Generative studies on Creole languages 9783111392844, 9783111030340

Table of contents :
Introduction
1. Admissibility conditions on final consonant clusters in the Jamaican continuum Glenn Akers
2. Multifunctionality as a derivational problem Jan Voorhoeve
3. The development of the category of number in Tok Pisin Peter Mühlhäusler
4. Pursuing Creole roots Bill Washabaugh
5. A re-evaluation of the predicate in Ile-de-France Creole Chris Come
6. The developing complementizer system in Tok Pisin Ellen Woolford
7. Marking WH-Movement in Afrikaans Hans den Besten
8. Creole tense/mood/aspect systems: the unmarked case? Pieter Muysken
9. Haitian Creole Pu Hilda Koopman and Claire Lefebvre
Addresses of contributors

Citation preview

Generative Studies on Creole Languages

Studies inGenerative Grammar The goal of Studies in Generative Grammar Is to publish those texts that are representative of recent advances in the theory of formal grammar. Too many studies do not reach the public they deserve because of the depth and detail that make them unsuitable for publications in article form. We hope that the present series will make these studies available to a wider audience than has been hitherto possible.

Jan Koster Henk van Riemsdijk

Pieter Muysken (Ed.)

Generative Studies on Creole Languages 1981 FORIS PUBLICATIONS Dordrecht - Holland / Cinnaminson - U.S.A.

© 1980 Foris Publications - Dordrecht. No part of this book may be translated or reproduced in any form, by print, photoprint, or any other means, without written permission from the publisher. ISBN 90 70176 106 Printed in the Netherlands by Intercontinental Graphics.

Introduction

This volume brings together a number of studies which try to cope with data from Creole (and in a few cases, Pidgin) languages from a generative point of view. As such its aim is two-fold: to interest scholars and students working on Pidgin and Creole languages in recent work in generative linguistics; and to interest people who have been part of the mainstream of generative linguistics in new types of insights which can emerge from the study of Pidgin and Creole languages as evolving and dynamic systems. Many serious scholars specialized in Pidgin and Creole linguistics have considered the generative paradigm to be hopelessly inadequate, for two reasons. Not only did they consider that artificial rigidity and uniformity were imposed on complex and variable data (as several critics have charged with regard to Beryl Bailey's 1966 generative grammar of Jamaican basilect), but also that structures directly imported from English were imposed on languages which diverge markedly from the European mold (see Roberts, 1975). Most generativeists, on the other hand, have shied away from Pidgin and Creole languages because of the variability hinted at or described in the literature. This situation may be changing somewhat as generative theory is becoming more focussed on ultra- and interlinguistic variation. The resulting relative independence of the study of Pidgins and Creoles from the generative paradigm has led to a number of refreshing insights into the complexity of Creole grammatical phenomena. On the other hand, we find that their analysis has remained relatively shallow, in spite of a feverish interest in the field over the last ten years or so. In contrast, the contribution of Pidgin and Creole linguistics to various brances of sociolinguistics, particularly variation theory, has been considerable and profound. This volume originated under the conviction that the study of Pidgin and Creole languages constitutes an important part of the research program of generative grammar for three reasons, which merit some discussion here. First, an analysis of the stages through which a Pidgin develops into a native language, a Creole, can give us insights into the minimal require-

VI

Introduction

ments for natural languages. Within the generative paradigm the process of creolization is seen in mentalistic terms. Systems of communication which have had a parasitic and grammatically peripheral status as Pidgins or other secondary uses, acquire native speakers. This means that they will have to adapt so as to be definable by the language acquisition device (cf. Naro, 1973). No matter what the Pidgin ancestor was like, the Creole will have to be learnable by the child. Since within the generative paradigm language acquisition is seen as an interaction between imperfect data input and a complex set of hypotheses on the part of the child, the process of creolization can give us direct insight into what kinds of hypotheses the child will formulate. An alternative view of minimal adequacy conditions on natural languages appears in Labov (1971) and Sankoff & Laberge (1973). These articles are the beginning of a series of papers within a functionalist paradigm, and have in common that they focus on the position of natural languages within the speech community, and their adequacy as systems of daily communication. This paradigm, which has centered on data from Tok Pisin, has stressed two points: (a)

(b)

There is no sudden break or qualitative jump between the Pidgin (non-native) and Creole (native) variaties of Tok Pisin, but rather an increase in complexity as the Pidgin develops. The differences which do occur are mostly stylistic, having to do with the amount of variability the system allows (Labov) or the amount of redundancy of the system (Sankoff & Laberge).

It is not clear to what extent the generative and the functionalist research paradigms are in conflict. The type of conditions on learnability formulated in generative grammar are of necessity quite abstract. Furthermore, it is not obvious that structural conditions on natural languages are necessarily distinct from functional ones. To give but one example, the emergence of ia bracketed relative clauses in Tok Pisin: [NPw[ s

ia]]

Sankoff & Brown argue that this construction is due to the specific discourse function of foregrounding in relative clauses (1976). On the other hand, one could claim that this development shows the generality of aa simple X' expansion rule as: X'"

->

X"

ia

where X ranges over N and V. The fact that Tok Pisin relative clauses are

Introduction

VII

optionally bracketed by two /a's would then be relevant from the point of view of discourse, but from the point of view of the syntax, relativization results from the interaction of several independent processes, among them the generation of X" determiners. A second contribution the study of Pidgin and Creole languages can make to the generative research program is in the domain of a theory of markedness. While most earlier theorizing on this point had assumed that Pidgins, given their reduced nature, represented unmarked systems, this idea is not tenable, and Bickerton (1975) launched the idea that Creole systems represent the unmarked case. If we believe that grammatical markedness develops in languages through lexical accretion, borrowing, the influence of factors of ease of speech perception and production, etc., then it is plausible that Young languages', such as recently emerged Creoles, represent the unmarked case. This idea is also problematic, however, in that Creoles can also be said to be 'mixed' languages, that is, they emerged in a multilingual contact situation. A case in point is verb serialization (cf. Schachter, 1974). While typologically one might distinguish between prepositional (or postpositional) languages and serializing languages, several Creole languages, e.g. Sranan Tongo, show both systems. For this reason, the idea that Creole languages constitute the unmarked case should be treated with caution; nonetheless, it can be a fruitful research strategy. The third contribution is in the field of variation. Many Creole languages have as oppressed languages undergone the influence of dominant languages spoken in the same area, but what makes the case of Creoles special is that these languages are often lexically closely related to the Creole in question. The result is that we find very complex speech communities, where a continuum of varieties is spoken ranging from the original Creole to a colonial standard. An example is Jamaica, but there are many more such cases (cf. DeCamp, 1971). The extreme variability which goes far beyond the lexical and the phonological, even in the range of speech styles of individual speakers, poses particular problems for generative grammar. Even if it were possible to abstract away from the variability with individual speakers, and focus on one single system, as Bailey tried to do in the case of Jamaican, how then can we relate the various individual systems to each other? Rather than claiming that the generative paradigm is inadequate, models can be developed which are plausible from the point of view of grammar as well as accounting for the variability in the data. In this volume the articles of Glenn Akers, Peter Mühlhäusler, Bill Washabaugh, Ellen Woolford, Hilda Koopman & Claire Lefebvre attempt to show how grammars vary and change, by providing and analyzing data from different historical and developmental stages of a number of languages.

VIII

Introduction

The papers of Jan Voorhoeve and Chris Corne stress the independence of Creole grammars from their European lexical ancestors. In the contributions of Hans den Besten, Hilda Koopman & Claire Lefebvre, Ellen Woolford, and Pieter Muysken we find attempts to substantiate the universal claims made in Bickerton (1975) and to redefine them in a generative framework: what universal characteristics do Creole languages have due to their being Creoles? In no way can this volume claim to summarize important work of the last ten years. Let us hope it marks a beginning. It is certainly the case that only on the base of a large amount of reliable data the generative study of Pidgins and Creoles will progress, as several authors remark in this volume, and most articles here do present a considerable amount of data. On the other hand, the data needed are defined by the theoretical questions asked, and we can hope that this volume contributes to those questions. BIBLIOGRAPHY Bailey, B., 1966. Jamaicain Creole Syntax. A Transformational Approach. Cambridge University Press, London. Bickerton, D., 1975. 'Creolization, Linguistic Universals, Natural Semantax and the Brain'. Presented at the Hawaii Conference on Pidgins and Creoles, Honolulu, Hawaii. DeCamp, D., 1971. 'Toward a generative analysis of a post-creole speech continuum'. In: D. Hymes (ed.), Pidginization and Creolization of Languages. Cambridge University Press, London. Labov, W., 1971. "The adequacy of natural language. I: The development of tense', manuscript. Naro, A., 1973. 'Creolization and Natural Change', manuscript. Roberts, P.A., 1975. "The Adequacy of Certain Linguistic Theories for Showing Important Relationships in a Creole Language'. Presented at the Hawaii Conference on Pidgins and Creoles, Honolulu, Hawaii. Sankoff, G. and S. Laberge, 1973. On the acquisition of native speakers by a language'. Kivung 6: 32-47. Sankoff, G. and P. Brown, 1976. 'The origins of syntax in discourse: a case study of Tok Pisin relatives.'Language 52: 631-666. Schachter, P., 1974. Ά non-transformational account of serial verbs.' Studies in African Linguistics, Supplement 5.

Table of Contents

Introduction 1. Admissibility conditions on final consonant clusters in the Jamaican continuum Glenn Akers

1

2. Multifunctionality as a derivational problem Jan Voorhoeve

25

3. The development of the category of number in Tok Pisin Peter Mühlhäusler

35

4. Pursuing Creole roots Bill Washabaugh

85

5. A re-evaluation of the predicate in Ile-de-France Creole Chris Come

103

6. The developing complementizer system in Tok Pisin Ellen Woolford

125

7. Marking WH-Movement in Afrikaans Hans den Besten

141

8. Creole tense/mood/aspect systems: the unmarked case? Pieter Muysken

181

9. Haitian Creole Pu Hilda Koopman and Claire Lefebvre

201

Addresses of contributors

223

Glenn Akers*

l. Admissibility conditions on final consonant clusters in the Jamaican continuum 1. INTRODUCTION

A prominent characteristic of English-based Creole languages is the absence of final consonants, particularly in clusters, in English derived lexical items. This paper describes the distribution of final clusters within the Jamaican continuum by first delimiting those clusters which are never reduced by any speaker.1 A series of grammatical stages are then postulated to account for the order of acquisition of clusters beginning with these clusters and ending with the full set of final clusters present in a fully-clustered variety of Standard Jamaican English. The distribution of clusters in the Jamaican continuum depends upon two sets of factors. One set of factors involves the phonological and morphological character of the clusters which determine the hierarchiallyordered pattern of cluster production. The second set of factors involves the conditions which promote or inhibit the retention of underlying final consonants in connected speech. These conditions include the presence or absence of stress, the phonological character of the following environment, and tempo factors. The present analysis assumes a simple rule of final consonant deletion, C ->· 0 / C # . The rates of application of the deletion rule are ordered according to the stages of cluster production. The deletion rule applies categorically to all clusters which are prohibited in the system at any given stage, while it applies variably to permitted clusters in accordance with a hierarchy of conditions on rule application.2

* The present paper is a revised version of Chapter 4 of my Ph.D thesis, referred to here as Akers 1977. The interested reader should consult my thesis for more complete details on the topics treated in this paper. The original research for this paper was supported by a research fellowship from the Organization of American States and by the Department of Linguistics, Harvard University, whose support provided for the research assistance of Ms. Pansy Ben during the period of data collection. I would especially like to thank Nick Clements for his advice and careful reading of several earlier versions of this paper. Of course, I alone am responsible for any inadequacies which may remain.

2

Glenn Akers

2. DATA The present investigation analyzes the occurrence of word final consonants in two-consonant clusters in the Jamaican continuum. The complete set of final two-consonant clusters found in standard American and British English contain one set of clusters which may occur in monomorphemic items, and another set of clusters which occur only through morphological affixization. The monomorphemic clusters investigated in this study are represented in Table 1.3 All were attested in the speech of at least one speaker. All of the clusters investigated occur in standard British English (Received Pronunciation) as represented by Gimson (1976:248). ks ps wp4 ns nz5 nt nd nc nj

'polities' 'perhaps' 'jump' 'dance' 'greens' 'elephant' 'hand' 'branch' Orange'

rjk sp st sk ft Ip Ib // Iv

'drink' Vasp' 'nest' 'desk' 'left' 'scalp' 'bulb' 'himself twelve'

Is lz6 It Id lcn //" Ik /m 8 In

'false' 'measles' 'salt' Old' 'belch' 'bulge' 'milk' 'palm' 'kiln'

Table 1. Standard English final two-consonant monomorphemic clusters. Bimorphemic clusters occur through affixization of the Z and D morphophonemes, or through affixization of . The bimorphemic clusters investigated in this study are represented in Table 2. mz ηζ ps fs os ts bz vz δζ

Barns' liangs' 'hips' 'cuffs' 'mouth's' 'goats' 'crabs' 'knives' 'sheathes'

dz gz β md rjd θt st ct

Voods' 'legs' 'fifth' 'screamed' 'banged' toothed' hashed' 'punched'

vd 8d Jd gd mQ ηθ ρθ ίθ

'curved' 'bathed' 'judged' Pegged' 'warmth' 'lenght' 'depth' 'eigth'

Table 2. Standard English final two-consonant bimorphemic clusters. The attested final two-consonant clusters are represented in Table 3.9 In Table 3 the consonants /, ή, η, ζ, δ, and g are omitted from the C2 posi-

Admissibility conditions on final consonant clusters C2: p

b

t

d

k

+

- + -

f

v

s

ζ

θ

s

c

f

m

n

-

-

-

-

-

Cl: p b

0 - 0 -

t

_ _ o - - - - + - + - - - - -

d k

- - -

£ /

--+ - - - + - - - - - __ + _ _ o - + - + - - - - -

v s

_ +

_ -

_ +

+ -

_ +

_

Θ δ

_ _

_ _

+ _

_ +

_ _

_ _

2

_ _ _

+

_

_

c / m

_ _ + _ _ +_ +

+ +

_ _ _ _ _ _ _ ( ) - _ _ _ - - _ _ _ o _ + __ + + _ _ _ o

y /

_ +

+ +

+ +

_ +

- 0 + -

_ +

0

-

- + - -

- - - - +

-

+

- + + + -

-

-

-

o + - - - 0 - - - -

_ + _ _ _

+

_

+

-

-

-

-

-

-

_ +

o _

_

_

_

_

_

_

_

_

_

_

_

_

+ +

+ + +

+

+

+

-

+

Table 3. Attested word final RP clusters: -C1C2# Note:

+ 0

indicates an attested cluster indicates a nonoccurring cluster indicates Cl = C 2

tion since they are not attested in any cluster. The cluster -mt tempt' is not listed by Gimson, nor is -ηθ. The former omission is an oversight, while the latter is a notational variant given by Gimson as -ηθ. The cluster -Is Welsh' is listed by Gimson, but was not investigated in the present study. Gimson lists -mf as occurring in RP, although only in triumph' and 'humph', but this cluster is restricted in the acrolect where it occurs only in literary forms.

4

Glenn A ken

3. A MODEL OF CONSONANT CLUSTER EXPANSION

3.1. Creole Clusters The model of linguistic competence for members of the Jamaican speech community must provide a basis for the production by each speaker of two codes, one of which corresponds to the individual's basilectal-modeled speech productions, and one of which corresponds to the individual's acrolectal-modeled speech productions.10 The total productions of Jamaican speakers may be arranged as a continuum of speech forms ranging from acrolectal Jamaican English to basilectal Jamaican Creole. The actual speech production of any given speaker may occur at any point along the continuum as determined by a combination of factors including the identity of the speaker's dominant code and his degree of control of the non-dominant code. Hyperforms result through generalization of the correspondences between the two codes when speakers incompletely control their non-dominant code. This view of a Creole continuum may be regarded as a bilingual model insofar as each speaker is regarded as controlling at least two codes, one of which is usually dominant, and for both polar codes there is internal variation among speakers who are dominant in that code. However, a Creole continuum differs from many bilingual situations insofar as the fundamental direction in historical evolution of the basilectal code is based upon acquisition of acrolectal properties at the expense of basilectal ones. Due to such cumulative accretions, the complete range of forms may be described synchronically by a series of discrete grammatical stages which begin with the basilect and end in the acrolect. It will be my goal to contruct a theory of consonant cluster production according to which the stages in acquisition can be arranged as a table of stages, in each of which a new group of clusters found in the acrolect is added to those already produced. The stages in the development of cluster production correspond to the implicational relations holding between cluster types, such that, if in a given stage we find a cluster of type n, we find clusters of type n-1 (i.e. of the type permitted at the immediately preceding stage) at the same or higher frequency. There are four stages of cluster production in the Creole versions of the data, as displayed in Table 4:

A dm issibility conditions on final consonant clusters Placeholders Stage 1 2 3 4 5

χ χ χ χ χ

Voice

χ x x x

Laterals

x x x

5 Voice

R

x x

x

Table 4. Stages of cluster production: Jamaican Creole monomorphemic clusters. The first stage includes the placeholder clusters, -mp, -If, -ns, -nc, and -nk, and spirant-final monomorphemic clusters -ks and -ps. The placeholder clusters are so named to emphasize the role of the set of final consonants, p, f, s, c, and k in the placeholders clusters. Each one of these consonants corresponds to one of the oral positions of articulation used in English, and there is one and only one sonorant-initial cluster per point of articulation. Each final placeholder consonant is voiceless. In any given articulatory position, the final consonant of the placeholder cluster is a continuant, if such a cluster exists in the acrolectal model, otherwise the final consonant is noncontinuant. Each placeholder cluster has an initial nasal sonorant segment if such a cluster exists in the acrolectal model, otherwise the initial segment is a lateral sonorant (the labiodental cluster -If). In the second stage, the final voiced versions of the placeholder clusters are added, if they exist in the acrolectal model. From the second stage on, only the placeholder clusters are affected, since -kt and -pi are not produced except in English versions of the data. Expanding the number of final two-consonant clusters to include the final-voiced versions of the placeholder clusters account for production of -Iv, -nz, -nj, while -mb and -ng are not found in the acrolectal model. In the third stage, the class of permissible clusters is expanded by adding lateral-initial versions of the placeholder clusters, which do not occur in stage 1, as well as the cluster composed of the sequence lateral plus alveolar stop. As with stage 1 clusters, the clusters added in stage 3 end in final voiceless consonants. In other words, the cluster added in stage 3 are the lateral-initial versions of the nasal-initial placeholder clusters -Ip, -Is, -Ic, and -Ik, as well as -It. This description of stage 3 shows that the set of final two-consonant clusters is expanded in two ways that it is restricted in stages 1 and 2. The primary expansion involves the addition of the lateral-initial versions of nasal-initial placeholder clusters. The secondary expansion involves the addition of a cluster ending in the alveolar position with a stop consonant, but only the lateral-initial version -It, is permitted, not the nasal-initial version -nt, as well. In other words,

6

Glenn Akers

a formal description of the clusters produced by stage 3 must explicity exclude -nt, but include all other clusters which have initial nasal and lateral segments and end in final voiceless consonants.11 In stage 4, parallel to stage 2, the set of clusters in the system is further expanded by including the final voiced versions of the stage 3 clusters, if they exist in the acrolectal model. In stage 4, the existing clusters -Ib, -Iz, -Id, and -// are produced, while -lg is not present in the acrolectal model. The clusters introduced in stage 4 do not typically occur in Creole productions due to the application of final consonant deletion, and also, in the case of -Ib, -Iz, and -If, due to the restricted number of English items containing these clusters. The stage 4 clusters may be added through advanced decreolization, and introduced into Creole productions. However, -nt, -nd, -sp, -st, -sk, -ft, -kt, -ηθ, -ηθ, -ΙΘ, -In, and -Im, constitute the remainder class R, and correspond to the remaining English monomorphemic clusters, which never occur in Creole productions. Although subsets of these clusters occur at different stages of decreolization when data from English versions is considered, from the viewpoint of the Creole, they are equivalent in that they are not part of the linguistic system. Table 5 shows the implicational ranking of speakers according to the percentage of cluster production in the Creole version of the data. The claim is that, if for any speaker we find a cluster of a given type then the Placeholders 1 (n-6)

Voice 2 (n=3)

Laterals 3 (n=5)

Voice 4 (n«4)

R 5 (n=7)

BC 6 (n-25)

(is)

loo

(so)

so

ι«

oe

PD-7

100

100

100

50

0

0

NJ-9

100

100

100

25

0

0

BP-8

1 0 0

1 0 0

8 0

0

0

0

LW-10

1 0 0

1 0 0

8 0

0

0

0

LS-3

1 0 0

1 0 0

8 0

0

0

0

DF-2

100

100

60

(25)

0

0

DG-5

100

(67)

60

(25)

(u)

0

CT-6

100

100

40

(25)

0

0

PT-4

1 0 0

1 0 0

4 0

0

0

0

Table 5. Implicational ranking of speakers according to percentage of cluster production in the Creole version of the data. Note: Circled cells are deviant. Coefficient of reproducibility:12 Colums 1-3 = 90%; 1-4 = 85%; 1-5 = 86%; 1-6 = 88.3%.

