical. It appears to extend morphologically to the class of nouns, and phonologically to any sequence of a mid plus high vowel sequence. Phonological extension is observed in nouns, which contain fewer e…u sequences than would be expected. This suggests that the harmony pattern is gradually extended in such a way that any high vowel lowers before a mid vowel. The next two papers in the volume deal with Limburgian. While normally classified as a tone-accent language, both show that the phonological status of tone in Limburgian is in fact a matter of contention. In “The phonological representation of Limburgian tone accents”, Ben Hermans argues that the two tonal word patterns in Limburgian, Accent 1 and Accent 2, are predictable from prosodic structure. Based on fieldwork data from the dialect of Maasbracht, Hermans observes a number of important regularities in the distribution of Accents 1 and 2. One is that a non-high vowel requires Accent 1 in a preceding syllable, while a high vowel requires Accent 2 in a preceding syllable. To account for this difference, Hermans adopts a model similar to that of Halle & Vergnaud (1987), but in which syllables are located in the same dimension as feet. In Hermans’ model, trochees can be bisyllabic or monosyllabic; the former have Accent 2, the latter have Accent 1. In this way, Hermans accounts for the distribution of the two accents. For example, a non-high vowel in the second syllable of a word is sufficiently sonorous to project its own foot. The initial syllable will therefore form a foot on its own (a monosyllabic trochee), which receives Accent 1. In contrast, a high vowel in the second syllable of a word is not sufficiently sonorous to project its own foot. Therefore, it forms a foot together with the initial syllable (a bisyllabic trochee), which receives Accent 2. In this analysis, there is thus no need to posit tones underlyingly, as in previous accounts of Limburgian tone. In “Quantity or durational enhancement of tone: the case of Maastricht Limburgian high vowels”, Carlos Gussenhoven offers further support for the claim that the high vowels of Maastricht Limburgian contrast in terms of length rather than tone (see Gussenhoven & Aarts 1999). Gussenhoven reports on a production experiment in which the length difference between tonal minimal pairs and minimal pairs with high vowels is compared. The results show that the length difference in high vowels, i.e. between /i, u/ and /iٝ, uٝ/, is treated differently from the length difference that is due to tone, in non-high vowels. The results of this experiment raise the question of how Maastricht Limburgian can maintain its tone contrasts along with the other vocalic contrasts that it employs. Gussenhoven suggests that the language is in the process of losing lexical tone. In the past, the situation in Maastricht


Bert Botma & Roland Noske

Limburgian was similar to that in the dialect of Cologne, where both high and non-high vowels can bear contrastive tone. The Maastricht dialect is probably developing towards the situation found in the dialect of nearby Weert, which has lost all lexical tone contrasts. In “Using constraint conjunction to discover constraints: the case of Mandarin Chinese”, Jeroen van de Weijer discusses a hitherto unnoticed restriction in Mandarin Chinese. Barring very few exceptions, this language lacks syllables with a voiced stop followed by a vowel with (a rising) tone 2 and a nasal (e.g. *dán, *bíƾ). As Van de Weijer points out, at least part of this restriction is phonetically natural, since the low tone that is associated with the voiced onset stop is antagonistic to the high tone of the following vowel. (It is less clear whether a similar antagonistic relation obtains between the high toned vowel and the nasal.) Van de Weijer analyzes the restriction in terms of two conjoined OT constraints, one banning sequences of a voiced onset and a high toned vowel, and one banning sequences of a high toned vowel and a nasal. By themselves these constraints are relatively low-ranked, but the conjunction of the two constraints is ranked high in the grammar of Mandarin Chinese. In “Implications of harmonic serialism for lexical tone association”, John McCarthy, Kevin Mullin and Brian Smith compare tone association in two versions of OT, parallel OT (Prince & Smolensky 1993[2004]) and Harmonic Serialism (e.g. McCarthy 2010). In parallel OT, output candidates may show the effect of several phonological changes simultaneously. Harmonic Serialism, on the other hand, is derivational: here output candidates show the effect of one change at a time, with the winning candidate being run through the grammar once again so that it can accumulate additional changes. McCarthy et al. observe that although there is growing support for Harmonic Serialism, the model is unable to account for the tonal pattern of languages like Kikuyu, if these tones are construed of as being lexically linked. This leads the authors to return to an assumption of early work in autosegmental phonology, viz. that tones are never associated in underlying representations — a position that is in accordance with the OT principle of Richness of the Base. In “A constraint-based explanation of the McGurk effect”, Paul Boersma provides an OT account of the McGurk effect, a phenomenon that illustrates the low-level interaction of visual and auditory cues in speech perception. Boersma formalizes the interaction between cue constraints (which evaluate the relation between sensory and phonological representations) and structural constraints (which evaluate phonological representations) using perception tableaux. The procedure of lexicon-driven perceptual learning then explains how the constraints come to be ranked as they are. Boersma goes on to show that the same cue constraints and structural constraints are used in production by speakers.



Frans Hinskens’ “Liquids in a case of unfolding early L1 Dutch” describes the development of /l/ and /r/ in a Dutch acquiring girl between the ages of 1;5 and 3;3. Hinskens shows that that the liquid system develops from one in which neither /l/ nor /r/ is realized, via free variation and a phase in which the realizations are probabilistically conditioned by internal constraints, into the prevalent adult system, where /l/ and /r/ are fully contrastive. Hinskens relates his case study to what is known about the unfolding of consonant inventories in the early L1 acquisition of Dutch, as well as to more general aspects of the phonology of liquids. In “The Tibetan numerals segmentation problem and how virtual learners solve it”, Diana Apoussidou compares the performance of two models in the acquisition of the morphological segmentation of Tibetan numerals. One model combines OT with the Gradual Learning Algorithm (GLA, Boersma 2008); the other model combines Harmonic Grammar (HG, Boersma & Pater 2008) with GLA. HG differs from OT in that it uses positive weights as well constraint violations represented by negative integers. Candidate forms are evaluated by multiplying the violations of a constraint with its corresponding weight. Apoussidou shows that her two groups of virtual learners (one using using OT/GLA, the other HG/GLA) acquire the Tibetan system, but do so in different ways. The OT/GLA learners use a restrictive lexicon, and so assign a greater role to the grammar. The HG/GLA learners, on the other hand, fall into two groups: one group uses a restrictive lexicon (like the OT/GLA learners) while the other makes extensive use of allomorphy. The difference between the latter group and the group of OT/GLA learners results from an important difference between OT and HG: in HG, the cumulative effect of lighter-weighted constraints can outweigh a heavier-weighted constraint. Apoussidou suggests that in the Tibetan numerals case, faithfulness and the respective lexical constraints together outweigh a heavier-weighted lexical constraint. This results in the production of more errors, which leads the latter group of HG/GLA learners to resort more to allomorphy. It remains for us to thank the contributors to this volume for meeting our rather stringent deadlines. We are also grateful to 35 anonymous reviewers for constructive comments and for their willingness to review papers at such short notice. Some of the papers in this volume were presented at a workshop at the University of Amsterdam, on the day of Norval’s retirement. We would like to thank the speakers, and Enoch Aboh and Kees Hengeveld for helping to organize the workshop research.