Admissibitity conditions on final consonant clusters

1

cluster-type occurring to its left on the table is found with the same or higher frequency. An ideally scaled implication table contains no violations of arithmetic order between any two adjacent cells. In the case of Table 5, an ideal scaling requires that the value of any cell, expressed as the percentage of occurrence, be greater than or equal to the value of all cells to the right of and below it, and less than or equal to the value of all cells to the left of and above it, for any speaker. 3.2. English Clusters The implicational array shown in Table 6, which represents the stages of cluster production in the English version of the data, may be taken to reflect the order of acquisition of acrolectal clusters by basilectal-dominant speakers.

Stage 1 2 3

4 5 6

CC

SC

χ χ χ

χ χ

χ χ χ

χ χ χ

CZ

CD

C

SS

χ χ χ χ

χ χ χ

χ x

x

Table 6. Stages of cluster production: Jamaican English Stage 1 core clusters (CC) correspond to the Creole clusters produced in stages 1, 2 and 3 of Table 4, as well as -nt, which is absent in basilectal Jamaican Creole, due to categorical deletion of the final consonant. With the exception of -ks and ps, the core clusters have sonorant-initial consonants. The secondary clusters (SC) introduced in stage 2 of Table 6 correspond to clusters produced in stage 4 of Table 4, as well as the remaining stop-final monomorphemic clusters -nd, -ft, -sp, -st, -sk, and -kt. The bimorphemic clusters (CZ) produced through affixization of the morphophoneme Z are added in stage 3. These clusters correspond to the expression of plurality, possession, and person-number agreement in the acrolectal system. The clusters introduced in stage 3 are -ps, -fs, -ts, -mz, -bz, -vz, -dz, -#z,and -gz. The bimorphemic clusters (CD) produced through affixization of the morphopheneme D are added in stage 4. These clusters arise in the acrolectal system through regular past marking of verbs, as in the past tense

8

Glenn Akers

form of Vash' in 'Lizzy washed the car', and in forming derived adjectives, as in The curved road'. The clusters introduced in stage 4 are -si, -ct, -md, -bd, -vd, -jd, -rjd, and -gd. The introduction of bimorphemic spirant-final clusters prior to bimorphemic stop-final clusters in stages 3 and 4 is parallel to the favored presence of monomorphemic spirant-final clusters -ns, -Is, and -ks in Creole productions over monomorphemic stop-final clusters -nt, -It, and -kt. Introduction of final monomorphemic spirant clusters prior to final monomorphemic stop clusters is a phonological condition of Creole. This condition also provides a phonological motivation for the introduction of final bimorphemic spirant clusters prior to final bimorphemic stop clusters. The absence of non-strident fricatives in Jamaican Cerole, and their marginal occurrence in Jamaican English, is reflected by the introduction of clusters (Co) containing these segments in stage 5. Parallel to the major constraining properties of stage 1, production of monomorphemic clusters -ηθ and -ΙΘ is favored more than production of the bimorphemic clusters -mQ, -ρθ, -β, -ίθ, -ηθ, -0s, -θί, -θζ, and -6d. The least favored clusters, which consist of sonorant sequences (SS) -Im and -In, are introduced in stage 6. Table 7 shows the implicational ranking of speakers according to the percentage of cluster production in the English version of the data. Both Table 5 and Table 7 provide support for the model of cluster production outlined in this section. Table 8 summarizes the Creole and English stages, showing which clusters are admitted in each stage.

4. THE FORMAL TREATMENT OF STAGES

Two alternative formulations for the stages of cluster production are presented in the following sections. One formulation is the traditional treatment, which describes the morpheme structure conditions to which the occuring clusters conform at each stage. The other formulation is my alternative treatment, which- describes admissibility conditions to which clusters must conform at each stage. As will be shown, these two treatments differ in at least one essential way. Morpheme structure conditions characterizing any given stage describe the conditions which must be met by clusters introduced at that stage and all previous stages. Admissibility conditions, on the other hand, describe the cluster-class introduced at each stage and show how any cluster-class relates to the cluster-classes of previous stages. First, it will be shown that the description of the morpheme structure constraints on each stage fails to provide a clear characterization of how the data of each stage relates to the other. Then my alternative treatment,

Admissibility conditions on final consonant clusters Cluster Type CC

SC

CZ

CD

ce

SS

(n=15)

(n-9)

(n=9)

(n-8)

(n-ll)

(n-2)

PD- 7

100

(89)

©

88

82

50

NJ- 9

100

(89)

100

88

©

50

DG- 5

100

100

100

88

©

0

BP- 8

100

100

©

75

45

0

CT- 6

100

100

100

63

27

0

LW-10

100

100

100

©

18

0

RM- 1

100

89

78

25

0

0

DF- 2

100

78

78

13

0

0

PT- 4

100

Ϊ7

67

©

0

0

LS- 3

93

22

11

0

(T)

0

Speaker

Table 7. Implicational ranking of speakers according to percentage of cluster production in the English version of the data. Note: Circled cells are deviant. Coefficient of reproducibility = 85%.

Creole Stage

1 2 3 4

mp

If h

nc nj Ic U

ns nz

IP

ls.lt lz,ld

Ib

qk

ps, ks

Ik

English Stage

1 2 3 4 5

nt sp mz, bz md, bd ηιθ,ρθ

ft

nd,st

fs,vz

nz, ts, dz

vd fl

ηθ,ΙΘ

9s, ίθ, dt δζ,δα

6

Im

In

Table 8. Clusters present according to stage

sk gz s et, jd, yd, gd st, ηθ

pt,kt

10

Glenn Akers

formalized in terms of admissibility conditions, will be shown to provide a simpler characterization of the data. 4.1. Morpheme Structure Conditions The set of morpheme structure conditions, henceforth MSC, is a formal representation of the clusters permitted in each grammar stage described for Creole and English in section 3. It will be shown that although the MSC are empirically adequate descriptions of discrete stages, they do not provide a clear description of the underlying processes by which the set of occurring clusters is expanded in each succeeding stage. The morpheme structure conditions given in this section are stated as if-then propositions. The six morpheme structure conditions required to permit only the Creole stage 1 clusters are summarized below. MSC-0 requires final two-consonant clusters to end in obstruents. MSC-0, which serves to exclude final consonant sequences ending in sonorants (-Im, -In), remains unchanged throughout all Creole and English stages until the final stage of English in which it is lost. MSC-1 permits homorganic nasal clusters with final consonants which are anterior, coronal, and continuant; and clusters with final consonants which are either non-anterior or non-coronal and non-continuant. The Stage 1 MSC-0:

[+consonant]

[-syllabic]+ [-sonorant]

MSC-1:

+nasal

+anterior

α place

+coronal -anterior -coronal

4

α place

(

a-»-c

Admissibility conditions on final consonant clusters [-syllabic]+ Ψ [+continuant] [-syllabic]*

MSC-2:

-nasal +consonant

MSC-3:

[+lateral]

MSC-4:

- continuant! -nasal J

[-syllabic]*

-coronal -voice [+consonant]

+coronal] +anterior] [-syllabic]*

MSC-5:

11

I [-voice]

combined effect of MSC-1 and MSC-5 is to permit all nasal-initial and final-voiceless sequences found in the acrolectal model except for -nt and -mf. MSC-2 requires the nonnasal-initial clusters to end in final continuants. The combined effect of MSC-2 and MSC-5 is to permit only final voiceless sequences, -// -ks, and -ps. MSC-3 requires the lateral-initial cluster to end in a labial. Since MSC-2 requires the final to be a continuant, MSC-2, MSC-3, and MSC-5 uniquely specify the cluster-//. MSC-4 requires the nonnasal, noncontinuant-initial clusters to have initial voiceless noncoronal consonants and terminate in coronal, anterior consonants. Since MSC-2 requires the final consonants to be continuant, MSC-2, MSC-4, and MSC-5 uniquely specify the clusters -ks and -ps. MSC-5 requires all clusters in the system at this stage to end in voiceless segments. As mentioned above, MSC-5 interacts with MSC-1, MSC-2, MSC-3, and MSC-4, in order to correctly restrict the set of permissible clusters. It is a striking observation that six morpheme structure conditions are required to describe only seven different clusters. This is evidently an uneconomical way to represent stage 1 consonant cluster productions. The morpheme structure conditions required to describe the clusters permitted in the system at stage 2 are summarized below. Stage 2 MSC-0: MSC-1: MSC-2:

same as in stage 1 same as in stage 1 same as in stage 1

12

Glenn Akers MSC-3: MSC-4':

same as in stage 1 Γ- continuant"! [-syllabic]+ [-nasal J

E

I

-coronafl -voice J

MSC-5:

l

+coronal +anterior -voice

lost

In stage 2, MSC-5 is lost, so that the final voiced versions of the placeholder clusters, -Iv, -nz, and -«/, are admitted to the system. MSC-0, MSC-1, MSC-2, and MSC-3 are the same as in stage 1, while MSC-4 is complicated to MSC-4' by addition of the lost restriction that final elements be voiceless, so that -kz and -pz are still prohibited. In stage 2 only the final voiced versions of the placeholder clusters are acquired, while the formal statement describing the complete set of clusters produced in stage 2 also requires the complication of MSC-4 to MSC-4' by addition of the restriction that final consonants be voiceless. The morpheme structure conditions required to describe the clusters permitted m the system at stage 3 are summarized below. Stage 3 same as in stage 1 same as in stage 1 lost [+lateral]

MSC-0: MSC-1: MSC-2: MSC-3':

ΓΙ-labiall . +voice _ Φ b d conditions: a ->· b MSC-4'

Γ-continuantl 1 -nasal J

[-syllabic]+ 1

f

+ continuant + coronal + anterior _- voice

i

-coronal -voice

1

Admissibility conditions on final consonant clusters

13

The morpheme structure conditions are further generalized in stage 3 by loss of MSC-2, which requires nonnasal-initial sequences to have final continuant segments. MSC-3 is complicated to MSC-3' in order to permit sequences of lateral plus nonlabial voiceless consonants. This condition permits sequences of lateral plus voiceless segments, -Ip, -Is, -Ic, -Ik, and -It, as well as the sequence lateral plus voiced labial continuant, -Iv. Prohibited by MSC-3' is the sequence of lateral plus voiced noncontinuant labial, -Ib. The loss of MSC-2, which requires nonnasalinitial segments to end in final continuant segments, results in the complication of MSC-3 to MSC-3' and MSC-4' to MSC-4'' by addition of the continuancy restriction to both MSC-3 and MSC-4'. The morpheme structure condition approach to the description of consonant cluster stages fails to capture, in its formalism, the fact that a single restriction — that the final member of oral-initial clusters must be a continuant — has been loosened at this stage. Its description of this generalization, which admits the new clusters -It, -Ip, -Ik, is formally arbitrary. The only difference between the sets of MSCs required for stages 3 and 4 is that one less MSC is found in stage 4. The loss of MSC-3' in stage 4 permits the occurrence of all lateral plus obstruent sequences, and thus admits -Ib, -Id, and -/;'. The morpheme structure conditions required to describe the clusters permitted in the system at stage 4 are summarized below.

MSC-0: MSC-1: MSC-3': MSC-4'':

Stage 4 same as in stage 1 same as in stage 1 lost same as in stage 3

While the MSCs uniquely describe the set of clusters produced at each stage, the formal relationships between successive MSCs, such as that obtaining between MSC-3 and MSC-3' and MSC-4 and MSC-4 do not reflect the underlying principles which account for the successive stages in the expansion of occurring clusters. Specifically, they fail to reflect the fact that each successive stage is a generalization of the earlier stage, in the sense that new clusters, in addition to all the previously allowed clusters, are permitted. 4.2. Admissibility Conditions 4.2.1. Creole Stages In the present section, an alternative to the morpheme structure condition

14

Glenn Akers

treatment of the stages in cluster production is given, by formalizing a set of admissibility conditions which describe the subset of clusters added to the system at each given stage from the total set of clusters found in the acrolectal model. The full set of clusters found at any given stage is determined by the admissibility conditions for that stage and for all previous stages. It will be shown in this section that a description of the admissibility conditions for each stage clearly indicates the significant relationship between the stages of cluster production. The model of cluster production underlying the admissibility condition treatment assumes point of articulation as a primitive notion, since every stage involves adding clusters present in the acrolectal model at each possible point of articulation. The present model of cluster production accounts for a continuum of speech forms in terms of individual differences in learning the implicationally-related admissibility conditions. In the formal description of the admissibility conditions, henceforth ACs, the parenthesis notation indicates that the parenthesized feature is chosen if the resulting cluster exists in the acrolectal model, otherwise the feature enclosed by parenthesis is suppressed, and the clusters described by the remaining features are selected. The negation of a feature in the rule conditions indicates the absence of that condition, and not the opposite value of the feature. The ACs function as a filter in permitting only the acrolectal clusters which satisfy the ACs at any given stage. The ACs are cumulative in function, such that the clusters produced at any given stage are permitted by the ACs for that stage and all previous stages. Two ACs are required to specify the clusters permitted in stage 1. One condition permits the group of placeholder clusters, and the other condition permits -ks and -ps.13 The placeholder clusters are permitted by AC-1. The role .of the subscript c will become apparent below. AC-1:

+consonant" +sonorant (+nasal) [+lateral]

-sonorant -voice (+continuance

AC-1 permits the nasal-initial placeholder clusters through selection of the feature +nasal, and it permits the lateral-initial placeholder cluster through selection of the feature +lateral. AC-1 can be read: Admit the clusters present in the acrolectal model which have initial sonorant, consonant (and nasal, if possible, lateral, if not) segments and final nonsonorant, voiceless (and continuant, if possible) segments for each oral articulatory position, labial, labiodental, alveolar, alveopalatal, and velar,

Admissibility conditions on final consonant clusters

15

found in the inventory of acrolectal consonants. The final alveolar continuant cluster is permitted by full expansion of the second feature matrix, as well as the labiodental and alveopalatal-final clusters, and final labial and velar stop clusters are permitted by suppression of the feature +continuant. Note that there need be no special mention of homorganicity for the nasal-initial clusters, since all such clusters in the acrolectal model (in respect to which the ACs act as a filter) are homorganic in any case. According to AC-1, -If is selected as the labiodental placeholder by selection of the features +lateral and -(-continuant, since the nasal-initial cluster -mfis not present in the acrolectal model. AC-1 admits -rjk as the velar placeholder through selection of the feature + nasal and suppression of the feature -(-continuant, since a final velar continuant cluster is not present in the acrolectal model, but the nasal-initial cluster with a final noncontinuant segment is present in the acrolectal model. The stage 1 stop-initial clusters -ks and -ps are described by AC-l'. The condition that each member of the cluster be voiceless need not be specifically stated, there being no monomorphemic clusters -gz and -bz in the acrolectal model. AC-1':

-sonorant -continuant -coronal

"-sonorant + continuant +coronal + anterior

The clusters added in stage 2 are described by a single feature change of AC-1, which permits the final voiced versions of the stage 1 placeholder clusters to occur as well. The AC corresponding to stage 2, AC-2, may be derived from AC-1 simply by changing the specification -voice found in AC-2 to +voice. AC-2:

[+voicej Recall that since the ACs are cumulative in function, the net effect of AC- and AC-1 and AC-2 is to permit clusters of the form: +consonant r (+ nasal) ι I [+lateral]'

-sonorant 1 (+continuant)!

The clusters added in stage 3 are described by AC-3, which permits clusters that have initial lateral segments and final voiceless segments (stops or

16

Glenn Akers

continuants). This set of clusters corresponds to the lateral-initial initial versions of the placeholder clusters and -It. The latter cluster is permitted by AC-3 through deletion of the parenthesized feature (+continuance of AC-1 in the second feature matrix of AC-1.14 The total set of clusters permitted in stage 3 are therefore those which meet the description of either AC-1, AC-l', AC-2, oiAC-3. AC-3:

-voice>

a-»b AC-3 is related to AC-1 by deletion of the parenthesized features (+nasal) and (+continuant) of AC-1. The full set of dusters (ignoring -ks, -ps) permitted by stage 3 could be described as follows: +consonant

-sonorant +labial t

+continuant -voice d

b-*c By comparing this statement with AC-3 we see that assigning the ACs a cumulative function allows us to achieve considerable simplification.15 The clusters added in stage 4 are described by changing a single feature of AC-3 in order to permit the final voiced versions of the clusters added in stage 3 to occur. AC-4, which describes the group of clusters added in stage 4, is derived from AC-3 simply be replacing the specification -voice found in AC-3 with + voice: AC-4 +voice

A summary of the ACs which describe the cluster added in stages 1-4 of Jamaican Creole is given below.16

Admissibility conditions on final consonant clusters

17

STAGE 1

AC-1:

+ consonant -sonorant AC-1' +sonorant -voice (+ nasal) (+continuant)c

-sonorant -sonorant -continuant +continuant •coronal +coronal +anterior

[+lateral] STAGE 2 AC-2:

[. ..]

+voice STAGE 3

AC-3:

b a-»-b

STAGE 4

AC-4:

[...] I+voicej

The treatment of the acquisition of consonant clusters according to ACs provides a simple, illuminating treatment of how each stage is a generalization of an earlier one. The ACs serve to enumerate the set of clusters permitted in the system at any given stage. The ACs provide a direct synchronic analogue of the successive historical processes by which, it may be hypothesized, new acrolectal clusters were acquired in the development of Jamaican Creole. Thus, the present analysis provides a characterization of the notion linguistic change in progress', explaining how synchronic grammars reflect their historical sources in a Creole situation. 4.2.2. English Stages The admissibility conditions determine which clusters present in a speaker's English representations are produced in Creole. The production of English clusters in Creole and English versions of the data can be viewed as reflecting a single line of historical development involving successive generalization of the class of permitted clusters. Inspection of the English versions of the data reveals that the first stage of English cluster production involves

18

Glenn Akers

further generalization of the admissibility conditions in order to permit speakers to produce all nasal or lateral sonorant plus voiceless obstruent clusters present in their English representations. There is one potential problem with the present view. The data indicates that -nt must be excluded from Creole stages 3 and 4, since it is never produced; but in the English version of the data, it occurs at a higher rate than do the stage 4 clusters. If -nt occurs in underlying English representations for all speakers, then its absence in Creole is accounted for by the fact that it is not admitted by any of the conditions. On the other hand, the Creole stage 4 clusters ~lb, -Iz, -If, and -Id cannot occur in the underlying lexical representations for all speakers, since there are speakers who do not produce them even in their English productions. If they do not occur in underlying lexical representations for a speaker, then AC-4 cannot, of course, require them to be produced. Therefore, if in stage 5 the class of permitted clusters is generalized to include all sonorant plus obstruent sequences, we expect speakers to produce -nt, but not, for instance, -Id, if -Id does not occur as an underlying cluster in their grammars. Accordingly, we can describe stage 1 English productions (see Table 6 of section 3) by AC-5.

AC-5:

+consonant + sonorant j+nasal I lateral I

Γ-sonorantΙ [_- voice J

Stage 2 in the production of English clusters results from further generalization of the class of permitted clusters by the addition of the remaining monomorphemic clusters excluded by the previous stage. The clusters added in English stage 2, other than -nd, have initial obstruent segments, so the resulting set of clusters produced by English stage 2 includes all consonant plus obstruent sequences (except -rC clusters; see note 5) present in speakers' lexical representations. Stage 2 English productions are permitted by AC-6.

AC-6:

[+consonant]

[-sonorant] £

+nasal i v +latera1

\> a-»-b

The grammar of stage 3 English productions permits the monomorphemic

Admissibility conditions on final consonant clusters

19

clusters associated with stages 1 and 2, as well as the acrolectal bimorphemic clusters produced through affixation of the morphophoneme -Z as the plural, possessive, and person-number agreement marker. Whereas stages 1 and 2 permit monomorphemic clusters confirming to admissibility conditions, stage 3 clusters require acquisition of the acrolectal morphological systems which are marked by the morphophoneme -Z.17 Since the final slit strident alveolar spirant -s is not prohibited in Patois, the occurrence of CZ clusters may be accounted for as grammatically-determined variable insertion of the acrolectal marker, rather than through its variable deletion. Stage 4 in the grammar of English productions permits bimorphemic clusters produced by affixation of the morphopheme -D as the acrolectal marker of past tense. The production of stage 4 clusters requires the acquisition of the acrolectal morphological system marking past tense by the morphophoneme -D.ls Stage 5 in the grammar of English productions permits clusters found in stages 1-4, as well as monomorphemic and bimorphemic clusters containing the alveolar nonstrident groove spirants θ and δ. Stage 5 clusters require two additional members in the inventory of phonological segments. For basilectal-dominant speakers, a lexically restricted phonological rule maps the appropriate occurrence of alveolar stop segments, t and d, into the acrolectal segments θ and δ. Acrolectal-dominant speakers with θ and δ in lexical representations may apply the converse rule, Θ, δ ->· t, d, in producing Creole forms.19 Stage 6 in the grammar of English productions corresponds to an acrolectal grammar which permits final sequences of sonorant consonants. Stage 6 of English cluster production results from the complete generalization of the admissibility conditions of the cluster classes present in English. The clusters added in stage 6 end in sonorant sequences, so that the generalization, formalized as AC- 7, permits all acrolectal clusters. AC-7:

[+ consonant] [+consonant]

The stages of cluster production for Creole and English versions of the data are summarized in Table 9. The stages outlined above for the grammar of English productions account for the implicational order found to obtain between the types of final twoconsonant clusters present in the acrolectal model. For the English versions, the stages in the expansion of cluster production begin with the generalization of the admissibility conditions on Jamaican Creole clusters found in stages 1-4. Each successive stage of English grammar production expands the number of clusters either through further generalization of

20

Glenn Akers CREOLE

STAGE l AC-1:

+consonant +sonorant j(+nasal) \ l[+lateral]J

-sonorant -voice (+ continuant) c

AC-1':

-sonorant -continuant -coronal

STAGE 2 AC-2:

[...]

|_+ voice STAGE 3

AC-3:



a-»-b

"c

STAGE 4 AC-4:

(...}

[+voicej STAGE l

AC-5:

AC-6:

GC-Z GC-D PRAC-7

ENGLISH

+consonant Γ-sonorant] [-voice J +sonorant r+nasal \ l+laterali STAGE 2 [+consonant]' [-sonorant] a r+nasal i UlateralJ b

STAGE 3 + plural -»· -Z STAGE 4 +past -»· -D STAGE 5 t5d-»0,5 STAGE 6 [+consonant] [+consonant]

Table 9. Conditions on stages of cluster production

-sonorant + continuant + coronal +anterior

Admissibility conditions on final consonant clusters

21

the admissibility conditions, as in English stages 1,2, and 6, or through acquisitions of acrolectal grammatical systems, as in stages 3 and 4, or by addition to the phonological segment inventory, as in stage 5. In each successive stage, a new group of clusters is added to the set of clusters produced in all preceding stages following the implicational order outlined above. 5. CONCLUSION

This paper purposes a model for the distribution of final consonant clusters which formalizes the conditions clusters must meet at each stage in cluster expansion, beginning with a restricted set of basilectal clusters and ending with the full set of acrolectal clusters. The present model utilizes a set of admissibility conditions, which describe the cluster-classes introduced at each stage and show how any cluster-class relates to the cluster-classes of previous stages. Point of articulation is regarded as a primitive notion, since clusters are added at each possible position. Thus, the initial set of placeholder clusters is described in relationship to the full set of acrolectal clusters, in such a way that only the acrolectal clusters which are permitted by the admissibility conditions occur at any given stage. It is further shown that the admissibility conditions provide a more economical description of the clusters admitted in each stage than do morpheme structure conditions. This is due to the fact that morpheme structure conditions must enumerate the complete set of clusters permit ed at each stage. Admissibility conditions, however, describe the clusters introduced in each stage in terms of previous stages. Therefore, the effect of the admissibility condition is cumulative, and clusters permitted in a prior stage need not be mentioned in a succeeding stage. The present model of cluster production accounts for the continuum partly in terms of individual differences in learning the implicationally ordered admissibility conditions.20 The asymmetry between perception and production is reflected in this model by stating the admissibility conditions (which account for the production of clusters) in terms of the full set of acrolectal clusters. The model presented in this paper is supported by the implicational ranking of speakers according to cluster-type for both versions of the data. The ranking of speakers for the English version of the data reflects the relative degree to which speakers produce acrolectal properties. Analysis of this type for the remaining linguistic variables in Jamaican speech will serve to further delimit the notion 'linguistic continuum'. The present paper may be regarded as a first step towards a characterization of this notion.

22

Glenn Akers

1. The uninitiated reader may wish to consult De Camp 1971, Bickerton 1973, and 1975 for background about the significance of Creole continuum for general linguistics. The present description is based on data collected from a sample of ten speakers representative of the full linguistic range found in the Jamaican continuum. The sample was constructed in such a way as to reflect the actual distribution of the population in terms of social class membership. The data considered in this description consists of a Creole and an English elicitation from each speaker for every cluster investigated. For further details on the social characteristics of each speaker and the methodology followed, see Appendix 1 and 3 of Akers 1977. 2. For detailed discussion of the conditions effecting the application of consonant deletion, the interested reader should consult chapter 4.2 of Akers 1977 and the references cited there. 3. The consonant clusters treated in the present paper do not include glide + consonant sequences, nor -rC clusters. An analysis of -rC clusters is given in Chapter 5 of Akers 1977. That analysis shows that while the distribution of -rC clusters is accounted for by a different set of environmental conditions then the clusters treated herein, the same type of grammatical conditions, i.e. admissibility conditions introduced below in section 4.2, may be used to describe the distribution of -rC clusters as well. 4. Nasal plus voiceless obstruent clusters are included in the present investigation even though phonetically these clusters may be realized as nasalized plus voiceless obstruent sequences. Cf. Bailey (1973:216) 5. The Patois value for -nz is based upon the speakers' regular production anz for the English item 'ant'. On the other hand, the English value for -nz is based upon the production for the item 'greens', a collective noun meaning vegetable. 6. The form 'measles' was included in the questionnaire since it occurs with singular verb forms. However, a bimorphemic two-cluster item such as 'tails' would have been more appropriate for the present purpose since "measles' contains the final sequence -zlz, and a higher rate of reduction may be found than would be the case for a bimorphemic -Iz cluster. 7. The variant bei s 'belch' is not attested in the present data. 8. The sequence -Im in 'palm' occurs as a refading pronounciation in monitored productions, Furthermore, Jones (1967:272) notes that people who are concerned with the working of kilns produce kil 'kiln'. 9. Adopted from Gimson (1976:248). The variants flftd fifth', tu:8d 'toothed', wo:mp6 Varmth', and lerjkQ 'lenght' are not attested in the present data. 10. The terms basilect and acrolect were introduced by Stewart (1965). 11. Note that if -nt is formally included in stage 3, the description of permissible clusters is simplified. Prior to stage 3, there is strict prohibition against alveolar stopfinal clusters. Although occurrence of alveolar stop-final clusters are permitted by stage 3, their production is conditioned by environmental factors including stress and the nature of the following environment. If -nt is formally permitted by stage 3, its absence in speech is accounted for by categorical application of final consonant deletion, which applies variably to -//. 12. A deviant cell violates the implicational pattern. The coefficient of reproducibility of an implicational table in computed as the percentage of non-deviant cells in the table. A value of 85% or more is usually accepted as confirming the validity of the ordering. See Guttman (1944), Torgerson (1957), and Matalon (1965) for detailed treatments of implicational analysis. 13. In interpreting the formalism for admissibility conditions, recall that the conventions are somewhat different from those used in stating phonological processes. See the paragraph above.

Admissibility conditions on final consonant clusters

23

14. Note that the formal interpretation of the negation of a condition of the form (± feature) is the absence of that condition when the AC describes the class of consonants added in a given stage. 15. The conflated statement given above is an abbreviation for the three feature matrices given below: [+consonant"! 1+nasal I

Γ-sonorant ~[ k+ continuant)

1+consonant! |+lateral J

-sonorant +labial +continuant

R-consonantl klateral

Γ-sonorant -voice

16. Note that AC-1' is present in each stage. 17. The order of acquisition of the acrolectal system for marking plurality, possession, and third person-number agreement is not investigated in the present study. See Wolfram (1969) for a discussion of their acquisition in Detroit Black English. 18. Bickerton (1975) contains a detailed analysis of the acquisition of tense marking in the Guyanese continuum. 19. This analysis is supported by Robson (1975:286-289). 20. Individual differences also occur in underlying representations and in the adaptation rules which derive the nondominant code.

BIBLIOGRAPHY Akers, Glenn A. (1977), Phonological variation in the Jamaican continuum, Ph.D. thesis, Harvard University. Bailey, Charles-James N. (1973), Variation resulting from different rule orderings in English phonology. In: Bailey, C.-J. N. and Shuy, Roger (eds.), New ways of analyzing variation in English, Washington, Georgetown University Press, pp. 211-252. Bickerton, Derek (1973), The nature of a creole continuum, Language 49,640-669 Bickerton, Derek (1975), Dynamics of a Creole system, Cambridge, Cambridge University Press. DeCamp, David (1971), Analysis of a post-creole speech continuum. In: Hymes, Dell (ed.), Pidginization and creolization of languages, Cambridge, Cambridge University Press, pp. 349-370. Gimson, A.C. (1976), An introduction to the pronunciation of English, 2nd edition, London, Edward Arnold. Guttman, Louis (1944), A basis for scaling qualitative data, American Sociological Review 9, 139-150. Jones, Daniel (1967), English pronouncing dictionary, edited by Gimson, A.C., Thirteenth edition, London, J M . Dent and Sons. Matalon, Benjamin (1965), L 'analyse hierarchique, Paris, Gauthier-Villars. Robson, Barbara (1975), Jenepher revisted: adult language change. In: Shuy, Roger W. (ed.), Analyzing variation in language, Washington, Georgetown University Press, pp. 283-290.

24

Glenn Akers

Stewart, WA. (1965), Urban negro speech; sociolinguistic factors affecting English teaching. In: Shuy, Roger W. (ed.), Social dialects and language learning, Champaign, HI., National Council of Teachres of English, pp. 10-18. Torgerson, Warren (1957), Theory and methods of scaling, New York, Wiley. Wolfram, Walter A. (1969), A sociolinguistic description of Detroit negro speech, Washington, Center for Applied Linguistics.

Jan Voorhoeve*

2. Multifunctionality as a derivational problem

1. INTRODUCTION

The loss of morphology in Pidgins and Creoles must have affected in the first place those derivational affixes which did not convey independent semantic information. I assume that the distinction between /// and illness can be expressed completely in terms of major categories. What, in that case, is the use of the affix -ness, given the fact that syntactic position and other dependency relations keep both items apart? Subsequent grammatical expansion in a creolization process does not offer incentives to create semantically empty affixes. This may explain why multifunctionality seems to be more frequent and more regular in Pidgins and Creoles than in the model language.1 Multifunctionality offers a descriptive problem. This was mentioned in Voorhoeve (1961:47). Mühlhäusler (1974:119) discussed the problem in greater detail and remarked: 'An adequate treatment of this phenomenon is one of the tasks now facing grammatical theory.' He took up this task himself a year later in an interesting contribution on multifunctionality in New Guinea Pidgin (Mühlhäusler 1975). I refer to these studies for bibliographical references. In short the problem is as follows: [viewjy and [viewJN have so much in common that it seems to be superfluous to include both words fully specified in the lexicon. Formally they are identical and it seems that the semantic distinction can be expressed by the semantic correlate of the categorical symbols V (verb) and N (noun). One can avoid duplication in the lexicon by interpreting multifunctionality as the result of a derivational process. The peculiarity would be that this derivation does not use any affixes. This solution is mentioned in Jackendoff (1974:661) as an explanation for the different functions of the wordform smoke and their interrelations: * This paper was given on the Conference on Theoretical Orientations in Creole Studies at St Thomas (March 28 - April 1,1979). An abridged version was presented at the Morfologiedag (April 11,1979) at Leiden and will be published in GLOT.

26

Jan Voorhoeve smoke! smoke2 smoke3 smoke4 smokes smoke6

+N. +Vjnt, GIVE OFF SMOKE ('the cigar/chimney smokes'). With the same relation to smoke! as steam, smell, piss, flower, signal. +Vtr ('John smokes the cigar'). With the same relation to smokej zsmilk. +V^, PERMEATE OR COVER WITH SMOKE ('John smoked the ham'). With the same relation to smoke1 as paint, steam, water (a garden), powder (your nose), flour, cover. +Vjnt ('John is smoking'). With the same relation to smoke3 as the corresponding use of verbs like eat, draw, read, cook, sing. +N, SOMETHING THAT IS SMOKED ('a cigar is a good smoke'). With the same relation to smoke3 as the nouns drink, desire, wish, dream, find, experience.

The conclusion (p. 662) is phrased as follows: We thus see a rich variety of partial regularities in lexical relations: their expression in a transformational theory becomes hard to conceive, but they can be expressed quite straightforwardly in the lexicalist framework.' I think that Jackendoff and others proved their point that the lexicalist position presents definite advantages. My problem is rather to know in what way lexical relations are regular. The phrase 'partial regularities' is rather vague. I think this vagueness is related to the fact that Jackendoff does not admit subcategories in his redundancy rules. I have the feeling that subcategories would enable him to capture more regularities. I would like to suppose that for this same reason Jackendoff is forced to avoid the problem of the direction of the derivation. The fact that all relations in the preceding uses of the different items smoke are expressed with regard to smoke 1 and smoke3 suggest a different conclusion. But let us consider this problem by studying the relation to causative and noncausative break (p. 659):

χ +V +[ΝΡι-] NPj W

«*· χ +V +[ΝΡ 2 -ΝΡχ] NP2 CAUSE (NPj W)

I do not regard the conflation of the morphological and semantic component as essential. More important is the double arrow, which means that a derivational direction is avoided. Aronoff 1976 uses a direction of derivation, which is convertible, in which case the rule functions as an interpretive rule. I have the impression that the avoidance of a derivational

Multifunctionality as a derivational problem

27

direction is related to a fear for irregularities. In this case one would not like to produce a causative verb *die in stead of the correct verb kill. This could, however, be avoided, if the rule contains the subcategorisation [+stative] in the leftmost member, which would also exclude die to produce a causative verb. Another hypothesis is implicit in Jackendoff s work. Jackendoff is convinced that the relations between lexical elements are based on so much chance (p.ex. the existence of a causative verb kill) that directional rules would make necessary the use of a filter to filter out undesired results. I agree with him that a filter should be avoided. But I think that a proper subcategorisation acts as a filter. I assume therefore that the relations between lexical elements are so regular that one indeed can use directional derivations. This hypothesis is a stronger one, because it can easily be falsified. Multifunctionality is the proper field for a study of this problem. In multifunctional sets the direction of the derivation cannot be deduced from the morphological structure. In a case like [view]y and [re-view]v the direction is given, because one assumes the simple form to be the basis of the derivation. Thus: view -*· re-view. Even in the case of [re-view-er]N one can say that the less complex form is the basis of the derivation: [re-view]v -*· [re-view-er]^. Moreover, there is often a semantic correlate which may indicate the direction: re-view-er would be 'somebody who reviews'. The semantic relation between [ill]A and [illness]N seems to be completely defined by the categorical symbols A (adjective) and N (noun), whichever way one would like to describe them semantically. But in that case one can determine at least the direction of the derivation from the measure of complexity of the words involved. If one is unable to determine the derivational direction, either on the grounds of complexity or on the grounds of a semantic change, one is faced by a precarious question. The problem could be avoided by assuming uncategorised semantic units that obtain a function in the sentence with the semantic correlate of this function. There would then be a semantic unit [view] which in one case would have a semantic element ACTION added to it and in the other case a semantic element EXISTENCE. This solution seems rather unapealing, because restrictions would become unpredictable. Why can [view] easily opperate as a noun whereas [read] can't? A related solution via complex categorical symbols as VN (verb + noun) as proposed by Hockett (1959:226) increases the number of categories and by enumeration abandons the possibility of an explanation.