Bert Botma & Roland Noske

References Boersma, Paul (2008): “A programme for bidirectional phonology and phonetics and their acquisition and evolution.” ROA-868. Boersma, Paul & Joe Pater (2008): “Convergence properties of a gradual learning algorithm for Harmonic Grammar.” ROA-970. De Rijk, Rudolf (1967): “A propos of the Dutch vowel system.” Ms., MIT (electronic edition at Gussenhoven, Carlos & Flor Aarts (1999): “The dialect of Maastricht.” – In: Journal of the International Phonetic Association 29, 55–66. Halle, Morris & Jean Roger Vergnaud (1987): An Essay on Stress. – Cambridge, Mass.: MIT Press. Hinskens, Frans (2001): “Hypocoristische vormen en reductievormen in het hedendaagse Nederlands.” – In: Neerlandica Extra Muros 39: 37–49. Lappe, Sabine (2007): English Prosodic Morphology. – Dordrecht: Springer. Lowenstamm, Jean (1996): “CV as the only syllable type.” – In: Durand, Jacques & Bernard Laks (eds.), Current Trends in Phonology: Models and Methods, 419– 441. CNRS/ESRI, Paris X. àubowicz, Anna (2003): Contrast Preservation in Phonological Mappings. PhD dissertation, University of Massachusetts Amherst – Amherst, Mass.:GSLA. ROA554. – (2012): The Phonology of Contrast. – London: Equinox. McCarthy, John J. (2010): “An introduction to Harmonic Serialism.” – In: Language and Linguistics Compass 4, 1001–1018. Montrueil, Jean-Pierre (2006): “Contrast Preservation Theory and historical change.” – In: Gess, Randall S. & Deborah Arteaga (eds.), Historical Romance Linguistics. Retrospective and Perspectives, 111–129. Amsterdam/Philadelphia: John Benjamins. Prince, Alan & Paul Smolensky (1993[2004]): Optimality Theory: Constraint Interaction in Generative Grammar – Malden, Mass./Oxford, UK: Blackwell. Smith, Norval S. H. (1973): “The phenomenon of D-deletion in Dutch.” – In: Spektator 2, 421–437. Smith, Norval S. H., Roberto Bolognesi, Frank van der Leeuw, Jean Rutten & Heleen de Wit (1989): “A propos of the Dutch vowel system 21 years on.” – In: Bennis, Hans & Ans van Kemenade (eds.), Linguistics in the Netherlands 1989, 219–230. Dordrecht: Foris. Van der Hulst, Harry (ed.) (2010): Recursion and Human Language. Berlin: Mouton de Gruyter. Van der Hulst, Harry & Norval S.H. Smith (1986): “On neutral vowels.” – In: Bogers, Koen, Harry van der Hulst & Maarten Mous (eds.), The Phonological Representation of Suprasegmentals, 233–279. Dordrecht: Foris. Vijver, Ruben van de (1997): “The duress of stress: On Dutch clippings.” – In: Coerts, Jane & Helen de Hoop (eds.): Linguistics in the Netherlands, 219–230. Amsterdam: John Benjamins.

Geert Booij

Allomorphy and the architecture of grammar*

1 Introduction One of the tasks of phonology is to provide an adequate theory of how allomorphy, the variation in the phonological shape of morphemes and words, should be accounted for in the grammar of natural languages. In this article, I will argue that lexically governed allomorphy, and even allomorphy in general, should be accounted for primarily in the lexicon. This view has various implications for the nature of the lexicon and the architecture of grammar, as we will see below. One of the reasons for choosing this topic is that my esteemed colleague Norval Smith wrote some articles on this issue at the start of his linguistic career at the University of Amsterdam, in the 1970s. In those days, the phonological analysis of allomorphy received a new impetus through the rise of generative phonology, and in particular through the publication of Chomsky & Halle (1968). This also applies to the study of Dutch allomorphy, the topic of the present chapter and also of some of Norval Smith’s early publications. The classical generative approach to allomorphy is that all allomorphs of a morpheme are derived from a common underlying form by means of a set of (possibly ordered) phonological rules. A nice example of applying this idea of a common underlying form and a set of ordered rules to Dutch allomorphy phenomena can be found in Smith (1973). In this article (probably the first English contribution to Spektator, the journal for Dutch language and literature), Smith dealt with two types of allomorphy: the alternation between /d/ and the glides /j, w/, and the alternation between /dԥ/ and σ. The two types of alternation are illustrated in (1):

–––––––—–– * The research for this article was supported by a fellowship from the Alexander von Humboldt-Stiftung, which is gratefully acknowledged here. I thank Matthias Hüning (FU Berlin) and two anonymous reviewers for their comments on an earlier version of this article.