28

Jan Voorhoeve

2. THE DERIVATIONAL BASIS

The first question we must ask ourselves is whether multifunctionality in Sranan is random or governed by rules. In other words: Does every word have its own individual pattern of multifunctionality so that the lexicon must indicate the functional valence for every word, or is it possible to formulate rules that produce the multifunctionality? Should we include words in the lexicon as [view]v>N next to other words as [read]y, or should one rather expect rules as [ ]y -*· [ ]N (t° ^e rewritten later as [V] -> [N]) together with conditions that exclude an element as [read]y from this rule. Obviously the linguist will only resort to enumeration when rules prove to be impossible. But rules are only possible when the derivational direction can be indicated. Because, in the case of multifunctionality , the direction cannot be determined on the bases of morphological or semantic complexity, one can assume multifunctional sets (to be called derivational strings later on) as (1) (2)

[siki]A, [siki]N, [siki]Vjntr. [siki]vtrans 'sick', also valid for hebi 'heavy', blaka 'black', nati 'wet', doti 'duty', do fit 'deaf, etc. [dede]A, [dedej^j , [dedejvintr 'dead', also valid for bigi 'big' bun iz 'happy', dipt 'deep^dunru 'dark',o?o« 'dumb'.

The distinction between the two strings is small. The words from (1) allow a transsitive verb, the words from (2) don't. One can say a e siki a man 'he makes the man sick', but not *a e dede a man 'he kills the man'. Instead one says a e kin a man 'he kills the man' . Is this a coincidence, because there already exists a verb km to kill' that blocks the derivation or can one discover a systematic distinction between the words belonging to either string which excludes transitive use in the second string. The derivational direction could be responsable. The distinction between the two derivational strings would be explained if the bases of (1) would be Vtr and of (2) would be Vjntj, and furthermore the derivational direction would be [V]tr -»· [V]^ (and not the other way around). This is a completely hypothetical solution, which is certainly contrary to my own intuitions, but which can still illustrate the type of solutions involved here. The lexicon would then have to look like (3)

[siki]vtr [dedejvmtr Wti

'sick' 'dead'

Multifunctionality as a derivational problem

29

This type of solution seems to me valuable, but should be tested on richer material. This is the main reason for the research reported in this article. There are nouns such as: (4)

[alen]N [botro]N [brifi]N

'rain' 'butter' 'letter'

These words can only be used as nouns and therefore they are not multifunctional. Another group of nouns: (5)

[bro]N [brasa]N [bribi]N

'breath' 'embrace' 'belief

are multifunctional, and at least have a verb in their multifunctional string. Given the hypothetical lexicon in (3) they would be included there as verbs and produce the corresponding nouns in (5). One may want to include the nouns in (4) in the lexicon and the corresponding verbs of (5). As soon as the derivational direction has been determined, rules can be made which produce the desired multifunctionality. Up to this point the lexicon in (3) can be maintained with the necessary additions. If one, however, includes the following group of nouns in the investigation (6)

[bosro]N [wipi]N [froiti]N

'brush' 'whip' 'flute

problems emerge. These nouns are also part of a derivational string containing a verb. If one, in order to give the derivational rules in (3) an as wide as possible application, accepts here too the verb as basis for the derivation, one has to conclude that [bro]N 'breath' is the product of [brojy and [bosroJN 'brush' is the instrument with which one (does) [bosrojv · Thus, two clearly distinct derivations are used. In English, one could express this distinction by the derivational direction (7)

[brush]N -»[brush]v to use N' [breath]y -> [breath^ 'product of V.

If one chosses this solution, one has to distinguish two types of nouns: nouns as in (4) and nouns as in (6), where only the last group can form the

30

Jan Voorhoeve

basis of a derivation as in (7). This can be done by putting the nouns of (6) in the subcategory 'instrument'. The Sranan lexicon would then have to contain a rule such as: (8)

[NJinstr^tV]

'to use N'

This would also explain why lexicon (3) doesn't produce an adjective bosro 'brushed', but does produce an adjective broko 'broken'. For the derivational basis [broko]vtr to break' produces an adjective according to (3), but [bosroJvtr 'to brush is a derivation itself. One would have to assume in that case, that a derived word (at least with this type of 'zeroderivations') cannot function as the basis for a further derivation. This restriction had not yet been assumed in the hypothetical lexicon (3). So the first derivational rule should be [V]ti -»[V]^, [A], [N].

3. DERIVATIONAL STRINGS

The above was a preliminary attempt, with as its only objective an investigation as to the possibility of providing rules for multifunctionality. A comprehensive investigation of the Sranan lexicon has to be based on a overview of the derivational strings that are allowed in the language. All lexical elements of the lexicon have to be categorized ascit were on the basis of the function or functions (if multifunctional) in which they occur. I hereby assume that the derivational strings can be expressed in syntactic surface categories such as A (adjective), N (noun), Vtr (transitive verb) and Vjntr (intransitive verb). Other categories such as preposition, pronoun, proper name, auxiliary, verbal tense or aspect markers, article etc. will be excluded for the time being. These are mostly closed categories. For the time being the categorical symbols can be defined as follows (9)

A N Vtr

= is followed by a N that is not the object of A = is accompanied by an article = is followed by a N that is an object of Vtr = is not followed by a N that is an object of Vjntr.

These definitions are obviously incomplete and are not intended to distinguish the four described categories from preposition, adverb, article, quantifier, etc. There can be observed a certain relation to the syntactic features in Jackendoff (1974:13). If the preceding descriptions are translated in Jackendoff s features, one would get A = [ -Subj, +Comp], N = [-Obj, +Comp], Vtr = [+Subj, +0bj] and Vjntr = [+Subj, -Obj]. Jackendoff does not distinguish between transitive and intransitive verbs as pri-

Multifunctionality as a derivational problem

31

mary categories. I think, however, that this will prove inevitable in his conception. It does not seem feasible to regard an intransitive verb as [+0bj]. Jackendoff describes the feature ±0bj as one that 'distinguishes those categories whose complements include an NP 'direct object' after the head from those categories whose complements do not contain NP'. If a primary categorical distinction between transitive and intransitive verb is accepted, the relation between the two readings of mi wasi kaba Ί have washed already' with Vtr and Ί am already washed' with Vjntr can just as easily be described as a derivational relation as the relation between [wroko]vtr to work' and [wroko]N 'work'. This has important consequences for stative formation, which will be treated later as a derivational process. The definitions above are certainly not meant to be heuristic procedures; in which cases one would have to state potentialities as X = can be followed by Y. Heuristic procedures are in principle impossible in a language with multifunctional elements, because the same word form can constitute different categories depending on the environment. The definitions in (9) are to be regarded as concise indications of what in reality are categorical meanings as for instance V^ = a state in which a following NP is brought, Vjntr a state in which a preceding NP is. I will here treat a sample form from the Sranan lexicon, taken principally from the letters A - F of the Woordenlijst van het Sranan Tongo (1961). Particular emphasis is put on the derivational strings in the material. Special problems, as the existence of a relation between [doroj^ 'door' and [dorojvintr to arrive' (cf. M lh usler 1975:54), have been avoided. These can be explained by collapsing sound changes.1 The material has been classified in groups of identical derivational strings. The strings are represented in a matrix in which χ indicates the suposed derivational basis, + the existence of a derived form, and - the absence of it.

(10)

Vte

V**,

Ν

A

+

χ + + + χ -

+ + χ χ

+ χ

+ +

-

+ χ χ χ

-

siki 'ill' (cf. also 1) dede 'dead' (cf. also 2) bosro 'brush' (cf. also 6) botro 'butter (cf. also 4) wasi 'wash' wroko 'work' brasa 'embrace' #'give'

32

Jan Voorhoeve

The categorical symbols represent surface categories. The fact that botro 'butter' is not countable and brifi 'letter' in the same group is countable, not expressed. The basis of the derivation has been determined intuitively. These intuitions are not based on etymology. Justification can only be provided by the complete grammar: does one succeed in producing all regular multifunctional sets or not? Several uncertainties remain for the moment, wroko 'work' is considered here an intransitive verb . This type of verbs can have the derived noun as its object (mi wroko wan wroko, litt. I worked a work) or can have an indefinite object (mi wroko a sani, litt. I worked the thing), both meaning Ί did something'. Verbs like brasa 'embrace' and bribi 'believe' (cf. also 5) are transitive, but one can say mi brasa en wan brasa (litt. I embraced him one embrace). They form therefore a separate group. I feel however that this problem has not been sufficiently studied yet. If my intuitions are correct, subcategorization should distinguish between the nouns bosro 'brush' and botro 'butter'. It has been proposed already to use a subcategorization 'instrument'. One should further distinguish the verbs wasi Vash', brasa 'embrace' andgi 'give'. The following subcategorization is proposed (11)

wasi Vash' brasa 'embrace'

[V] eig+nom [V]eig+ ioc

gi

[Vleag + nom + toe

'give'

The distinction is based on the number of arguments and on the semantic case of their objects. Finally the intransitive verbs are subcategorized as (12)

siki 'ill' wroko 'work'

[V]nom [V]eig

on the basis of the semantic case of their subjects. Derivations can be applied to such a subcategorized base, so that there will be derivations like (13)

[VLsg+nom -»

[V]nom

to account for the distinction between (14)

a wasi den krosi den krosi wasi

'he washed the clothes' the clothes are washed'

33

Multifunctionality as a derivational problem I would like to call this the stative formation. The causative formation is executed by the exactly opposite rule

(15)

[V]eig +

[V]nom

nom

to account for the distinction between (16)

a man siki Leba siki a man

the man is ill' 'Leba (a God) has made the man ill'.

4. CONCLUSION

The proposals made in this paper are summarized in lexicon (17). Each item in this lexicon stands for a number of items with identical properties. I have only studied verbs, nouns and adjectives here.

(17)

bosro botro brasa dede 8t siki wasi wroko [NJinstr [ » Jerg + nom

[V]nom Wnom Mnom [A] [A] ι » J erg + nom

'brush' 'butter' 'embrace' 'dead' 'give' 'ill' 'wash' 'work'

-> -» -»· -> -> -> -> ->·

[NJinstr

[N] [V]erg + loc [A] IMerg + nom + loc [V]nom I" Jerg + nom Merg I.VJ(erg + ) nom

[N] [N] [A]

'to use NT 'product of V

1 · Jerg + nom

[N] [V]nom ΙΎ Jnom

This represents a highly simplified lexicon, of course, that leaves many problems unsolved. I believe, however, that most problems can be solved along the lines indicated here.

34

Jan Voorhoeve

NOTE l. Irregular (accidental) relations may be found between lexical items in a Creole language, that are caused by homonymy. Perfectly regular sound correspondences create homonyms. In Sranan the form ati corresponds in a regular way to the English forms hurt, hat, hot and heart; tapu corresponds in the same way to English top and stop, and to Portugese tapar 'cover'; doro corresponds to English door, Dutch door 'through' and Akan dru 'arrive'. It is quite possible that Sranan speakers assume a derivational relationship between doro 'door' and doro 'arrive', or between tapu 'top' and tapu 'cover', but these are clearly irregular and will not be taken into account in this contribution.

BIBLIOGRAPHY Anderson, John M. (1976), The Grammar of Case. Towards a Localistic Theory. Cambridge Studies in Linguistics 4. Cambridge. Aronoff, A. (1976), Word Formation in a Generative Grammar. Cambridge, Mass. Hocket, Charles (195 9), A Course in Modern Linguistics. New York. Jackendoff, Ray (1975), Morphological and semantic regularities in the lexicon, Language 51,639-671. Jackendoff, Ray (1974), Introduction to the X convention. Indiana University Linguistics Club. October. Mühlhäusler, P. (1974), Pidginization and Simplification of Language. Pacific Linguistics, Series B, 26. Canberra. Mühlhäusler, P. (1975), The functional possibilities lof lexical bases in New Guinea Pidgin. Int. Conf. on Pidgin and Creoles, Hawaii. Voorhoeve, Jan (1961), Linguistic experiments in syntactic analysis. Creole Language Studies II, 37-60. Woordenlifst van het Sranan-Tongo. (1961), Paramaribo.

Peter Mühlhäusler*

3. The development of the category of number in Tok Pisin 1. INTRODUCTION

In a paper discussing the state of the art of developmental studies in the field of pidgin and Creole linguistics (Mühlhäusler 1979a) I have pointed out that the scarcity of data-oriented studies is one of the reasons for the lack of progress in our field. The present study is meant to rectify this situation in a small area of grammar, i.e. that of number marking. The phenomenon of pluralization, though a predominantly syntactic one in Tok Pisin, requires attention to other levels of grammar as well. The following points are of particular interest: i) ii) iii)

the absence of morphological complexity in Tok Pisin when compared with its lexifier languages the dependence of overt plural marking on semantic criteria the phonetological properties of the plural marker

In addition, the development of plural marking can provide valuable insights into general principles of the development of pidgin languages and language mixing. When I began to look at number marking in my large corpus of Tok Pisin data it struck me how little the numerous statements made on this aspect of Tok Pisin grammar were borne out by the data. There is a considerable discrepancy between what grammarians say speakers are doing and the reality of the spoken language. Most grammars in the past were written for pedagogical reasons (often with the implicit aim of 'elevating' the language at the same time) or to serve as easy reference grammars to newcomers to Papua New Guinea. The few linguistic descriptions available were mostly written within a structuralist framework which called for static, abstract, overall-pattern descriptions. * I wish to thank Pieter Muysken (Univ. Amsterdam) and Don Laycock (Australian National University) for comments on the draft, Tom Dutton (Australian National University) for making available unpublished data on Queensland Kanaka English and my wife Jackie for numerous corrections and suggestions.

36

Peter Mühlhäusler

In this paper I will abandon considerations such as neatness or symmetry and describe what actually happened (and is still happening) at a much more concrete level. In addition, my descriptive approach will be dynamic, in that I treat Tok Pisin as a continuously expanding system. In sketching the development of number marking I will also consider the many unsuccessful forms, some of which have developed further in other closely related Pacific Pidgins.

2. STAGES IN THE DEVELOPMENT OF PIDGIN LANGUAGES, PARTICULARLY TOK PISIN

One of the central problems of this paper was alluded to in Schuchardt's analysis of Saramaccan (1914: iv), though only recently has this idea received the attention of linguists. Speaking about the role of substratum influence in the development of pidgins and Creoles, Schuchardt remarks: 'Under what circumstances could these African languages be of considerable influence? No longer once the Creoles had become the mother tongue of the majority; not yet during its formation as a makeshift language. Both master and slave were interested only in making themselves understood. The former stripped the European language of all its special features, the latter held back all special features. They must met halfway. The master, for instance, recognized right from the beginning that European specifications of plural, the -s of stone-s or piedra-s, or even the (unstressed) de of des pierres, would certainly meet with a total lack of understanding. And he thus adopted the radical solution of saying 'Stein Stein' or *Menge Stein' or 'Stein viel' ['stone stone' or 'multitude stone' or 'stone many', P.M.] or, if the third person plural pronoun had already been established, 'Stein sie' or 'sie Stein', [they stone' or 'stone they', P.M.]. The slave resorted to the same means of communication, since he could do nothing with the inherited prefixes and suffixes. He would even himself have been reluctant to adopt them; a Duala would form a plural ma-stone, based on his ma-dala 'stones', much more readily in his own language than in the foreign one. (author's translation) Thus, physical contact between two languages and the absence of 'mechanical barriers' to language mixing (cf. Whinnom 1971: 94) alone does not result in language mixing. In fact, at the time when language contact is maximal very little language mixing occurs. The reasons for this phenomenon have to be sought in the ever-changing, dynamic character of pidgin languages. These languages not only change with regard to their grammati-

The development of the category of number in Tok Pisin

37

cal and lexical make-up but also with regard to their susceptibility to outside influences. A closer analysis of the changes found with pidgins and Creoles reveals that they can be ascribed to two principal factors: 1. development, in the sense of an increase in overall complexity; 2. restructuring, i.e. changes due to contact with other languages, in the present case contact with Tok Pisin's original lexifier language English. As a rule, these changes do not affect the overall complexity of the language. The above changes can be summarized in the following scheme: Table I. Directions of Change in Pidgins and Creoles

developmental continuum

Jargon (Bush Pidgin) Stable (Rural) Pidgin Expanded (Rural) Pidgin Post-Pidgin (Urban Pidgin) English Creole Post-Creole (Urban Creole) English restructuring continuum

Whilst the development of a pidgin language proceeds along a continuum, it seems convenient to distinguish between qualitatively different phases, as one distinguishes a one-word phase from a two-word phase in the discussion of child language development. It has become customary to distinguish the following phases in the lifecycle of pidgins and Creoles: Table II. Developmental Phases and Their Grammatical Properties jargon

stable pidgin

expanded pidgin

Creole

one or two-word sentences, tendency towards CVCV word-structure, no deictic markers, referentially impoverished gradual emergence of simple sentence structures, some shallow embedding, consonant clusters becoming more common complementizers and relativizers emerge, development of a productive word-formation component, discourse-structuring devices complex (multiple) embedding in syntax and multiple derivation in word-formation component, numerous stylistic devices

38

Peter Mühlhäusler

J

The order in which plural marking appears in a pidgin appears to be governed by universal principles of language development. In practice this means that what is more prominent in a sentence has a greater chance of attracting additional grammatical categories such as number or gender than what is less prominent. Prominency, the data suggest, can be defined in terms of two parameters, namely animacy and semantic case. It is important to keep in mind that linguistic development can proceed along dimensions other than the temporal one. Changes found there are recapitulated along the social and geographical dimensions: 'This means nothing else but that, with the continuous spread of Pidgin, its linguistic history is repeated, so that, for instance, the contact situation in a very remote area today resembles that of a developed area fifty years ago. The linguistic phenomena encountered, too, are very similar in kind. In its expansion Pidgin has become repeatedly pidginized and brought back to the norm that had developed in the meantime.' Mühlhäusler 1975b: 62) For example, the relation between linguistic stability and temporal and spatial factors can be illustrated as follows: Table III. Time and Space in the Development of Stable Tok Pisin

Duke of York Islands

Gazelle Peninsula

New Guinea coast

New Guinea interior

New Guinea remote interior

space axis

I will now proceed to illustrate these general principles with some data on number marking in Tok Pisin.