Geert Booij (1)


/d/-glide alternation kwade rode houden gouden


[kȣaٝdԥ] [roٝdԥ] [h:udԥn] [Ճ:udԥn]

kwaaie rooie houwen gouwen

[kȣaٝjԥ] [roٝjԥ] [h:uwԥn] [Ճ:uwԥn]

‘angry(inflected)’ ‘red (inflected)’ ‘to keep’ ‘golden’

slee la broer veer

[sleٝ] [laٝ] [bruٝr] [veٝr]

‘sledge’ ‘drawer’ ‘brother’ ‘feather’

/dԥ/-σ alternation slede lade broeder veder

[sleٝdԥ] [laٝdԥ] [brudԥr] [veٝdԥr]

Smith proposed to account for these alternations by means of two rules: a rule that deletes /d/, followed by a rule that inserts a glide. In those cases where glide insertion does not apply, a rule of schwa deletion removes the unwanted post-vocalic schwa, as in sled[Ω] > sle[Ω] > slee. This analysis was criticized in Zonneveld (1975), and Smith replied in Smith (1975). Zonneveld developed his analysis in more detail in his dissertation (Zonneveld 1978). Zonneveld’s basic objection was that the two alternations cannot be subsumed under one rule of /d/-deletion. Instead, there is one rule turning /d/ into a glide, and another one that deletes /dԥ/ intervocalically and at the end of a word. The details of this debate will not concern us here. Instead, I will comment on the underlying methodology of these analyses, and the problems that they evoke. The major problem of these analyses lies in the leading idea of classical generative phonology that all alternation patterns have to be accounted for by rules. These rules apply to underlying forms, and derive the various surface alternants. As far as the alternations in (1) go, a problem for this approach is that the relevant rules are lexically governed. That is, they do not apply to all words that meet the structural description of the relevant phonological rules, but only to a subset of them. For instance, the word slede ‘sledge’ cannot be realized as sleje [sleje], nor the word broeder ‘brother’ as broeier [brujԥr]. Also, the word woede ‘rage’ cannot be realized as woe [wu]. Therefore, a rule-based analysis is forced to mark all relevant individual words with a rule feature [+Rn], and to make that feature part of the structural description of Rule n. Alternatively, if the majority of words undergo the rule, the words that do not undergo it must be marked with a negative rule feature [–Rn], marking these words as negative exceptions to the rule. In this way, phonological rules can be blocked from applying to the wrong words. A second example of the rule-based approach to allomorphy in Dutch is the classical analysis of the phonological variation displayed by the diminutive suffix, with its five allomorphs -tje, -je, -kje, -pje, -etje: traan-tje ‘tearDIM’, kat-je ‘cat-DIM’, konin-kje ‘king-DIM’, riem-pje ‘belt-DIM’, zonn-etje

Allomorphy and the architecture of grammar


‘sun-DIM’. The choice of a specific allomorph is governed by the phonological shape of the stem. Hence, it was argued that these five allomorphs could be derived from one underlying form /tjԥ/, by means of a set of ordered phonological rules (Haverkamp-Lubbers & Kooij 1971). This topic has been extensively discussed by Dutch linguists; see the references in Booij (1995) and Van der Hulst (2008). The problem in this case is that certain diminutives have exceptional forms. For instance, the diminutive of brug ‘bridge’ is either the regular brug-je, or the unexpected brugg-etje. Similarly, for bloem ‘flower’ we find not only the regular diminutive noun bloem-pje but also the irregular form bloem-etje, with both the regular meaning ‘small flower’ and the idiosyncratic meaning ‘bunch of flowers’. In the case of diminutive allomorphy there is another problem: the rules that we need cannot be considered general phonological rules of Dutch, since they apply to diminutive words only. For instance, the allomorph -pje shows up after stems ending in /m/ preceded by a long vowel (riem – riem-pje). This looks like a rule of nasal place assimilation. However, whereas it is normally the case in Dutch (and universally) that the nasal consonant adapts its place of articulation to the following obstruent, in this case the order of assimilation is the reverse: the underlying /t/ assimilates to the preceding /m/. Therefore, the rules for the alternations in the diminutive suffix must be qualified as morpholexical rules (Booij 1995). A morpholexical rule is a rule whose application is governed by the presence of a specific lexical or morphological feature, in this case the feature [+diminutive]. Hence, two types of phonological rules have to be distinguished: automatic phonological rules, which apply whenever the phonological structural description of the rule is met, and morpholexical rules, which have a more restricted application. The distinction between these two types of rules, and its relevance for the organization of phonology, was argued for in detail by Anderson (1974). Anderson’s claim is that normally morpholexical rules precede automatic phonological rules. The non-automatic nature of the allomorphy of the diminutive suffix has prompted some Dutch linguists to come up with alternative analyses. For instance, Van der Hulst has recently proposed an analysis in which the five allomorphs are listed, but summarized in a schema with variables and a fixed common part je (/jԥ/) (Van der Hulst 2008). A similar solution is offered in Van Zonneveld (1978): (2)

-(ԥ) ({[–son], ([+cor])}) je

The selection of the variable material in this schema is then determined by the phonological properties of the stem: the variable material of the suffix can be omitted if this is necessary for the well-formedness of the output form. The basic idea is therefore that the allomorphs of this suffix are listed, but


Geert Booij

that the choice of a particular allomorph is still governed phonologically. A similar approach to allomorphy in Polish, in the framework of Optimality Theory, has been proposed in Rubach & Booij (2001). This approach also reflects the idea that if there is a clash between the phonological properties of a stem and those of a suffix, those of the stem prevail (Borowsky 2000). So far, it seems that the problem of restricted application of rules can be solved by means of the devices mentioned above: rules can be governed by lexical and/or morphological features. Alternatively, allomorphs may be listed, but selected by phonological constraints, in which case stem properties may be given precedence to suffix properties. However, there is another type of analytical problem in dealing with allomorphy phenomena, which we have already seen for the diminutive pair bloempje - bloemetje: the two allomorphs of a lexical morpheme may differ in meaning, or in stylistic value. For instance, the Dutch words broeder and broer ‘brother’ do not have the same range of meanings. The word broeder does not only mean ‘brother’ in the literal sense, but also ‘male nurse’, ‘male member of a religious order’, and ‘male member of a protestant church community’. These additional meanings are not available for the short form broer. In the case of the alternation rode – rooie mentioned in (1a), the second form is more informal. Moreover, it has a lexicalized meaning, i.e. ‘socialist’, that rode does not have. In such cases, the two forms must therefore be stored in the lexicon for non-phonological reasons, along with their specific meaning or stylistic value. The process of de-deletion is no longer productive, and the allomorphy is a relic of the past. This means that the de/σ alternation can at best be expressed in terms of a redundancy rule which states that some words have a corresponding de-less form with the same or a similar meaning. The same holds for the d/glide alternation, which cannot be extended to new words. However, while these alternations can be expressed in terms of redundancy rules, it is not obvious that speakers of Dutch do so. In fact, they do not need to do this, if we allow for the possibility that allomorphy can be lexically stored. As mentioned above, in diminutive allomorphy there are some words with two diminutive forms, e.g. kip-je – kipp-etje ‘chicken-DIM’ and brug-je – brugg-etje ‘bridge-DIM’. Moreover, there may be semantic differences between the two forms, as in bloem-pje ‘small flower’ vs. bloem-etje ‘bunch of flowers, bouquet’. This implies lexical listing of the diminutive nouns with the various suffix allomorphs. There are also diminutive nouns without corresponding base words, such as meis-je ‘girl’, for which the base word meis is not available (except for some speakers as the result of back formation). These words must be listed, despite the fact that the diminutive suffix has the regular shape -je that is required after an obstruent.