The development of the category of number in Tok Pisin

39

3. THE DEVELOPMENT OF NUMBER-MARKING IN TOK PISIN 3.1. Developments During the Jargon Phase

As I have discussed elsewhere (Mühlhäusler 1978b) much of Tok Pisin's early development took place outside New Guinea on the Samoan plantations of the Deutsche Handels- und Plantagengesellschaft. Reliable data on early contact Tok Pisin and on the bush varieties of present-day Tok Pisin are hard to come by and the statements in this section must therefore remain somewhat tentative. In a contact situation, two points have to be considered: i) ii)

the input the learners' are exposed to, their ability to learn the language they come into contact with.

An examination of examples of Europeans' pidgin before 1900 reveals remarkable inconsistency with regard to plural marking. Consider the following examples representing the use of Pidgin English by a recruiter of labour (Thomas 1886): (1) (2) (3)

one fellow so, all fellows so (p. 273) plenty shops (p. 328) man Tanna want Englishman (p. 260)

Such inconsistencies also appeared in a 'foreigner-talk' test which I administered to a small group of Australian students. I found that, whilst more than 90% of the third person singular -s was deleted, plural -s was deleted in only about 40% of all cases. Inconsistent input may thus be one of the reasons why certain lexical items have plural endings in jargon texts, including those which become fossilized in this form, such as anis 'ant', bis Tjead' or has 'card'. Similarly, some German plural endings were preserved in items such as hebsen 'pea',binen 'bee', sirsen 'cherry' and bonen •bean'. The learner's ability to cope with linguistic input can be ascribed to two main sources: i) the natural second-language (interlanguage) hierarchy and ii) the grammar of his first language. In the past, under the impact of structuralism and contrastive linguistics, the second factor was given much more attention, though the claims of contrastive and comparative models were seldom subjected to empirical verification. Unfortunately, the early contact jargon was spoken by learners from many linguistic backgrounds and the task of providing contrastive data is thus considerable. An interesting observation is that by Schuchardt (1881: 157) on New Caledonian Pidgin English:

40

Peter Mühlhäusler 'The most elementary grammatical differences tend to be neglected. They are made only if absolutely necessary, and then in a very clumsy manner, such as him fello all = they' ... I must mention, in this connection, that in the Melanesien languages 'all' is used, among other things, to express the idea of plurality where required.' (author's translation)

Walsh (1978: 188-9) suggests that some of the properties of present-day Tok Pisin pluralization may have an East-Austronesian origin, though number marking appears to be a less clear instance than other constructions: The TP plural indicator ol has no structural counterpart in English, whereas Raga has the plural marker l-ra which can occur with nouns referring to people, e.g. l-ra vavine plural woman Vomen'. The affinity of the TP plural indicator ol with ol they/them' is parallelled by the affinity of Raga l-ra with ra- they' and -ra 'them'. The syntactic value of the TP form is again more plausibly derivable from the model represented by Raga than from English. Walsh's main shortcoming is his use of a static comparative model to determine relationships between a pidgin language and its lexifier and substratum languages. He ignores the fact that the number marking system of present-day Tok Pisin is not a simple continuation of plural marking in the days of early contact but the result of developments that proceed along lines not found in the early substratum languages. My view is that the similarity between Tok Pisin and Raga is a chance one (though explicable in terms of universal conventions for plural marking) and not due to shared history. The substratum argument is further weakened by the fact that, in his early stages of second-language leraning, a learner simply does not have functors such as plural in his grammar (cf. Dulay & Hurt 1974: 37-53), two reasons being that plural is a marked semantactic category and thus 'late' in any developmental hierarchy and that inflectional morphology gets reduced or lost in language-contact or incipient learning contexts. The first point has been discussed in more detail by Mühlhäusler (1974: 91 ff). One would predict, as suggested by Schuchardt (cf. quotation p. 2), that, irrespective of whether speakers of substratum language have a morphological category plural in their vernacular, they will not carry over any conventions for plural marking into the pidgin in its formative period. However, we can expect attempts to cope with the semantic notion of plurality where disambiguation is called for.

The development of the category of number in Tok Pisin

41

The data examined include information i) on pronouns and ii) on nouns. As can be expected, the development of conventions for number marking in pronouns comes first chronologically, though it does not, as suggested by Cassidy (1971:213), appear right at the beginning. In Samoan Plantation Pidgin (SPP), Tok Pisin's predecessor, no plural pronouns are found in early texts, such as those collected by Schuchardt (1889) where sg. he is also used as the third person plural pronoun. At this stage two formatives were competing as indicators of semantic plural, one being -pela, the other one ol (from 'all'). By 1910 the pronoun system of SPP had taken the following stable form: sg. first mi second yu third em (him) hi

pi. (subject) mi ol yu ol em (him) ol

sg. first (bilong) mi second (bilong) yu third (bilong) em (him)

pi. (object) (bilong) as (bilong) yu ol (bilong) dem

However, even after 1910 the second and third person singular continued to be used instead of the plural forms as the anaphoric referent to a previously introduced plural noun phrase or pronoun. As can be expected with a developing early pidgin, the semantic plural is expressed in an analytic manner, i.e. by means of quantifiers such as plenti 'plenty', ol 'all', olgeta 'alltogether'. As a rule plurals are not signalled overtly where a plural meaning can be derived from contextual or textual information, i.e. there is little or no redundancy in the area of grammatical categories. 3.2. Developments in the Stabilization Phase Around 1880 Tok Pisin became structurally stable in the Duke-of-York New Britain and New Ireland area, and later on other islands and the New Guinea mainland. Stabilization occurred at a time when Tolai was the most important language in the young colony and thus a potential source of structural and lexical innovations. During this phase the pronoun system of SPP underwent a number of changes, which were apparently conditioned by two forces: i) that towards unitary representation and ii) that towards naturalness. The Samoan pluralizer ol was given up in favour of pela, resulting in the plural forms mipela Ve', yupela V011' and hipela they'. Whilst the first two forms have be-

42

Peter Mühlhäusler

come standard usage, hipela (documented in Thurnwald 1908) has since disappeared. The basic weakness of this system was that in spite of its symmetry, it was unnatural in that the first person plural should not be encoded as (I + plural), (cf. Mühlhäusler 1979a: 31). The solution was the introduction of an inclusive (yumi) and exclusive (mipela) first person plural pronoun. The distinction is also made in Tolai. For speakers who do not make this distinction in their first language it remains a source of interference until very late in their acquisition process. The final outcome of the various restructuring processes combines features of the old Samoan and other systems and is a compromise rather than an optimal solution: 1st inc 1st excl 2nd 3rd

singular mi yu em, (h)i

plural yumi mipela yupela (em)ol, olgeta

Texts after 1920 show a gradual disappearance of hi and olgeta as third person pronouns. As regards the noun system, three formatives competed as noun pluralizers: -pela (which is still found with pronouns) is documented only in some very early texts, for instance: (4)

no likee German feller (Burnell 1915)

Brennickmeyer (1924: 2) writes: "The plural is formed by a preposed 'all' or a more forceful 'altogether'.' (author's translation) Instances of the competive forms ol and olgeta in about the same meaning can be found in many early texts. An example are some plural forms found in Detzner (1921): (5) (6) (7)

altogether German alltogether boy he like kill alltogether man, all marries and pikanninies

the Germans' 'the indigenes in European service' 'he wants to kill men, women and children alike'

The fact that ol, olgeta and pela were competing pluralizers at one time may explain why they cannot be combined in present-day Tok Pisin.1 Compare:

The development of the category of number in Tok Pisin (8) (9) (10) (11) (12)

sampela ol man *olgeta ol man olgeta man *olfelaman olman

43

'some men' 'all men' 'all men' 'men' Ίηβη'

The above translations suggest that, in later stages of Tok Pisin, speakers have applied the principle One form-one meaning', i.e. olgeta has come to mean 'all'. Apart from a straight plural one also finds evidence of a plural which indicates 'groupiness'. Compare: (13) (14)

ol pater pater ol

the priests' the priest and his flock'

Whilst it has not been recorded in any grammar, this form can be commonly heard in present-day Tok Pisin. According to Mosel (personal communication 1979) this construction corresponds to Tolai pater dital. Unfortunately it is not documented in early Tolai grammars so that the direction of borrowing is not clear. One may be dealing with an internal development of Tok Pisin, a suggestion whose plausibility is reinforced by the fact that a comparable construction is found in a number of historically unrelated pidgins and Creoles (cf. Muhlha'usler 1974: 20-21). Throughout Tok Pisin's stabilization phase pluralization is an optional category. In the following text, spoken by Noah Maikan of Smain village near Dagua (East Sepik Province, recorded in 1972, speaker about 75 years old), plural ol is used only in three out of twenty possible instances. As a rule, plural is indicated only with the first pronoun of a sentence, otherwise the third person singular pronoun / is used: (15)

Siaman i kamap. Nambawan samting bipo dispela graun no gat masta. The Germans arrived. At first this land had no Europeans. / no gat masta. Ol i raun noting i kamap long Wallis, i_ kamap. It had no Europeans. They (predicate marker) sailed about and they arrived on Wallis, they arrived. I_givim masket long man bilong Wallis. They gave guns to the men of Wallis.

With slightly younger speakers from the New Guinea coast one notes a gradual increase in the use of the plural marker ΌΓ. An examination of a large corpus of data from speakers in the 50-70 years age-group indicated that the appearance of the plural marker ol is related to i) the semantics

44

Peter Mühlhäusler

of the nouns involved and ii) the syntactic environment in which they appear. This means that the development from a numberless jargon to a pidgin in which number indication is obligatory in many contexts follows a well-defined program, in which the rules for the use of number marking are implicationally ordered. Lack of space prevents me from reproducing the texts and tables used in charting these developments. However, as some of the texts are to be found in readily available publications, the reader can check the validity of my statements for at least those sources. The next representative text is by Dziru from the Madang area (about 55 years of age) recorded by Fischer (1966). As this is a rather long text, the tendencies are easy to see. They are: I) Plural marking is favoured in the case of nouns denoting a) humans, b) animates, c) count nouns and d) mass nouns, in that order. II) The grammatical environments favouring plural marking are a) subject position, b) direct object position, c) after preposition long (indirect object or locative) and d) after preposition bilong (possessive), in that order. III) There is a marked tendency to keep redundancy to a minimum within sentence boundaries. There are thus a number of contexts which discourage the use of plural markers: a) olgeta 'all', b) sampela 'some', c) kainkain 'all kinds of, d) numbers, e) the use of the anaphoric pronoun ol, and f) enumerations when the first noun is marked for plural, in that order. Here follow two extracts illustrating some of these tendencies. Note that nouns referring to humans always take ol if they are semantically plural, whereas all other nouns are variably marked for this category. An interesting feature is the distinction between ten siling 'ten individual shilling pieces' and ol ten siling ten shilling notes'.

(16)

D: I-tokem mi wanfela stori olosem: I-gat wanfela pies bilong ol. I-gat planti kakaruk i-stap. Kakaruk men na meli i-stap long disfela pies, orait, wanfela man em i-lail.im kisim mani na em i-go singaut long em. Orait i-go was long em na i-stap i-stap nau, olo disfela kakaruk i-kamap, singaut na i-laik holim pas i-nogat. Mekim mekim i-nogat. (p. 68)

Translation: He told me the following story: There was this place of theirs. There were many chickens. There were roosters and hens in this place. Well, one man wanted to make money and he called them. He watched for them for a long time. These chickens came, they were cackling, he wanted to grab them, but he was unsuccessful. He tried again and again without success.

The development of the category of number in Tok Pisin (17)

45

D: Yes. Na i-kisim, orait, siubim long olo man. I-tokim disfela mani i-kamap. Olo man i-lukim pinis, em i-putim bek i-stap. Na i-go, ten siling i-kamap. I-go kamap long ten siling na i-tok: Nau i-kamap long ten siling, inap long ten siling. Na baimbai ol ten siling tasol i kamap. (p. 78)

Translation: Yes and he took it and showed it to the people. He said: This money has come. The people looked at it, and he put it back. This went on for some time and ten shilling pieces appeared. When ten shilling pieces had appeared he said: We have reached ten shillings, from now on only ten shilling notes will appear. Whereas present-day Tok Pisin is an indigenous lingua franca, the role of European speakers was much greater during the stabilization period (cf. Mühlhäusler forthcoming). It thus seems necessary to examine also plural marking among the Europeans' languages, in particular among the very influential German missionaries who began to standardize the language in the 1920s. My main impression after an examination of written missionary Tok Pisin is that there is a marked tendency for the pluralizer ol to appear whenever one is dealing with a semantic plural, even in those cases where contextual information already expresses the idea. Thus, lexical items which are low on the animacy hierarchy and lexical items in constructions that discourage the use of overt pluralization are marked for plural. The following sentences were taken from a hagiography, published in Alexishafen in 1936: (18) (19) (20)

Mama i lainim kwik long ol gudfelo fashion. 'His mother quickly taught him good manners.' Mama i lainim tu long ol liklik beten. 'His mother also taught him little prayers.' Papamama i lukautim em long ol rum bilong haus. 'His parents looked for him in the rooms of their house.'

The same tendency, as well as the use of redundant plural marking within sentences and phrases, is found in Tok Pisin course materials such as the one by Borchardt (1930): (21) (22)

Ql diwai i stop nambis, han bilong o/ i silip daon tumas. 'The branches of the trees on the beach hang down to the waterlevel.' 01 ensel crt i lukaotim yumi. 'The angels are looking after us.'

46

Peter Mühlhäusler

The fact that most of the expatriate speakers of Tok Pisin during its stabilization phase were of German origin may also have influenced the choice of the plural formative (pi or olo rather than olgeta or plena), as Tok Pisin ol was often identified with the German quantifier alle. 3.3. A theoretical assessment of stabilization in Tok Pisin The stabilization stage is the most central stage in the life cycle of a pidgin/ Creole; it is at this stage that the direction for the future development of the language is determined. In the case of the Tok Pisin plural marker, the choice of the phonetic form ol or olo creates certain conditions for phonetological rules which may emerge at a later developmental stage of the language, while the syntactic fact that the pluralizer preceeds nouns or noun phrases may eventually be reflected in the development of inflectional plural formatives. In other words, during the stabilization stage a pidgin becomes a linguistic system with a well-defined set of potentialities for further expansion. The development of a well-defined system of pluralization in Tok Pisin will now be discussed in terms of four main parameters: 1) the emergence of a new form for a meaning that was not previously encoded in the language 2) the wave-like spread of this form to more and more grammatical environments 3) a gradual change in the meaning of this form 4) optional and obligatory grammatical categories 1) The emergence of a new form for a new meaning typically involves the reduction of the random or free variation of the jargon stage to ordered variation or to the assignment of a single meaning to a single form. Whereas free variation between ol and olgeta is still documented in Borchardt's grammar (1930: 5), grammars and vocabularies written slightly later distinguish the meanings ol 'plural' and olgeta 'all'. In addition, the Tok Pisin plural has a meaning which is not found in either the early Jargon or in English, i.e. with animates it refers to three or more or four or more depending on the speaker's regional dialect. Details of the pluraldual-trial distinction are given by Nevermann (1929: 256). In a recent paper, Bickerton distinguished the following two types of linguistic change (1979: 4): 'In natural change, an already existing form or structure acquires a new meaning, function or distribution. In decreolisation, an already existing function or meaning acquires a new form or structure.' The data just discussed necessitate the introduction of a third category

The development of the category of number in Tok Pisin

47

of change which I have labelled expansion (cf. Mühlhäusler 1979a). Expansion is encountered in a situation where a language acquires new categories which make it referentially or otherwise more powerful. Expansion can be found with both second-language pidgins and with newly developing Creoles. 2) An important characteristic of the development of number marking is that it is only found in a very limited structural environment, the restrictions for the application of the pluralization rule gradually becoming relaxed as the language develops further. It seems that the order in which these restrictions are relaxed is more or less preprogrammed and that the implicational order of plural marking is the same for all speakers of Tok Pisin (with the exception of those on a post-pidgin continuum; this will be discussed below). The factors involved, include pragmatic factors, textual factors and semantactic factors at sentence or sub-sentence level. Whilst the operation of individual factors (e.g. the animacy hierarchy) is more or less clear, the complex interplay of all the factors needs to be studied. For this a far more extensive analysis of texts would seem to be necessary, which may not be possible in view of the lack of early data. Even if all the details of the development of plural marking cannot be documented, they support the view (put forward by Mühlhäusler 1979a) that pidgins in later stages of their development are simpler, in the sense of more rule governed, than pidgins at earlier stages. Overgeneralization, rather than being the first stage in rule acquisition or language expansion, in fact appears very late. 3) The gradual change of the meaning of ol and olgeta illustrates a more general process which is common in the expansion of a pidgin, namely the transition from lexical to syntactic forms. That ol typically appeared when a large number of referents, and not simply more than three, were involved is documented in many early sources (e.g. Thurnwald 1913: 95-96). As late as 1943 Hall makes a similar observation (1943: 2): 'Nouns in Pidgin do not have any plural. You must use the singular form at all times: one-fellow man, two-fellow man, plenty-fellow man. If you want to emphasize the fact that more than one is meant, you may use all or all- in the meaning the (plural)': all man, all- man 'the men'. A few English plurals are used as singular nouns in Pidgin: one-fellow matches One match', One box of matches'.' In some of the texts published in Hall's description one can observe the transition to a simple (often redundant) plural marker. The development of ol into a plural marker is comparable to the development of complementizers out of verbs, prepositions or adverbs (cf. Mühlhäusler 1976a:

48

Peter Mühlhäusler

314 ff), or the development of relativizers out of the adverb hia 'here' discussed by Sankoff and Brown (1976). 4) Whilst plural marking is becoming increasingly common in the later years of Tok Pisin's stabilization (after 1930) it is still not obligatory in any environment. This state of affairs continues to be the rule in conservative rural societies where the language is spoken as a second language. Thus, Laycock (1970: xix) writes: Nouns in Pidgin are used without articles, and have no number or gender; thus kapal means a possum, the possum, possums, the possums, female possum, male possum, and so on. Sometimes the third person pronouns are used to indicate number, and in these contexts are comparable to the definite article in English: em kapul hia the possum here, ol kapul i stop long diwal the possums in the trees. The numeral wanpela is often weaker than in English, and may be translated as the English indefinite article 'a' in sentences like wanpela pukpuk i stop long dispela baret there is a crocodile in this stream. Laycock's data illustrate the interdependence of temporal, social and spatial factors in the grammatical expansion of a pidgin.

3.4. Plural marking in the Expansion Phase Grammatical expansion is closely linked to functional expansion, i.e. in those communities where Tok Pisin has taken over most of the functions of the traditional vernacular it is structurally most complex. The developments during this stage are a continuation of those found at the previous stage. Whilst in theory Tok Pisin is developed enough to borrow grammatical categories from other linguistic systems, the very multitude of languages with which it is in contact makes borrowing less likely. It appears that the developments at this stage are the result of internal natural growth. Outstanding characteristics are that plural marking is generally redundant with pronouns and animates, that nouns lower down on the animacy hierarchy can collocate with the plural formative and that semantic rather than lexical factors determine the choice of plural marking. These tendencies can be found in the following account of the first German contacts with Wallis Island told by a young man (aged around 30) from Wallis. This story should be compared with text number (15) where the same contact story is told by an older speaker. (23)

Na long moning olgeta man klostu o/ go sanap long nambis. And in the morning all men nearby they went and stood on the beach.

The development of the category of number in Tok Pisin

49

O/ / wet sip. Ol i sanap lukluk, bikpela sip i kam . . . They waited for the ship. They stood and looked, a big ship arrived. Dispela man, bikman, i go antap na toktok long ol masta nau. This man, the important man, went up and talked with the Europeans. Ol i mekim long han tasol, long wanem ol masta i no save Tok Pisin. They used sign language, because the Europeans did not know Tok Pisin. A similar concentration of plural marking is found further on in the same story: (24)

Pints, olJeman igo nau. Dispela taim em ifestaim tru After that, the Germans went away. This was the very first time ol i bin lukim wanpela masta. Orait, o± i stop liklik,... that they saw a European. Well, they stayed on for a while ... Ö/ man bilong pies bilong mipela i bin lukim pastaim o/ masta. The people of our village were the first to see the Europeans.

Categorical use of plural with animate nouns contrasts with variable use with inanimate nouns in the following text spoken by a middle-aged speaker from near Maprik (four hours drive from Wewak, the nearest coastal town): (25)

Mi toktok long o/ pipal insait long pies, toktok long rot long I speak to the people in the villages, I talk about roads and skul samting. Ol i no bin wok long helt, nogat, ol i save schools and so on They didn't do anything about their health, they sindaun noting. Na ol i no save troimwe pekpek long bus. just sat around. And they used not to throw away their excrements in the bush. O/ i save sindaun wan taim ol pekpek. They used to live with their excrement^.

An analysis of the speech of a group of young men in Tuman village (a conservative village seven truck hours from Wewak) has shown that the use of ol with inanimates is as high as 80% in subject position and 60% in direct object position. This increase is partly due to the fact that many so-called mass nouns are now treated as plurals. A number of writers (e.g. Milhalic 1971: 12) have drawn attention to the fact that a few Tok Pisin nouns have implied plural signification. Thomas (1969: 13) writes:

50

Peter Mühlhäusler

A noun taken in itself is singular, generally speaking . . . . haus, bolus, diwai, pinas, buk. There are some words, however, which of their very nature usually have a plural meaning, e.g. morota, daka kaukau, taro, su (shoe), sandal, etc. In Tok Pisin as spoken and written by speakers under the age of 30 one finds numerous instances of these and similar lexical items with the formative ol. The following sentences illustrate that plural marking of "muss nouns' implies plural marking of other inanimate nouns: (26)

(27)

Ol hetman bilong olfaktori i wok long kukim bret i wari tru The directors of factories that bake bread are vey worried long ol plaua i sot. that there is a shortage of flour. Ol sigaret na ol Kapstan samting tu i no stop long stua Cigarettes and tinned tobacco and other things are not in our store Mipela i putim ol rais na ol mit na olgutpela samting tasol. We put rice and tinned meat and only good things.

The full text, from which these examples are taken, is printed in Button (1973: 29). Another factor promoting the increase in inanimate nouns that are marked for plural is the tendency for abstract nouns to be regarded as plurals if: i) they are objects of a sentence with a plural agent, as in: (28) (29)

(30)

oligatolfan they had fun Ostrelia i gat laik tru long strongim ol prensip wantaim Papua Nougini. Australia wants to stengthen the friendship with Papua New Guinea. olkros i bin kamap namel long ol the bad relations that existed between them

ii) the abstract concept can be identified with several concrete items or events, as in: (31) (32) (33)

em i helpim ol inkam bilong ol wokman he increased the income of the workmen strongim ol kalsa bilong ol kantri strengthen the culture of the whole population ol kain projek i mas givim ol trening long ol lokal pipel.

The development of the category of number in Tok Pisin

(34) (35) (36)

51

various projects must provide training for the local people. minista i bosim olmani2 the minister that controls the money (= finance minister) ol divelopmen i no go het gut the development does not progress well olwok politik politics

It is too early to predict whether the syntactic distinction between mass and count nouns will disappear in Tok Pisin. There is certainly a noticeable tendency for inanimate nouns to be neutral with regard to this distinction, so that reclassification has become virtually open. The same tendency is also evident with animate nouns referring to non-countable groups. Whereas in the text samples of older speakers and those recorded before 1950 animate group nouns are treated as syntactic singulars, there is now a strong trend towards regarding them as plurals. The assignment of the plural category appears to have taken place in a number of stages: stage i ol preceding names of contries, islands or towns indicates 'inhabitants of a country', for example, ol Ostrelia 'Australians', ol Amerika the Americans', ol Rabaul the inhibitants of RabauT, olBuka the inhibit ants of Buka Island'. stage ii ol preceding common nouns referring to localities indicates 'inhabitant of that locality', e.g. ol nambis 'beach dwellers',ol bikbus 'people who live in the bush', ol taun 'people who live in town',olmaunten 'mountain dwellers'. stage iii ol preceding names of organizations indicates the members of that organization'. This construction is particularly common with abstract nouns borrowed from English, for example ol ami the army, soldiers', ol misin the mission, missionaries', ol malaria sevis the malaria service, people working for the malaria service', ol pablik helt sevis the public health service, people working for the pub lie health service', olgavman the government, members of government'. For most speakers there is considerable inherent variability of plural marking in the the above three cases. Thus, the following sentences were found in the same letter written by a young man from Yangoru (East Sepik Province):

52

(37)

Peter Mühlhäusler

Igutpela aidia longgivim laisens long Broken Hill Kampani It is a good idea to give a licence to the Broken Hill Company long painim samting olsem gol na kopa. to find things such as gold and copper.

(38)

Sapos ol kampani i painim samting olgol bai sampela mani igo If the Company finds gold and such things, then some money goes bek long ol. back to them.

3.5. The Increase of Redundancy during Structural Expansion The main difference between the stabilization and expansion phases in Tok Pisin can be characterized as follows. During stabilization a systematic grammar is created, whilst during expansion this system is put to an optimal use. This includes the introduction of more redundancy to make up for an increase in tempo. Redundancy also reflects the fact that optimalization of encoding, which leads to an analytic grammar with little redundancy is supplemented by strategies optimalizing perception.3 In section 3.2. we saw that, at the end of Tok Pisin's stabilization phase in the 1920s, plural was often redundantly encoded at sentence level, i.e. the same idea of plurality was expressed more than once within a single sentence. During the expansion stage we find that redundancy is beginning to become common at phrase level. Whereas speakers recorded before 1920 or those of my informants who were older than sixty showed no redundancy of-plural marking at phrase level, younger speakers exhibit a gradual increase in the environments in which ol can (optionally) appear. My data suggest that the sequence of the environments into which redundancy is introduced is the following: i) in enumerations Informants between the age of fifty and sixty were found to use constructions such as ol kago samting 'European type goods and such things' next to kago samting. ii) pronoun followed by noun Constructions of the type mipela ol man *we men\yupela ol wantok *you fellow-countrymen' or yumi kanaka Ve (incl.) uncivilized people' are found next to constructions such as yumi ol kanaka. With younger people the redundant use of ol is definitely favoured. iii) ol with numbers Ol preceding a number can either signal distribution of items or people that are grouped together or simple plural. In some contexts this dis-

The development of the category of number in Tok Pisin

53

tinction is difficult to make. Speakers of Tok Pisin who have compulsory dual and trial marking use ol for numbers over the value of two and three respectively. Others, particularly younger speakers, use ol before any number as long as the concept is plural. Whilst the redundant use of ol with numbers is chronologically early, it is not a construction that is favoured above its non-use by any one group of speakers examined. The following sentences uttered by younger speakers in rural areas, illustrate what can be found. However, a more detailed analysis of the problem needs to be undertaken, before a more definite assessment can be given: (39) (40) (41) (42)

(43)

em i lukim ol sikspela boi he saw six men ol tripela man i bin pait three men were fighting, groups of three men were fighting ol tupela men ol tok two women were talking ol wanpela pies ol sut i kam the inhabitants of one village came down shooting

givim mani long wanpela ol bisnisgrup gave money to a business group

The last two examples illustrate the tendency for ol to mark semantic plural even in those cases where wanpela signals the syntactic singular. Ϊ) sampela 'some', plenti 'many' and similar quantifiers The use of ol together with the above quantifiers is common with most speakers of the younger and youngest generation and there are indications that the use of ol in this context is well on its way to becoming categorical. Before this happens, the problem of the position of the plural formative in the noun phrase has to be settled. At present there are three variants: (44)

sampela ol bisnesman 'some businessmen' ol sampela bisnesman „ „

ol sampela ol bisnesman

„

„

Often, the same speaker will use one or more of the above variants together with sampela bisnisman. There is considerable variability in this area of grammar. One can argue, that this variability will eventually give way to the more consistent pattern found in creolized Tok Pisin (discussed below). v) haumas 'how many'

54

Peter Mühlhäusler

In spite of the fact that English distinguishes between how much and how many, few speakers of Tok Pisin, even those who have had an English education, produce forms such as haumas ol man? 'how many men?', and no example of haumas ol + inanimate noun was found in my corpus. vi) ol kain, kainkain 'various, all sorts of As with v) the use of ol after ol kain and kainkain is found with animate nouns only and is rectricted to younger speakers. vii) olgeta 'all' The least common environment for pluralization is olgeta 'all', possibly because this item was in competition with ol in Tok Pisin's past. The widespread reluctance to collocate ol with olgeta may also be due to the semantic properties of the latter, which include a strong plural meaning. Thus, the combinations olgeta ol and ol olgeta are only found with speakers who also use ol in the environments i-vi above. One such speaker is a young man from Erima Nambis, a village in one of the oldest plantation areas of the New Guinea mainland, where Tok Pisin has been in use for almost 90 years. Here is a sample of his speech:

(45)

Orait, ol tupela men ol painim pis na ol i go long wara Well, two women went to catch fish and they went to the river orait, tupela igo daun, ol painim ol man bilongpait, ol man well, the two went down and encountered warriors, men bilong Bom i laik pait long ol Saiya. Orait, nau from Bom who wanted to fight with the Saiya villagers. Well,then ol tupela men ol i ron i go antap long pies nau, tokim olman the two women ran up to their village, told their husbands bilong ol, na olman bilong pait o/ kamkamap pints, na olgeta and the warriors arrived, and all the men ol man kirap na tokim ol: got up and said to them:

Further on in the same story, a similar degree of redundancy with pluralization is found: (46)

0/ karim ol supia, ol karim ol hap plang, ol karim ol samting They carried spears, they carried shields, they carried the things bilong helpim. Ol sapim kil bilong diwai, o/ sapim, ol putim hia, to help them. They carved the (exposed) roots of (mangrove) trees, they carved them, they put them here,

The development of the category of number in Tok Pisin

(47)

55

orait, ol i go, ol i go sutim banara, ol wanpela pies ol sut ika,... well, they kept shooting with bows and arrows, the population of on village approached shooting (their arrows) ... Ol bipo ol tumbuna ol wokim olsem, ol papa ol^ lukim long ai, na Our grandparents before did it like this, our fathers saw it, and mipela i no lukim long ai, gl stori long mipela tasol. we did not see it, they only told us so.

My findings in connection with redundancy parallel findings by Sankoff and Laberge (1972) on the behaviour of tense/aspect marking in Tok Pisin. 3.6. Plural Marking in the Creolization Phase Creolization (i.e. when a pidgin becomes the native language of a new speech community) can take place at any phase in the development of a pidgin (cf. Mühlhäusler 1977 and 1979a). However, if Creolization occurs with a little-developed pidgin much more intensive repair (restructuring) is needed than in the case of Creolization of an expanded pidgin. With Tok Pisin, Creolization most typically occurs after the language has already acquired considerable structural complexity as a second language. As a result, the difference between expanded and creolized Tok Pisin is not abrupt, but gradual. This is also confirmed in work by Sankoff & Laberge (1972) and Sankoff & Brown (1976). Arguing from what is widely maintained in the theoretical literature on Creolization I set up the following hypotheses: i) ii) iii) iv)

plural marking will become categorical in all environments, i.e. it is semantically determined the position of the plural marker in the noun phrase will become fixed differences in surface form will always be associated with differences in meaning semantic plural will be marked in parts of the sentence other than the noun phrase.

None of these predictions were borne out in full. Moreover, it appears that different solutions to the problems of plural marking are found in the three creolized varieties of Tok Pisin examined, i.e. those of Malabang (Manus Island), Yip (on the Keram River) and urban Lae (data recorded by Sankoff and Laberge). Unfortunately, the texts examined were not of sufficient length for a detailed syntactic analysis such as the one undertaken here, and elicitation and formal interviews were only used in Malabang. However, the following generalizations can be made with confidence: i) Plural marking remains variable for all creolized varieties examined;

56

Peter Mühlhäusler

the only instance of categorical plural marking was that of animate subjects in Malabang Tok Pisin. Still, plurals without ol are in the minority in all semantic environments of the varieties examined. My own feeling is that the trend towards a categorical marking of the semantic plural is blocked by the fact that ol is a relatively stressed free-standing formative and not an unstressed affix. Thus, its use in contexts where the idea of plurality is subordinate would seem slightly unnatural (for instance, in constructions such as lukautim pik to hunt pigs', hukim pis to catch fish', etc.). It seems that the choice of ol as a plural marker was not an entirely fortunate one because of the difficulty of reducing this form to a phonologically reduced affix. ii) My data suggest that the position of ol in the noun phrase is not fixed in creolized Tok Pisin and that variation is found not only across creolized varieties in different localities but also within the speech of individual speakers. However, the favourite form is the one in which ol appears at the beginning of a noun phrase. Compare the following data: (48)

(49)

(50)

Speakers from Yip aged between 8 and 12: ol adj. N

adj. ol N

ol adj. ol N

ol dispela lain this group ol planti dok many dogs

dispela liklik ol tumbuna ol narapela ol tumbuna those little grandchildren the other grandchildren olgeta ol pis all the fish

Speakers from Malabang (first generation Tok Pisin speakers aged 25-35):

ol dispela lain

dispela ol man

this group

these people

ol faivhandet masalai

bikpela ol man

five hundred spirits ol lokal pipel the local people

big men

Speakers from Lae (children recorded by Sankoff in 1971): ol adj. N

adj. ol N

ol adj. ol N

ol dispela ol man these men'

sampela ol man

ol dispela man these men' ol sampela man

'some men'

'some men'

The development of the category of number in Tok Pisin

57

More fixity is found with some pronominal modifiers, such asolgeta 'all', which is always followed by ol and lokal local', which is always preceded by ol. It is not clear to me to what extent the position of ol is linked to certain lexical items. iii) While the principle of one form-one meaning is realized to a greater degree in creolized Tok Pisin than in other varieties, I have not been able to find any consistent difference in meaning between, for instance, ol sampela man 'some men' and sampela ol man 'some men'. I would predict, however, that, unless speakers settle for one of these two alternatives, a difference in meaning will develop in creolized Tok Pisin.4 iv) The prediction that plurality will be marked in parts of the sentence other than the noun phrase is partly fulfilled in Malabang Creole Tok Pisin, where a kind of agreement between plural noun subjects and reduplicated verbs is developing (cf. Mühlhäusler 1975b and 1976b). Examples of this construction are: (51)

a. ol pikinini i pilaipilai

the children are playing

as against

(52)

b. wanpela pikinini i pilai c. * wanpela pikinini i pilaipilai planti man i lainlain ol manmeri i bungbung

a child is playing a child is playing many men were lined up the people gathered

In conclusion, the following generalizations can be made about the developments in Tok Pisin's creolization phase: i) ii)

iii)

the differences between non-creolized and creolized Tok Pisin are slight rather than drastic whilst there is a definite tendency for rules to become more productive (and less restricted by environmental conditions), the endpoint of maximum simplification (cf. Mühlhäusler 1979a) has not yet been reached the amount of redundancy found in creolized Tok Pisin is not greater than that found in late expanded Tok Pisin.

The linguistic nature of present-day creolized Tok Pisin appears to be the direct result of the communicative requirements to which the language is subject, i.e. the principal raison d'etre for Tok Pisin in New Guinea is a lingua franca for people who do not share a common language. Thus it is the second language for most of its users and the speakers for whom it is a first language will have to communicate with others for whom it is not (cf. Mühlhäusler 1977). I have observed that young children speak a much faster and structurally more advanced variety of Tok Pisin than

58

Peter Mühlhäusler

their parents, but that they revert to the more conservative norms of second-language Tok Pisin as they grow older.