Allomorphy and the architecture of grammar


How can we do justice to regularities in alternation patterns, and at the same time to the non-automatic and lexicalized nature of this allomorphy? This seems impossible in the classical approach sketched above. The problem is caused by the rule–list fallacy (Langacker 1987), viz. the idea that information that is stored cannot at the same time be specified as instantiating a regularity, and vice versa. Current views of the lexicon avoid this fallacy, by specifying both the various lexical forms and the abstract alternation patterns that they instantiate in the lexicon (see e.g. Booij (2010); Jackendoff (2002), and the references mentioned there). This means in effect that rules may function as redundancy rules that specify to what extent the information on lexical entries is predictable, redundant information (Jackendoff 1975). In the case of broeder – broer, for instance, both words are stored in the lexicon. They can be related by means of a phonological schema that specifies in which contexts this de/σ alternation can occur. Each word is a combination of three types of information, phonological (PHON), morpho-syntactic (SYN), and semantic (SEM). The lexical information about the words broeder and broer will therefore have the following structure (with arbitrary lexical indices 9 and 10): (3) /brudԥr/9


SYN9 ļ SEM9a, SEM9b, SEM9c



SYN9 ļ SEM9a

These two lexical entries share one meaning, i.e. the literal meaning of ‘brother’( SEM9a), whereas the other meanings are unique for the long word form broeder. The syntactic properties are identical. The phonological redundancy schema that expresses the relevant pattern may then be formulated as follows: (4)

Redundancy schema §

The parts between angle brackets are schemas (i.e. correlations of phonological form, morphosyntactic form, and meaning), and ‘§’ stands for ‘is paradigmatically related to’. Hence, (4) expresses a generalization about all pairs of lexical entries of the sort exemplified in (3): for words with a long vowel followed by /dԥr/ or /dԥ/ there may be a corresponding word without /dԥ/. This rule has to be labeled explicitly as being a redundancy rule, since it cannot be applied productively to new cases. A precursor of this conception of the lexicon in relation to phonological alternations can be found in Leben & Robinson’s (1977) theory of ‘upsidedown phonology’. Phonological rules may work upside-down, to undo the effect of rules. The idea is that, given the strong lexical governedness of many phonological rules, it is preferable to store complex words in the


Geert Booij

lexicon in their surface forms. Phonological rules then have a redundancy rule function, and can be used to determine the relatedness of various lexical items by undoing the effect of phonological rules. For instance, in their approach the English rule of Trisyllabic Laxing would serve to relate sanity [sænʏti] to sane [seʏn], despite the fact that these words have different initial vowels. In the case of the deadjectival noun obesity, an exception to Trisyllabic Laxing, this rule is not necessary to relate the word to the adjective obese. Robinson also applied this idea to the allomorphs of Dutch diminutives (Robinson 1980). For instance, Dutch riem-pje ‘belt-DIM’ is computed back to riem-tje, and consequently recognizable as the diminutive form of riem ‘belt’. In the case of broeder – broer, undoing the rule of de-deletion would mean that broeder is reconstructed by inserting de, which can then be related to the word broeder with the same meaning. However, notice that this incorrectly predicts that broer has the same range of meanings as broeder. Therefore, the upside-down phonology approach cannot be the whole solution, as it cannot deal with semantic differences between allomorphs. After this short introductory sketch of the issues that allomorphy raises with respect to the architecture of grammar, the next sections will discuss in more detail how allomorphy should be accounted for. The leading idea is that allomorphy is to a remarkably large extent a matter of the lexicon and of morphology, and not of phonology. Section 2 shows how the selection of an allomorph may be governed by morphological rather than phonological considerations. Section 3 argues that in a model which assumes paradigmatic word formation, selection of the appropriate allomorph is a straightforward affair. Section 4 discusses the role of phonological output conditions in the choice between allomorphs. Section 5 considers briefly the implications of allomorphy for the issue of storage vs. computation. Finally, section 6 summarizes my conclusions as to what allomorphy implies for the architecture of grammar.

2 Morphological implications of allomorphy Morphemes may vary in their phonetic shape due to the effect of automatic phonological rules. An example is the rule of syllable-final devoicing in Dutch; this explains why the lexical morpheme hoed ‘hat’ has the shape [hut] when used as a singular form, and [hud] in the plural form hoed-en /hudԥn/. The standard analysis is that the morpheme hoed has the underlying form /hud/. However, in many cases allomorphy has lost its synchronic motivation. This is the case for the words in (1b): these words were subject to a historical

Allomorphy and the architecture of grammar


phonological process of de-deletion that affected words one by one. Synchronically, we end up with two different words that may still be synonymous. However, in most cases there is independent evidence (semantic, pragmatic, stylistic, or other) for their status as separate lexical items. In some cases, language users may feel that the two words are no longer related, as is the case for a word pair like ijdel ‘vain’ – ijl ‘thin’. The same applies to Dutch words that differ in the presence vs. the absence of a final schwa, such as those in (5): (5) aard[ԥ] eind[ũ] er[ũ] keuz[ũ] leuz[ũ] wijz[ũ]

aard eind eer keus leus wijs

‘earth’ ‘end’ ‘honour’ ‘choice’ ‘slogan’ ‘manner’

There is no automatic rule of word-final schwa-deletion in present-day Dutch, and speakers of Dutch have to learn in which case this alternation applies. The process of word-final schwa-deletion is no longer productive. Therefore, we have to assume that both words are stored in the lexicon, and that they may be related by a phonological redundancy schema which states that nouns that end in a schwa may have a correspondent without schwa, with the same meaning. This schema, which expresses a paradigmatic relationship between two sets of words, will have the format shown in (4). Since this schema is not productive, language users can do without it, as they will have stored all relevant cases. When words of the kind in (5) function as constituents of compounds, it may be that one of the allomorphs has to be chosen obligatorily. For instance, the word eind ‘end’ can function as first part of a compound, as in the forms in (6): (6) eind-bedrag eind-gesprek eind-oordeel eind-verslag