3.7. Developments during the Post-Pidgin Phase In the current literature on pidgins and Creoles there is widespread confusion between expansion of the developmental type discussed so far and restructuring of pidgins or Creoles when they come into renewed contact with their former lexifier languages. Apart from confusing an increase in the referential and non-referential power of a linguistic system with mere restructuring, a number of writers also appear to argue that the kind of linguistic processes leading to restructured or expanded varieties are very much the same. The data on pluralization certainly do not confirm this. Instead they indicate, as has been seen suggested by Bailey (1977), that mixing of systems of comparable complexity leads to unnaturalness. Before discussing this point any further I want to present some data on, as one of my informants put it, i gat singular na plural i kam insait long namba tri Tok Pisin, i.e. the adoption of English plural marking in Urban Tok Pisin. The first one to draw attention to this phenomenon was Hall (1956: 99-100). Hall documents plural -s for the following lexical items: Tok Pisin

Gloss

bepis des traktas yams yias kreps mails pauns praisis silings taims wiks

babies days directors yams years crabs miles pounds prices shillings times weeks

Whilst Hall does not provide information on the grammatical environment in which these forms were found, it seems clear that the presence of the plural -s is not determined by the degree of animacy of a noun. My own data suggest that the presence or absence of plural -s is neither determined by the animacy hierarchy nor by the grammatical environment nor, in the case of written Tok Pisin, by spelling. The following data were taken from

The development of the category of number in TokPisin

59

letters written in anglicised Tok Pisin. Sentences by the same writer are grouped together: (53)

wok bilong kainkain gavamen;

(54)

ol man (subject) wanem gavamen (direct object) long African kantris bilong mipela ol men planti men wok olsem taipis, post office clerk, nurses, radio announcer na sampela wok moa

(55)

ol Pacific Island sampela boys (subject) ol men skulmanki go long haiskuls_ long ol boys olgeta kantris ol i laikim man na men citisens ol pipol bilong narafelo kantris bilong dispela tupela stejes bipo yu givim kain points ol skulmeri i save pulimapim spes bilong o/ boys6 putim boys wantaim men hevim seperet haiskuls long boys

(56)

olgels (subject)7 mipela olgirl (subject) pilai long ol girl

(57)

dia friends olgeta evangelis, Bible tisa, pastors,* waitmisin

(58)

ol dispela lain liklik lain boys na girls? helpim friends_ long dresim cuts_ or wounds ol lapun (subject)

the work of various government officials men what government officials in the African countries of us women many women work as typists, post office clerks, nurses, radio announcers and in other jobs Pacific Islanders some boys the schoolgirls go to the highschools to the boys all countries appreciate both their male and female citizens people from other countries of these two stages earlier on you gave a number of arguments the schoolgirls take up the places for the boys, put the boys with the girls have separate highschools for the boys girls we girls have sexual intercourse with girls dear friends all evangelists, Bible teachers, pastors and white missionaries this group of people a small groups of boys and girls help your friends to dress cuts or wounds old people

60

Peter Mühlhäusler

Whilst I could adduce numerous further examples of the emergence of -s plural marking in Urban Pidgin, the above data seem sufficient to show that: i) ii)

iii)

The emergence of -s does not follow any of the hierarchies that determine pluralization in the non-anglicized dialects of TokPisin. Both -s -plurals and ol-plurals are found with recent loans (girls, pastors, ol visitor) and both encodings are found simultaneously with nouns (ol councillors). On the other hand, plural -s can also be attached to some old established lexical items, such as yams Yams' and yias Years'. There is a great deal of variability even with individual speakers. Whilst in many cases the functions are old and the forms new (as in gavamen 'government officials', stafs bilong haiskul the staff of the highschooP or ol bisnisgrups 'a business-group), there are cases where both form and function are new in mesolectal Tok Pisin. Even in those cases, English grammar may not be the only source.

Summing up these findings, one can say that the kind of mixing processes found when two linguistic systems of comparable complexity are in contact are quite different from those resulting from contact between a developing pidgin and other languages. In the former case borrowing appears to be by and large unrestricted and free to increase the unnaturalness of the developing mesolect, whereas in the former case borrowing is highly selective and restricted by universal principles of language development.

4. PLURAL MARKING IN OTHER PACIFIC VARIETIES OF PIDGIN ENGLISH 4.1. Introduction

Comparative data on the Pacific pidgins and Creoles are rare. This is largely due to the fact that, in comparison with Tok Pisin, all other pidgin and Creole varieties of English have rarely been described in any depth. As a result, the idea of a dynamic comparison, i.e. a comparison of developmental trends, cannot be met in full. Nevertheless, a number of linguistically interesting facts can be gathered even from a fairly limited comparison, particularly insights into the linguistic affinities and possible historical relationships holding between these languages.

The development of the category of number in TokPisin

61

4.2. Chinese Pidgin English A brief analysis of number marking in Chinese Pidgin English is called for, as this language has featured prominently in the development of the Pacific varieties of Pidgin English. Whilst the exact nature of the influence of Chinese Pidgin on the various pidgin and Creole varieties of English remains unclear, the following points appear to be generally accepted: i)

ii)

Trade between North America and the China Coast in the early 19th century promoted the use of Chinese Pidgin English in various parts of the Pacific. In at least some instances, this language served as a model for Pacific Jargon English. Towards the end of the 19th century, Chinese workers settled in a number of areas where a stable Melanesian variety of Pidgin English had developed (e.g. German New Guinea, Samoa and Queensland). Many of these Chinese had a knowledge of Chinese Pidgin English and used it as a language for communication with outsiders, including Pidgin speaking Melanesians. This may have led to mixing between Chinese and Melanesian Pidgin English in some instances.

The following pronoun system appears to have been used in virtually all varieties of Chinese Pidgin English on record:

1st 2nd 3rd

Singular mai yu hi

Plural

hi

No firm conventions for plural pronouns appear to have ever been established. Hall (1944: 97) remarks: In 'pure' Pidin these are the only personal pronouns; the use of aj T, mi Ίηβ', w'i Ve' or 3e they' is an Anglicism. Dennys (1878: 170) notes another way of overcoming the absence of plural pronouns in Chinese Pidgin English: 'We' and they' are rendered by thisee man, that man, the context implying when they are used in a personal rather than in a demostrative sense. The sentence Ί saw him' thus becomes 'my have see he', while *we went out' would be rendered 'Allo thisee man go out'.

62

Peter Mühlhäusler

Dennys (ibidem) further notes that this circumlocutory construction is not a result of substratum influence: There is not here any analogy between the Chinese forms (resembling our own) and the rude substitutes adopted. All native dialects have I, he, we you and they, the possessives (in Mandarin) being regularly formed by the addition of ft', of: thus, wo, I; wo ti, mine. No plural formatives are reported for Chinese Pidgin English nouns: 'Strictly speaking the original pidgin . . . does not make any distinction between singular and plural.' (Ehlerding 1936: 23, author's translation) The idea of plurality with nouns is signalled by the use of pre-nominal numerals or other quantifiers. Interesting for comparative purposes is the fact that a suffix -piecee can be added to numerals as well as to a number of other quantifiers and adjectives. The conventions underlying the use of piecee are rather complex and can not be discussed here. (A rather extensive discussion has been given by Bauer (1974: 125-127)) However, there appear to be certain parallelisms between the use of piecee in Chinese Pidgin English and the use of -felal-pela in a number of Pacific varieties of Pidgin English. 4.3. Early Pacific Jargon English Only few reliable data on Pacific Pidgin English prior to 1880 are available. These have been scrutinized by Clark (1977), who has the following information about plural marking: i) ii)

iii)

the plural marker all is attested for Tanna (New Hebrides) as early as 1859. Plenty + N is used as early as 1827, apparently in the English meaning of 'many, a great deal' rather than the simple plural meaning. altogether is found in New Guinea Tok Pisin, Solomon Islands Pidgin, New Hebridean Pidgin and Torres Straits Pidgin. According to Clark (1977: 14) 'in SI (Solomon Islands Pidgin) this appears to have lost its universal force, being reduced to a simple plural (pronoun or determiner), a process which may be underway in the New Hebrides'.

The development of the category of number in Tok Pisin

63

4.4. Papuan Pidgin English This language has recently been described by Mühlhäusler (1979b). All as third person plural pronoun and pronoun pluralizer (you all = 2nd pers. pi.) is documented as early as 1886. However, in later texts all is generally replaced by olgeta which, as in Solomon Islands Pidgin, has lost its original force and become a simple pluralizer. It is interesting that in some of the later texts olgeta begins to appear variably as the pluralizer of inanimate nouns (olgeta rais *rice'). 4.5. Queensland Plantation Pidgin English ('Kanaka English') The external and internal development of Queensland Kanaka English is at present being investigated by Dutton and Mühlhäusler. As in Solomon Pidgin and Papuan Pidgin, olgeta at the beginning of a noun phrase is the favoured plural formative. It appears to have lost the force of the English etymon. Ol 'all' on the other hand, appears to correspond more closely to English 'all'. Very often, nouns (including animates) are unmarked for number. In one of the most extensive recordings, Dutton's recordings of Tom Lammon, the following instances of plural nouns were found: (59)

paiv yia eitiyia ova siks yia plenti.. . marika olgeta: waitm@n kau olgetha bigbla m@n ten eka sam banana mo blankit sam pateita tri paun rut tu stiki

five years eigthy years over six years plenty of Americans all Europeans cows all the big men ten acres some bananas more blankets some potatoes three pounds roots two sticks

Queensland Kanaka English never developed to the degree of grammatical sophistication found in Tok Pisin. Thus, redundancy at sentence and phrase level appear to be absent, and there is some variation with regard to the choice of a plural formative. At the same time, due to the continued presence of English as a target language, one finds restructuring in the direction of English plural marking next to the development of a new plural marking system.

64

Peter Mühlhäusler

4.6. Bichelamar (New Hebridean Pidgin English) There appear to be good reasons for distinguishing a number of regional dialects of New Hebridean Bichelamar, notably those described by Camden (1975) and Guy (1974), the former working with Tangoan speakers, the latter with Sakao people. To what extent the regional dialects can be traced back to recruiting for different plantation areas (Samoa and New Caledonia on the one hand and Queensland on the other) remains to be established. Pluralization in the Sakao variety of Bichelamar of the Northern New Hebrides appears to be relatively close to the system of Tok Pisin, the main difference being the lesser degree of redundancy found there. Similarities exist with both pronouns and nouns. The personal pronouns are: Singular 1st pers. 2nd pers. 3rd pers. Dual 1st pers. (inclusive) 1st pers. (exclusive) 2nd pers. 3rd pers. 3rd pers. (collective) Plural 1st pers. (inclusive) 1st pers. (exclusive) 2nd pers. 3rd pers.

mi I yu you em he, she, it yumituvala you and I he or she or it and I mituvala the two of you yutuvala tuvala the two of them both of them tuketa yumi we (including you) mivala we (excluding you) yavala you people they ol (according to Guy 1974: 13)

With regard to the pluralization of nouns, Guy (1974: 18 + 20) gives the following main rules: #44 The plural is shown either by a numeral or by the particles ol preceding the noun: [tu trak} two cars; [sam trak] some cars, cars (indefinite); [ol trak] cars (in general), the cars (definite). #83 When a substantive phrase is a complement of another phrase, it may contain the plural particle ol only if its meaning is definite, not general: [wan basket] plong [yam] a basket for yams, i.e. a yam-basket (yam is indefinite); but: [wan basket] plong [olyam]a. basket for THE yams (definite); [tsif] i [kilim ] [olpik] the chief kills the pigs (definite) It appears that the exclusive use of ol as a definite plural marker in objects is an independent development in Bichelamar. My Tok Pisin data suggest

The development of the category of number in Tok Pisin

65

that the choice of ol is not normally associated with definiteness, though the absence of plural marking with objects (as in painim pik 'hunt pigs', rausim dok to throw out the dogs') is often associated with indefinite objects. On the other hand, Tok Pisin permits combinations such as sampela ol man 'some men' in both subject and object position. The main difference between Sakaon and Tangoan Bichelamar is that in the latter the third person plural pronoun ol is used in a small number of syntactic enviromments only, whereas olgeta is used in all other cases. Camden (1975: 43) remarks: "There is limited use of the pronoun ol they' as a stylistic variant of olgeta in the form blong ol, usually sentence finally. This form ol never occurs with a subject noun, and never occurs as a pronoun subject.' Whilst the behavior of ol and olgeta is more reminiscent of Papuan and Solomon Islands Pidgin, the use of ol or oli as 'predicate marker' or anaphoric pronoun is just the same as in Tok Pisin. Camden (p. 42) is wrong in claiming that: "The occurrence of the third plural predicate marker oli in Bislama forms a difference between Bislama and Neo-Melanesian, where the third person plural predicate marker is /. In Bislama, the form occurs as a variant in some areas in dependent clauses, and also occurs in most areas rarely, with the sense of a collective. Thus olgeta kenu i stop long smol aelan 'the entire fleet of canoes was on the small island' occurs beside olgeta kenu oli stop long smol aelan 'all of the canoes were on the small island'.' Examples of how this construction is used in Tok Pisin ('Neomelanesian') are found in texts 45,46 and 47 above. 4.7. Solomon Islands Pidgin English Unfortunately, few developmental data on this language are available. The following information was taken from Young (1976). The pronoun system is as follows:

66

Peter MühlMusler

1st person exclusive:

inclusive:

Singular

Dual

Plural

mi (I, me)

mitufala (we two; another and I) iumitufala (we two, you and 0

mifala (we three or more; others and I) iumi (we three or more; you, me and one or more others)

—

2nd person

iu (thou,you)

iutufala (you two)

iufala (you three or more)

3rd person:

hem (he, him, she, her, it)

tufala (those two)

olketa (they, them)

Whilst the pronouns are much more like those of Papuan Pidgin and Tangoan Bichelamar, the treatment of mass nouns is like in Tok Pisin. However, one suspects that one is dealing with an independent development here. Young (1976: 10) writes: Mass Nouns (uncountable nouns) Pidgin does not distinguish, as English does, between things that can be counted (a coconut, coconuts) and fluids and the like that can not (water, salt, machinery, love). All of the latter are treated as plural; they take plural articles (samfala, olketa) and speakers who add -s tend to add it to them. The pluralization of count nouns is done either by means of a preceding oloketa (for older speakers), olketa (the most common form) or olta (for many younger speakers). Young (1976: 10) remarks that: 'Younger speakers, educated in English, feel the need to add -s to plurals, especially of words recently adopted from English. This should be discouraged. No-one would think of adding --s to a Pidgin word that does not occur in English, to make *pikininis or *kaikais, and it is making life harder to have two kinds of nouns, those that mark the plural and those that do not.' I assume that depidginization of Solomon Islands Pidgin proceeds in a similarly unsystematic way as Tok Pisin (see section III, VI). 4.8. Torres Straits Pidgin and Creole English Because of its complex external history, the linguistic history of pidgin

The development of the category of number in Tok Pisin

67

and creole varieties in the Torres Straits is also very involved. This complexity is manifested i) in the simultaneous presence of different first and second language varieties and ii) in significant regional variation. The statements made below can only be tentative and much more fieldwork is needed before more definite answers can be given. The earliest source about the language (Ray 1907) contains the following information: Pronouns. Personal. Singular. 1,/, me; 2.you', 3.he, him.Plural. 1. we; 2. you; 3. they. In the singular "he" is used for all genders: woman he go, woman goes; night he come, night comes. "Me" appears to be used with intransitive more than with transitive verbs: me growl, I disapprove; me one fellow, I am alone. "Me" and "him" are also used preceding "/" and "he": me I go,I go; hem he go, he goes; him he run, he runs. In the plural "fellow" is often used after "we", or "you": all you fellow, all of you; we fellow got him, we understand. A dual appears in "you me", we two. This is especially used in the Eastern Islands for the inclusive person, the exclusive person being "other man ". Nouns. Number. Occasionally the word "plenty" is used to express a plural: plenty man go, many people go. Data on Torres Straits pidgin for the period between 1900 and 1960 are not available, with the exception of Landtman's Kiwai data. Though located within the political boundaries of Papua, Kiwai Island was part of the area where Torres Straits Pidgin was used. Moreover, because of the intensive contacts between Kiwai Islanders and Torres Straits Islanders, one could even speak of a common speech community. Thus, the Kiwai data can be regarded as representative of Torres Straits Pidgin. The pronoun system of Kiwai Pidgin exhibits a considerable amount of variation (Landtman 1927: 455): Pronouns. Of personal pronouns, "I" and "me" are used almost indiscriminately, sometimes in one and the same sentence: "I think me (or me fellow) go now." "Him" (often pronounced "hem") is sometimes used preceding "he," the pronoun thus being doubled: "Him (hem) he go now," "he goes now." In the plural "me" and "we" are both used: "Me altogether man come this place," "we have (all) come here." "You me" expresses a kind of dual, by adding "alltogether man," "plenty man," etc., also plural. Expressions such as "He come kill we" are ordinary. As in the case of nouns, personal pronouns, too, are frequent-

68

Peter M hlh usler ly marked out by the accompanying word "fellow": "me fellow" (not "I fellow"), "you fellow," "him fellow" (not "he fellow"), in the plural: "me fellow" or "we fellow" (very often "you me two fellow"), occasionally "we me fellow"; "you fellow"; "them fellow" (also in the nominative; "they fellow" is seldom used).