‘final amount of money’ ‘final discussion’ ‘final judgment’ ‘final report’

In these compounds, the allomorph eind cannot be replaced with the long allomorph einde: a word like *einde-bedrag is ill-formed. On the other hand, the long form does occur as the rightmost constituent (i.e. as the head) of compounds, in forms as in (7): (7) gespreks-eind(e) levens-eind(e)

‘discussion end, end of discussion’ ‘life end, end of life’


Geert Booij

In these words, the long form may be replaced with the short form; for instance, we also find levens-eind, as shown in (7). The data in (6) imply that compounds with initial eind- have to be stored in the lexicon. This is an interesting conclusion in relation to the debate on the balance between storage and computation. These compounds are quite regular as to form and meaning. However, even though their meaning is fully transparent, they must be stored in order to specify the correct choice of allomorph. The regularity that it is the short form eind that has to be used in compounds can be expressed by a subschema for NN compounds of the following type (Booij 2005, 2009a, 2010): (8)

In this schema, the double-arrowed symbol denotes the correlation between form and meaning. The meaning ‘final’ correlates with this use of the short form eind. The preceding discussion shows that the choice of a particular allomorph may be morphologized, i.e. depend on its position in a complex word. This form of lexicalization may go hand in hand with the development of allomorphs into affixoids. For instance, the word eind- in the compounds in (3) has acquired the specific meaning ‘final’, and we might therefore claim that Dutch has acquired a prefix eind- with this meaning (Booij 2005). The term ‘affixoid’ is used here to denote words with a ‘bound’ meaning, i.e. a meaning linked to a word that is part of a complex word. This affixoid behaviour can also be observed in the distribution of the words ere [e:rԥ] and eer [e:r], both meaning ‘honour’. In isolation, the short form eer is preferred to the long form ere (which has an archaic flavour). However, when used with the meaning ‘honorary’, it is always the long form that is used, while the short form is used when other meanings are involved: (9) a. ere-lid ‘honorary member’ (litt. ‘honour member’) ere-voorzitter ‘honorary chairman’ (litt. ‘honour chairman’) b. eer-betoon eer-bied

‘mark of honour’ (litt. ‘honour show’) ‘respect’ (litt. ‘honour offering’)

This regularity concerning ere- can be expressed by a morphological subschema for NN compounds with ere as their first constituent. (10)

[erei Nj]Nk ļ [honoraryi Nj]k

This kind of variation is reminiscent of the morphomic phenomena discussed in Aronoff (1994). In many languages, a lexical item may have various stems that are used in inflection and word formation. In Latin, for instance, verbs have three different stem allomorphs, and the choice of a particular stem variant is determined by purely morphological considerations. In the case of


Allomorphy and the architecture of grammar

Dutch, the choice of allomorph may also be conditioned semantically, as shown in (9): one allomorph, used in a particular morphological structure, carries one meaning from the set of meanings of a lexical item. We have seen, therefore, that an allomorph which results from a historical phonological process of schwa-deletion may receive a new interpretation as an affixoid, that is, as a word with a specific meaning in a specific morphological structure.

3 Morphological selection of allomorphs The data discussed in section 2 show that the selection of allomorphs may shift from phonology to morphology (cf. also Booij 2002, Chapter 5.3). The morphological selection of allomorphs can also be observed in the systematic difference between native and non-native allomorphs of Dutch words. Consider the following sets of related words: (11)

Base word

Native suffix

Non-native suffix

a. filter /fʏltԥr/ ‘filter’ filter-en ‘to filter’ filtr-eer ‘to filter’ b. regel /reٝՃԥl/ ‘rule’ regel-en ‘to arrange’ regul-eer ‘to regulate’ c. orkest /Þrkİst/ ‘orchestra’ orkest-en ‘orchestras’ orkestr-eer ‘to orchestrate’

In (11a), the word filter has the allomorph filtr- in words coined with a nonnative suffix; in (11b), the word regel has regul- as its allomorph before the non-native suffix -eer; in (11c), the word orkest has the allomorph orkestrbefore the non-native suffix -eer. This may seem to imply that for each of these three words we have to list two stem allomorphs, a native stem that is the form when used as a word by itself (or in combination with a prefix, as in ge-regel ‘(the act of) arranging things’), and a non-native stem allomorph that is used before non-native suffixes. In some cases, it may look as though the relation between the two allomorphs can be captured by the phonology. Suppose we assume the underlying form /fʏltr/ for the word filter. From this underlying form we may derive the word filter by means of schwa insertion, a phonological rule that rescues the unsyllabifiable coda cluster /tr/. In filtreer, schwa insertion does not apply since the syllabification is fil.treer (the dot indicates a syllable boundary), with /tr/ forming an onset. However, this does not explain why a schwa is inserted in the (infinitive form of the) verb filteren, as fil.tren would also be well formed prosodically. That is, the presence of schwa within the stem is not due to phonological requirements. In the case of orkest, we might assume an underlying form with final /r/, i.e. orkestr. However, here the /r/ cannot be rescued by schwa-insertion (as is the


Geert Booij

case for German: Orchester) since the word in isolation is not orkester, but orkest. In the case of regel, with its non-native stem allomorph regul- , the situation is even more complex. What can be achieved as a descriptive generalization is that we assign the allomorphy to the stem, and not to the suffix. Consider the following Dutch complex adjectives and their base words: (12)

moment tekst ratio positie

‘moment’ ‘text’ ‘ratio’ ‘position’

moment-eel tekstu-eel ration-eel position-eel

‘at this moment’ ‘textual’ ‘rational’ ‘positional’


muziek rabbi ras dictator

‘music’ ‘rabbi’ ‘race’ ‘dictator’

muzik-aal rabbin-aal raci-aal dictatori-aal

‘musical’ ‘rabbinal’ ‘racial’ ‘dictatorial’

These facts are described in De Haas & Trommelen (1993) as cases of suffix allomorphy. The suffix -eel is said to have the allomorphs -ueel and -oneel, and the suffix -aal is described as having the allomorphs -naal and -iaal. Such a description misses the generalization that the extra elements i, u, n and on recur in the different types of non-native complex words. For instance, on also shows up in position-eer, and the n also appears in rabbinaat ‘position of a rabbi’ and in rabbin-isme ‘teachings of the rabbis’. Hence, these extra ‘bits of sound’ are stem extensions rather than initial parts of suffixes. It will now be clear that these alternations do not belong to the domain of phonology proper. Does this mean that the stem allomorphs of the relevant words have to be listed as such? This would mean that speakers of Dutch memorize allomorphs such as regul- and position-. The question is: how do speakers acquire these allomorphs? The obvious answer is: as part of the complex words that they come across and memorize. In other words, a stem allomorph like rabbin- is not stored in isolation, but as part of complex words such as rabbinaal and rabbinaat. The only reason why one might think that such stem allomorphs are stored as such is that they can be used for coining new words. For example, if we want to derive a word in -ist from the word positie ‘position’, the word will be positionist, not positist. However, we do not need to list a stem allomorph in isolation, because new words can be coined on the basis of paradigmatic relations between existing words (Booij 2002, 2010). That is, we can derive positionist from positioneel by replacing the suffix -eel with the suffix -ist. The necessity of assuming paradigmatic relationships is clear for independent reasons, from cases where there is no base word that is shared by the

Allomorphy and the architecture of grammar


word pairs (Booij 2010). Consider the following English word pairs in -ism and -ist: (14)

altru-ism aut-ism bapt-ism commun-ism pacific-ism

altru-ist aut-ist baptist communist pacif-ist

Even though they have no corresponding base word, the meaning of one member of a pair can be defined in terms of that of the other member. In particular, the meaning of the word with -ist can often be paraphrased as ‘person with the ability, disposition, or ideology denoted by the word in ism’. Hence, the following paradigmatic relationship can be defined for these two schemas: (15) §

where SEMi represents the meaning of the word in -ism, and the angle brackets mark the edges of a constructional schema. Thus, an altruist has a disposition for altruism, and a pacifist adheres to the ideology of pacifism. The paradigmatic relationship between these two schemas may lead to the coining of new words. For instance, if we know what determinism is, we can easily coin the word determinist, which predictably denotes a person adhering to determinism. In sum, stem allomorphy can easily be recognized and recovered on the basis of existing and hence listed complex words, from which other complex words can be derived by means of affix substitution. What remains to be accounted for is when and how language users recognize various allomorphs as being formal variants of the same word. I will not discuss this issue in this chapter; see Booij (2010: Chapter 10) for a brief discussion. Our focus here is on the implications of allomorphy for the architecture of grammar, and the facts discussed above lead to the conclusion that allomorphy is massively stored in the lexicon and encoded in the lexical representation of existing, listed complex words.

4 Phonological selection of allomorphs If the various allomorphs of a morpheme cannot be derived from one underlying form, this does not mean that their distribution cannot be governed by phonological conditions, as pointed out by Carstairs (1988): the


Geert Booij

two competing forms may be phonologically completely unrelated, yet their selection may be determined phonologically. A good example of this situation is the competition between the Dutch suffixes -er and -aar, discussed in another article by Norval Smith (Smith 1976). These two suffixes have a common historical origin; according to the Woordenboek der Nederlandsche Taal, the suffix -aar is a later, strengthened form of -er. Synchronically, it does not make sense to assume a rule that can derive -er from -aar or vice versa, as there is no independent evidence for such rules apart from the -er/-aar alternation. If we tried to capture such alternations in phonological terms, we would end up with a very complicated, and very probably unlearnable phonological system. Therefore, it is a better idea to come up with an analysis in which the phonological complementary distribution of allomorphs is captured in a more insightful way. For instance, the selection of allomorphs can be modeled as the result of a set of ranked phonological output conditions. This is proposed for -er and -aar in Booij (1998), and for certain types of allomorphy in Polish by Rubach & Booij (2001). The basic generalization for the Dutch cases is that -aar occurs after a stem ending in an unstressed syllable with a final coronal sonorant, and -er elsewhere (with the variant -der after /r/) (Booij 2002: 183): (16) a. eet judo Amsterdam wetenschap

‘to eat’ ‘to do judo’ ‘id.’ ‘science’

et-er judo-er Amsterdamm-er wetenschapp-er

‘eater’ ‘judoist’ ‘inhabitant of Amsterdam’ ‘scientist’

b. vereer vier Bijlmermeer

‘to worship’ vereer-der ‘worshipper’ ‘to celebrate’vier-der ‘celebrator’ ‘id.’ Bijlmermeer-der ‘inhabitant of Bijlmermeer’

c. looch[ԥ]n luist[ԥ]r knuts[ԥ]l Diem[ԥ]n Udd[ԥ]l

‘to deny’ ‘to listen’ ‘to tinker’ ‘id.’ ‘id.’

loochen-aar luister-aar knutsel-aar Diemen-aar Uddel-aar

‘denier’ ‘listener’ ‘tinkerer’ ‘inhabitant of Diemen’ ‘inhabitant of Uddel’

The use of -aar instead of -er after an unstressed syllable avoids the creation of a sequence of two unstressed syllables; the use of -der avoids the illformed phonological sequence /rԥr/. Hence, the selection of the various allomorphs can be stated in terms of phonological output conditions (Booij 1998). As in other cases of allomorphy, one finds irregular forms such as leraar ‘teacher’, derived from leer ‘to teach, to learn’ (we would expect leerder, which is the correct form for the meaning ‘learner’), and dien-aar ‘servant’ instead of the expected dien-er (there is also the irregular word dien-der ‘police officer’, with the allomorph -der). Again, this shows that the allomorphs have to be considered as different, competing suffixes, with a

Allomorphy and the architecture of grammar


certain degree of phonological similarity, and that lexical listing of complex words with the correct allomorph is necessary. This type of interaction between morphology and phonology receives a natural interpretation in a tripartite parallel architecture of grammar, as proposed in Jackendoff (2002), with three levels of representation: phonology, morpho-syntax, and semantics (as was illustrated in section 1), and interface components specifying the systematic relations between these levels. In the case of the competition between -er, -aar and -der discussed here, this means that at the level of morphology, morphological schemas will create words of the types X-er, X-aar (the restriction that -aar is allowed only after stems ending in a coronal sonorant reflects another type of interface between morphology and phonology, viz. phonological subcategorization) and X-der. The relevant interface principle is defined as follows: ‘from the set of competing complex words, choose the word with the optimal corresponding phonological form’. This explains why we have et-er ‘eater’ instead of *etaar or *eet-der, as eter has the optimal prosodic form of a trochee and the simplest syllable contact possible; and it explains why we have loochen-aar ‘denier’ instead of *loochen-er: the latter word has a final unstressed syllable that cannot be parsed as the right constituent of a trochaic foot, given that loochen /loxԥn/ already forms a trochee on its own. Existing exceptional words such as ler-aar, dien-der and opener ‘opener’ (instead of the expected openaar) are listed, and thus override this general selection principle.

5 Storage versus computation Allomorphy phenomena have interesting consequences for the issue of what kind of information is stored, and which information is computed by the grammar. As was shown above, complex words must be listed in the lexicon if they behave irregularly with respect to allomorphy, a conclusion also reached by Zuraw (2010) on the basis of an analysis of lexical variation in allomorphy in Tagalog. Therefore, some complex words must be listed, despite the fact that they are morphologically regular. From a psycholinguistic point of view, it is obvious that storage of regular forms is quite normal: if regular word forms exhibit frequency effects, they must be lexically stored (Baayen et al. 2003). This conclusion is supported by the facts mentioned in (1): the morphologically fully regular inflected adjectives kwaaie, rooie and the infinitive houwen must be listed in the lexicon because they allow for d > j weakening. This weakening does not apply to all inflected adjectives with a stem ending in /d/, as shown in (17a);


Geert Booij

similarly, (17b) shows that not all relevant infinitives in -en allow for dweakening. (17) a. rode ‘red’ > ro[j]e but wred-e b. lijden ‘suffer’ > lij[j]en but mijden

‘cruel’ > ‘avoid’ >

*wreje *mij[j]en

In classical generative phonology, one would have to assign negative rule features to all words whose inflected forms do not undergo the weakening rule. A more adequate conception of the mental lexicon is one in which language users can store inflected forms. Storage makes it possible for such form to acquire idiosyncratic properties, e.g. that they belong to an informal stylistic register or that they have a special meaning (for example, rooie has the additional meaning of ‘socialist’, which rod-e does not have). The claim that the outputs of phonological rules can be stored in lexical representations is supported by the observation that allomorphy is preserved even after the relevant phonological process has disappeared. This is for example the case for the early Germanic rule of vowel lengthening in open syllables (Prokosch’ Law). Relics of this alternation are found in some nouns, and also in some diminutives: (18) sch[ʏ]p p[:]d

‘ship’ ‘path’

sch[eٝ]p-en p[aٝ]d-en

‘ships’ ‘paths’

sch[eٝ]p-je p[aٝ]d-je

‘ship-DIM’ ‘path-DIM’

These facts, and their implication for what is stored in the lexicon, are discussed in more detail in Booij (2009b).

6 Conclusion Allomorphy is a pervasive phenomenon in the grammar of Dutch. Only a small part of it can be accounted for in terms of a common underlying form for the allomorphs, with phonological rules deriving the surface forms (but even there one might assume storage of surface forms, cf. Booij (2009b)). Most allomorphs have to be listed in the lexicon, as words or as word constituents. The choice of allomorph may depend on morphological structure (section 2), on paradigmatic relations (section 3), or on prosodic output conditions (section 4). There is positive evidence that regular complex words, even inflected ones, are stored in the lexicon. This allows us to specify allomorphs in the lexicon without the formal machinery of exception features. In our approach, stylistic and semantic differentiation between allomorphs finds a natural interpretation, reflecting the old idea (‘Humboldt’s

Allomorphy and the architecture of grammar


universal’) that if there is difference in form, there may also be difference in meaning. The preservation of the effect of phonological rules after their disappearance also implies the lexical storage of allomorphy (section 5). In sum, our conclusion is that a proper architecture of the grammar of natural languages has four basic ingredients: a tripartite parallel architecture with interface conditions, storage of complex words (whether regular or irregular), morphological schemas that specify allomorphs, and paradigmatic relations between complex words that can be actuated by coining new complex words.

References Anderson, Stephen R. (1974): The Organization of Phonology. – New York: Academic Press. Aronoff, Mark (1994): Morphology by Itself: Stems and Inflectional Classes. – Cambridge, MA: MIT Press. Baayen, R. Harald, James M. McQueen, Ton Dijkstra & Rob Schreuder (2003): “Frequency effects in regular inflectional morphology: revisiting Dutch plurals.” – In: Baayen, R. Harald & Rob Schreuder (eds.): Morphological Structure in Language Processing, 355–90. Berlin: Mouton de Gruyter. Booij, Geert (1995): The Phonology of Dutch. – Oxford: Clarendon Press. – (1998): “Prosodic output constraints in morphology.” – In: Kehrein, Wolfgang & Richard Wiese (eds.): Phonology and Morphology of the Germanic Languages, 143–163. Tübingen: Niemeyer. – (2002): The Morphology of Dutch. – Oxford: Oxford University Press. – (2005): “Compounding and derivation: evidence for Construction Morphology.” – In: Dressler, Wolfgang, Dieter Kastovsky, Oskar Pfeiffer & Franz Rainer (eds.): Morphology and its Demarcations, 109–132. Amsterdam/Philadelphia: John Benjamins. – (2009a): “Construction morphology and compounding.” – In: Rochelle Lieber & Pavol Stekauer (eds.): The Handbook of Compounding, 201–216. Oxford: Oxford University Press. – (2009b): “Lexical storage and phonological change.” In: Hanson, Kristin & Sharon Inkelas (eds.): The Nature of the Word. Essays in Honor of Paul Kiparsky, 497– 505. Cambridge MA: MIT Press. – (2010): Construction Morphology. – Oxford: Oxford University Press. Borowsky, Toni (2000): “Word-faithfulness and the direction of assimilation.” – In: The Linguistic Review 17, 1–28. Carstairs, Andrew (1988): “Some implications of phonologically conditioned suppletion.” – In: Booij, Geert & Jaap van Marle (eds.): Yearbook of Morphology 1988, 67–94. Dordrecht: Foris.


Geert Booij

Chomsky, Noam & Morris Halle (1968): The Sound Pattern of English. – New York: Harper and Row. De Haas, Wim & Mieke Trommelen (1993): Morfologisch Handboek van het Nederlands. Een Overzicht van de Woordvorming. Den Haag: SDU. Haverkamp-Lubbers, Roos & Jan Kooij (1971): Het Verkleinwoord in het Nederlands. – Amsterdam: Instituut voor Algemene Taalwetenschap, Universiteit van Amsterdam. Jackendoff, Ray (1975): “Semantic and morphological regularities in the lexicon.” – In: Language 51, 639–671. – (2002): Foundations of Language. – Oxford: Oxford University Press. Langacker, Ronald (1987): Foundations of Cognitive Grammar. Vol. 1: Theoretical prerequisites. – Stanford, CA: Stanford University Press. Leben, William R. & Orrin W. Robinson (1977): “‘Upside-down’ phonology.” – In: Language 53, 1–20. Robinson, Orrin W. (1980): “Dutch diminutives over easy.” – In: Zonneveld, Wim, Frans van Coetsem & Orrin W. Robinson (eds.): Studies in Dutch Phonology, 139–57. The Hague: Martinus Nijhoff. Rubach, Jerzy & Geert Booij (2001): “Allomorphy in Optimality Theory: Polish iotation.” – In: Language 77, 26–60. Smith, Norval S. H. (1973): “The phenomenon of D-deletion in Dutch.” – In: Spektator 2, 421–437. – (1975): “In suppport of D-deletion.” – In: Spektator 5, 17–22. – (1976): “-Aar.” – In: Leuvense Bijdragen 65, 485–96. Van der Hulst, Harry (2008): “The Dutch diminutive.” – In: Lingua 118, 1288–1306. Van Zonneveld, Ron (1978): “Verkleinwoordvorming.” – In: Van Berkel, A., W. Blok, G. Brummel & Th.A.J.M. Janssen (eds.): Proeven van Neerlandistiek, Aangeboden aan Prof. Dr. Albert Sassen, 279–302. Groningen: Nederlands Instituut Groningen. Zonneveld, Wim (1975): “A reanalysis of D-deletion in Dutch.” – In: Spektator 4, 231–239. – (1978): A Formal Theory of Exceptions in Generative Phonology. – Lisse: Peter de Ridder Press. Zuraw, Kie (2010): “A model of lexical variation and the grammar with application to Tagalog nasal substitution.” – In: Natural Language and Linguistic Theory 28, 417–472.

Camiel Hamans

From prof to provo: some observations on Dutch clippings*

1 Introduction Traditionally, clipping (also called shortening, abbreviation or truncation) of lexical material has been viewed as an irregular and rather eccentric process of word formation. For example, Stockwell & Minkova (2001: 10) observe that shortening may take any part of a word, usually a single syllable, and throw away the rest, like quiz from inquisitive, phone from telephone, plane from airplane, flu from influenza …The process often applies not just to an existing word, but to a whole phrase. Thus mob is shortened from mobile vulgus “fickle rabble”. Zoo is from zoological gardens. Ad and British advert are transparently based on advertisement … Many shortenings have entered the language and speakers have lost track of where they came from. How many people would recognize gin and tonic as coming from Genève?

According to Marchand (1969: 441), clipping is not a regular grammatical process but a stylistic and sociolinguistic phenomenon belonging to informal registers and special jargons. This leads him to claim that clipping forms part of the parole (speech) instead of the langue (system). The same claim, in a different theoretical framework, is made by Aronoff (1976: 20–21). However, the examples provided by Marchand (e.g. fridge for refrigerator, plane for airplane, maths for mathematics) show how common clipped forms are in Standard English. Marchand distinguishes between four types of clipping: (i) Back clipping

temp ad

> >

*boet *vreed *reed

‘fine’ ‘peace’ ‘speech’

These data could suggest a change from a non-trochaic to a trochaic metrical pattern. Further support for this comes from the accommodation of loans, e.g. pijler ‘pillar’ from Latin pila, with stress shift and subsequent diphthongization. Not all the evidence points in the same direction, however. Another loanword from the same base, pilaar, retained stress on the final syllable. In addition, the preference for a trochaic pattern is not manifested in allegro speech, where instead of an expected stress shift we find complete vowel reduction and concomitant monosyllabification (Awedyk & Hamans 1998), as in (16).

From prof to provo: some observations on Dutch clippings (16)

ba.'naan par.'tij ka.'non

> > >

bnaan p(r)tij knon


‘banana’ ‘party’ ‘cannon’

These data suggest that there is more variability in preferred minimal word structure than is sometimes assumed (though whether it is justified to compare the effects of fast speech with ‘normal’ grammatical patterns is an open question). These considerations aside, we can observe two competing clipping patterns in modern Dutch: an older pattern, illustrated in (4), of mainly monosyllabic forms, and a relatively recent pattern, illustrated in (5), (11) and (12), of mainly disyllabic trochees with a final, possibly suffixed, -o. The recent pattern emerged in the mid 1980s, due to influence of American English (Kuitenbrouwer 1987; Hamans 1996, 2004a, 2004b). Dutch has only two disyllabic clipped nouns with final -o that are older, viz. indo ‘Indonesian-Dutch half-breed’ (first attested in 1898) and provo (introduced by the criminologist Wouter Buikhuisen in 1965). (The clipped form prof, from professor, was first attested in 1875.) As noted, the preponderance of monosyllabic clippings in Lappe’s corpus masks the recent emergence of disyllabic truncation in (American) English. Lappe’s corpus consists of three dictionaries (one of standard English and two slang dictionaries) from the early 1980s, a few years before -o clipping became fashionable. The pattern was already productive in Australian English earlier, but the influence of Australian English on American and British English is small, and its influence on Dutch negligible. To date, no satisfactory explanation has been advanced for the emergence of -o clippings in American English, owing to the lack of reliable data. Since clipped forms, and in particular -o clippings, belong to informal registers, a reasonable scenario is that this pattern originated in the street jargons of youngsters in the big American cities, with a possible influence from Hispanic and perhaps Italian (see Hamans 2004b). It is interesting to observe in this respect that the -o ending has subsequently spread to full words which form part of and originated in youngsters’ slang: (17)

sicko weirdo creepo

< <

However, the Dutch preference for disyllabic trochaic minimal words, which predates the change from mono- to disyllabic clippings, has not yet become absolute, and most likely never will. This means that the language system displays some variability (which is not problematic in Prosodic Morphology). Unfortunately, we do not yet have sufficient diachronic data to refine this very sketchy picture. The account offered by Lappe also contributes to our understanding of the change, in particular because the change did not just involve truncation, but was also extended to full forms, with suffixation of -o. English examples of this were given in (17). Some Dutch examples are provided in (19) and (20): (19)

lullo duffo suffo