The system of noun pluralization is typical of an early stabilized pidgin,i.e. no obligatory plural marking and no redundancy at phrase level. Landtman (1927: 454) writes: Nouns. All nouns are used in the singular number, no inflection whatever occurs, and variations of meaning, corresponding to inflected forms in European languages, are expressed by separate words. In exceptional cases the natives show a certain notion of the existence of a plural "s"; thus the form "boys," of such frequent use in the speech of white people, is also met with in individual expressions in pidginEnglish The plural is shown by separate words which imply number: "plenty man," "full-up man," etc. Dutton's data recorded in the 1960s, do not differ greatly from those recorded by Ray and Landtman (Dutton 1970: 149): Pronouns. The following subject pronouns were observed: ai, mi, T; yd, yu:, *you' (sing, or pi.); i, em, 'he, it'; mitu:, Sve (2)';m/fw: Nako, 'Nako and Γ; yutu: "you (2)'; Semtu:, those (2)'; wi, mipeU, yumi:, 'we'; SemplA, they'. The same forms seem to be used for object pronouns, except that mi (not ai) is always used of'me'. These observations are similar to those of Ray except that the following pronouns were not observed: 1. 'he' for the female pronoun 'she'; 2. 'me' used preceding T, as, for example, in Ίηβ I go'; 3. 'fellow' suffixed to *we' for the pronoun 'we'. Nouns. As in Aboriginal English no distinction is made between singular and plural number of nouns in the form of the noun, for example, to:tl eg, turtle egg(s)'; plei ma:bl de: 'play marbles there'. When required, number will be indicated by some numeral adjective such as WAH, One'; plenti, 'plenty'. Unfortunately, few plural subjects can be found in Dutton's texts. I suspect that some plural marking would have occurred in this more favourable environment. An interesting construction also recorded by Dutton is

The development of the category of number in Tok Pisin

69

kese ol to:tl 'caught many turtles', where the original meaning of all is reduced. The grammatical sketch provided by Laade (1971) is interesting in that one finds those features of the early jargon recorded by Ray, which were absent in Dutton's data, for instance the use of he for 'she'. Unfortunately, Laade does not specify what regional or social variety is reflected by his analysis. He lists (115): PRONOUNS. The personal pronouns are the following:

I you

me (fellow) you (fellow)

he, she, it

he, him ('em); him ('em) often precedes he: 'em he go now you me (fellow) we (fellow), or, us fellow you two (fellow) you fellow them two (fellow) them fellow; or, all he (in contrast to singular em he)

we (2 persons) we (more than two) you (2 persons) you (more than two) they (2 persons) they (more than two)

The twofold plural forms correspond with those of the native language. With regard to noun pluralization Laade merely observes that: "The plural is indicated by plenty: plenty man he go 'many people are going (or: went)'.' My own data on pluralization from a Creole speaker from Darnley Island suggest a significant development with plural marking in the creolized varieties. Noun pluralization is indicated by means of ol and ol dem and a clearcut distinction is made between ol 'plural' and oltugeta 'all'. Examples from Torres Straits Creole English as recorded by myself in 1978 include: (50)

dem tu pikinini ol kain dem kaikai ol fofada bilong mi ol kalad res oltugeta dok sam ol man dem pikinini ol pikinini

the two children' 'various kinds of food' 'my ancestors' the coloured races' 'all dogs' 'some men' 'children' 'children'

It appears as if there is a semantic distinction developing between ol and dem, the latter corresponding to English definite plurals. Again, more data from different lects of the language are needed to establish the facts.

70

Peter Mühlhäusler

4.9. Assessment of the comparative evidence It is my conviction that only the comparison of the developments in pidgin languages can lead to a true account of their historical interrelationships. As the data on hand were mostly non-developmental, no strong claims can be put forward here. All of the languages investigated proceed from a numberless jargonstage to number-marking in certain semantactic contexts. The earliest type of number marking is that using plenty (documented for the earliest jargon as well as second-language varieties such as Torres Straits Pidgin up to the present). With regard to the pluralizers ol and ol(tu)geta there appear to be two clear sub-groups, one (ol) including some varieties of Bichelamar, Samoan Plantation Pidgin and Tok Pisin, and the other (plgeta) including other varieties of Bichelamar, Queensland Plantation Pidgin, Papuan Pidgin, and Solomon Islands Pidgin. If it is true that genuine plural marking develops in stable pidgins and not in unstable jargons, it would seem warranted to trace the two traditions to the Samoan and Queensland plantations respectively, where the first stable pidgins were found. Unfortunately, data on New Caledonian Pidgin (the third plantation pidgin) are very scarce. Schuchardt's (1881) data suggest that plenti was the only way of pluralizing there. The fact that New Hebrideans went to both Samoa and Queensland may explain dialectal differences in Bichelamar. In addition to these two groups there is Torres Straits Pidgin which exhibits a number of patterns of pluralization, those involving plenti, ol and possibly ol(tu)geta. It appears that the resulting conflict is now being resolved through the development of yet another pattern of plural marking, i.e. the use of the third person plural pronoun dem. My comparison suggests that in no case can the origin of a grammatical category such as pluralization be traced back to a proto-pidgin. It provides weak support for the claim that the linguistic conventions that were developed in the Pacific plantation areas served as starting points for the development of distinct varieties in the recruiting areas.

5. THE ROLE OF SUBSTRATUM AND SUPERSTRATUM INFLUENCES IN THE DEVELOPMENT OF PLURALIZATION

5.1. Introduction In the past, grammatical structures in pidgins and Creoles were regarded as reflecting either superstrate or substrate grammars. At the same time,

The development of the category of number in Tok Pisin

11

little room was given to the possibility that some of the grammatical developments in these languages might be independent developments. In a recent paper (Mühlhäusler 1979a) I have given arguments in favour of independent development, particularly in the morphological and syntactic components of pidgins and Creoles. The principal argument is that the development of grammar is, as it were, preprogrammed and that developing systems of restricted complexity are incapable of integrating complex structures from the languages with which they are in contact. For our specific problem this means that, at the time of maximum contact between English (the superstratum) and Tolai (the alleged substratum) Tok Pisin had not developed to an extent which would have made borrowing of plural marking from these languages likely. 5.2. The Role of Substratum Influence in Number Marking It is still widely accepted that Tok Pisin 'consists of loan words from different languages, native and European, used in a grammatical system that is basically Melanesien' (Reed 1943: 274-5). This belief has led to the practice of assigning a Melanesian origin to Tok Pisin grammatical structures for which no English source can be found. A static comparison between present-day Tolai and present-day Tok Pisin will in some cases give support to such a belief. As we will see, however, very little structural borrowing occurred at that point in Tok Pisin's history when it was in closest contact with Tolai and related Melanesian substratum languages. In the absence of an exhaustive grammar of Tolai definite statements about areas of overlap between Tolai and Tok Pisin pluralization have to remain tentative. However, some very useful information was found in Mosel (1979) and a number of points can be resolved. Plurality in Tolai and the related languages of the Duke of York Islands and New Ireland may be expressed by reduplication or pre-nominal plural markers which are either numerals or indefinite quantifiers. The plural markers are not identical with the third person plural pronoun. The use of reduplication to express plurality is a straightforward case of iconic encoding and one could thus expect this construction to appear early in the life of Tok Pisin, particularly if reduplication is also used as a mechanism for indicating plurality in its principal substratum language. This, however, is not the case. As I have documented more fully elsewhere (Mühlhäusler 1976a: 347 ff and 515 -533), reduplication entered the structure of Tok Pisin at a very late stage. Thus, Hall (1943: 194) states that 'reduplication is not of grammatical significance in Pidgin', a finding which was also arrived at by Lay cock in the 1960s (1970: xxv). Some writers have drawn attention to the item sipsip 'sheep' as an instance of pluralization signalled by reduplication. If this is so, the fact that only very few

72

Peter Mühlhäusler

other nouns appear in reduplicated form (e.g. meme 'goat' and kainkain lands of) would suggest that a productive pattern of Tolai became fossilized in Tok Pisin. Against such a view stands the fact that the earliest dictionaries of Tok Pisin (Brenninckmeyer 1924 and Borchardt 1926), i.e. the dictionaries written at the time of maximum impact of Tolai, list sip 'sheep' and me 'sheep, goat* in unreduplicated form. I have suggested elsewhere (1975a: 201) that the item sip 'sheep' was reduplicated to distinguish it from sip 'ship'. An even more plausible example of reduplication for the purpose of disambiguation is that of sip 'jib' which became sipsip in the 1930s, as it would have been dysfunctional to have the same form referring simultaneously to a ship and one of its main parts. In present-day Tok Pisin (particularly in the Sepik and Manus areas) the use of reduplication to signal noun plurality is gradually becoming more common (cf. Mühlhäusler 1976a: 530-31), as in planti bükbuk plasta i gat hulhul haphap graun siot i gat kalakala

'many protruberances' 'plaster with many little holes' 'plots of land* 'a many-coloured shirt'

Note incidentally, that pluralization in this case does not follow the animacy or accessibility hierarchies but appears to be determined by the semantic factor of 'a larger number of items that are more or less evenly distributed'. An important difference between the grammar of Tolai and Tok Pisin is that the use of reduplication excludes the simultaneous use of the plural marker umana, whereas in Tok Pisin the following sentence is perfectly well formed: (61)

ol i pulimapim kabora long ol bekbek

they filled the copra into the bags

Differences in the conventions for redundant plural marking are also found with nouns that themselves denote plurality. In Tolai the plural marker can be 'redundantly combined' (Mosel: personal communication) with such nouns. In present-day Tok Pisin this is possible (e.g. ol lain 'a group', ol pipol *people', etc.) but it is a recent development and phrase-level redundancy was not found in Tok Pisin when contact with Tolai was intense.10 The data discussed by Mosel (1979: 120-124) suggest that the conventions for obligatory (in the case of animates) and optional (for all other nouns) use of plural marking are similar in present-day Tok Pisin and Tolai. However, at the time of maximum contact between the two langu-

The development of the category of number in Tok Pisin

73

ages, no obligatory plural marking was found in Tok Pisin. That animates are more readily marked for plural than inanimates appears to be a function of communicative requirements rather than a result of language contact. A similar argument could also be put forward for the next case of parallelism between the two languages (quoted from Mosel 1979: 124): 'In both languages the equivalents of English mass nouns are combined with the plural marker or cardinal numerals, if a certain amount of specific quantities of the noun referent is to be expressed.' A better case for Tolai influence may be made for the restructuring of the pronoun system. The distinction of inclusive and exclusive first person plural and the distinction between dual, trial and plural is also made in Tolai. It was not made in Tok Pisin's predecessor Samoan Plantation Pidgin, at least not consistently). However, such distinctions are made in numerous Oceanic languages and should be regarded as a Sprachbund feature rather than due to borrowing from an individual language. Walsh (1978: 190) argues for an ΕΑΝ (Eastern Autronesian influence) but again, this may be too restricted a view. Whilst the preceding comparison between Tok Pisin and Tolai is not exhaustive, the following conclusions seem warranted: i) ii) iii)

many aspects of Tolai pluralization are not found in Tok Pisin many aspects of Tok Pisin pluralization are not found in Tolai if the developing grammatical system of Tok Pisin rather than an abstract synchronic state is taken as the basis for a comparison, substratum influence begins to appear an unlikely explanation for most aspects of Tok Pisin plural marking.

5.3. The Role of Superstratum Influence in Tok Pisin Plural Marking Opposed to the view that the grammar of pidgin languages is fundamentally that of a substratum language is the view that one is dealing with a slightly modified and simplified grammar of an European language.11 For Tok Pisin, this view has been most strongly put forward by Hooley (1962: 116-127) and Hall. The latter writes (1966: 117): 'In Haitian Creole, the proportion of French structure is both greater and more fundamental than that of African-type structure; and the same is true of Chinese Pidgin English, Neo-Melanesian [= Tok Pisin, PM.], Sranan, Gullah, etc., in relation to English and the various substrata involved.'

74

Peter Mühlhäusler

A close examination of English and Tok Pisin pluralization indicates that there are very few shared features in this area of grammar. It would lead us too far if all aspects of plural marking in English were to be discussed here. Instead I will point to some prominent areas of similarity and differences. i)

ii)

iii)

The loss of plural affixes can be regarded as either a simplification of the English and German systems of plural marking or else a direct carry-over from foreigner-talk. At no stage in the development of Tok Pisin, with the exception of the post-origin phase, was there pluralization by affixation. The presence of fossilized forms such as masis 'match, matches' or anis 'ant, ants' suggests that speakers of the superstratum language used bound plural formatives at least some of the time. Tok Pisin and German were in contact when the former language had already reached a considerable degree of stability. In spite of this, the German plural formative -en was not adopted as a productive grammatical device. It has survived in a few fossilized lexical items, including binen 1)ee, bees', palmen 'palm tree, or palm trees' and sirsen 'cherry, cherries'. English plurals are redundant at phrase level, i.e. plural -s appears without regard for quantifiers in the same phrase as in many houses, some cars, etc. Whilst this redundancy is also found in present-day Tok Pisin (through to a much lesser degree), it was completely absent in earlier developmental stages. One could thus only argue that this aspect of English grammar is being gradually borrowed by Tok Pisin. It could be argued that the tendency in English for syntactic plurals to represent semantic plurals was carried further in Tok Pisin, and that Tok Pisin wanpela trausis 'a pair of trousers' or wanpela sisis 'a pair of scissors' constitutes a simplification of English. However, such argumentation raises the question of the extent to which a pidgin can be simplified and changed and still be basically English.

Next to the debatable similarities of i) to iii), there are a number of important differences, including: iv)

v)

The English distinction between mass and count nouns is neutralized in Tok Pisin. Mass nouns can be pluralized in the latter language. For all but some very progressive (creolized ur urban) varieties of Tok Pisin no plural marking can take place in the predicate. Compare:

The development of the category of number in Tok Pisin

75

(62) (62')

ol i men ol i ol men

vi)

Animates are marked for dual, trial and plural in Tok Pisin (a fact which is reflected, for instance, in the choice of pronouns), whereas in many grammatical contexts the distinction singular vs. plural can be neutralized. Nothing comparable is found in English. Tok Pisin has an optional category of distributional plural (typically signalled by means of reduplication) which is not found in English.

vii)

they are girls they are girls

Some more remarks on differences in the grammatical functions of pluralization are discussed in Mühlhäusler (1976a). It seems that statements about the virtual identity of Tok Pisin and English grammar are the result of superficial and impressionistic comparisons and can not be upheld any longer for this area of grammar. 5.4. Parallelisms between the Development of Tok Pisin and Natural L j and L 2 Acquisition of English In recent years, considerable empirical support has come forward for the hypothesis that there are important parallelisms between all types of human language learning, including the processes that lead to the establishment of pidgin languages. In the case under discussion this means that one should be able to discover similarities between the structure of Tok Pisin pluralization and that found with very young native speakers of English or adult foreign learners. Whilst a comparison between these three systems is useful, there are certain limitations: i)

ii) iii) iv)

In the case of the development of a pidgin (after the initial phase at least) we are dealing with an untargeted development, whereas in the case of first and second language learners of English we are dealing with targeted learning. Most studies of targeted learning investigate whether learners have reached the target or not. Deviations from the target are seldom discussed in any detail. The distinction between plural as sementic category and morphological plural marking is not always made. None of the studies on first or second language learning investigated considers that plural marking may be dependent on semantic or syntactic hierarchies of the kind used in the discussion of Tok Pisin's development.

Early research into the acquisition of plurals by children (e.g. Berko 1958) concentrates on the purely formal aspects of English pluralization. An im-

76

Peter Mühlhäusler

portant finding is that -s and -z plurals are mastered before -iz plurals. Thus, according to Berko (1958: 163): 'the children can add /-s/ or /z/ to new words with a great deal of success. They do not yet have the ability to extend the /iz/ allomorph to new words, even though it has been demonstrated that they have words of this type in their vocabulary.' The data from Tok Pisin seem to suggest that this hierarchy was of no relevance in the early development of this language. English plural morphemes were simply dropped. In the few cases where they were borrowed they became part of the Tok Pisin word stem, e.g. masis12 'match, box of matches', trausis trousers, pair of trousers' or (in some varieties) gas 'guts'. More recent researchers (e.g. Anisfeld & Tucker 1967) concentrate on phenomena other than the learning of plural morphemes. They criticize Berko in a number of areas, notably (p. 1203): 'Although this was a carefully conducted study, several limitations restrict the inference which may be drawn from it. First, Berko tested only productive control of pluralization rules and did not attempt to study receptive control. Further, the children were required, at all times, to produce a plural form, having been given the singular form. A broader understanding of the child's concept of singularity-plurality may be gained from the analysis of his ability to produce the singular form when he has been given the plural form.' Anisfeld and Tucker (op. cit.) discuss a number of issues which are also relevant to the question of parallelisms between first-language acquisition and pidginization. Thus, they point out that the semantic idea of plural can be expressed by means other than plural morphemes (p. 1202): 'With regard to plurality, some investigators (...) have suggested that a numeral preceding a noun may be construed by the young child as a sufficient marker of plurality, resulting in such constructions as 'two book' and One-two shoe'. Such observations suggest that any attempt to study the nature of the child's pluralization rules must focus not only on his knowledge of 'adult' rules but also on any other rules which he may adopt temporarily.' This suggests that the lack of redundancy at phrase level is a feature of both pidgins and child language. Another finding concerns the iconicity of plural marking (p. 1207):

The development of the category of number in Tok Pisin

77

"The results clearly indicate a strong preference for longer forms as plurals. In only two of the 72 trials did Ss choose the original singular form to serve also as the plural form.' In the development of Tok Pisin (and pidgins in general) we find a similarly pronounced tendency towards iconic encoding of grammatical categories. A detailed discussion can be found in Mayerthaler's treatment of morphological naturalness (1978). Unfortunately, I have no data on the appearance of plural marking in semantic environments in child language. I suspect that the order in which plural marking appears in the various parts of sentences is similar to that found in the development of Tok Pisin and I would like to see empirical evidence on this point. Data on the acquisition of plural marking by adult second-language learners of English are similarly scarce. In addition, most of these studies are based on observations in a classroom situation and not on the observation of natural second language acquisition. According to Fathman (1977: 32), regular plural forms are among the first morphological and syntactic rules learned by second language learners, whilst irregular plurals are learned considerably later. Madden et al. (1974: 241) show that the order in which the grammatical category of plural is learned relative to other function words' differs in the various language learning contexts. Whereas different researches have found conflicting evidence, one clear fact emerges from their comparison, i.e. that the time at which pluralization is acquired by second-language learners does not coincide with the time at which this structure is introduced by the teacher. Compare the order of acquisition in a number of different learning situations as compiled by Madden et al. (1974: 241): Table IV. The order of Acquisition of Grammatical Categories Adults-ESL

children-ESL

children-Llj

children-Llz

ESL-Teachers

ing con cop plural articles con aux past irreg third per poss

plural ing con cop con aux articles past irreg third per poss

plural ing past irreg articles con cop poss third per con aux

ing plural past irreg poss articles third per con cop con aux

cop ing aux plural third per past irreg

None of the above statistics provides any evidence about the various subtypes of pluralization, in particular, the order in which plural formatives appear in phonetologically and semantically differing environments. As

78

Peter Mühlhäusler

I have not analysed a sufficiently large body of data, I can only make suggestions. With regard to the phonetological conditions, the order of acuisition of first-language learning appears also to apply in natural second-language learning by adults and children. This means that plural [s] and [z] appear before [iz]. In a text spoken by a 10 year-old learner of English13 the following plural forms were found: (63)

birds house mouth doors shoes feet apples

In my data I have also found evidence that the plural formative is added later to inanimate nouns than to animate nouns. The following data were found in English letters written by Papua New Guineans to the editor of Wantok newspaper: (64)

writer number

plural -s

no plural -s

1

high schools primary schools boys

high school

2

behaviours minutes aspects

school

3

girls boys stores

thing

The above data also illustrate the principle that when a rule is applied in new environments it tends to begin as a variable rule. In the same data I have found some supporting evidence and no counterevidence to the claim that the plural formative will appear first in the most prominent positions of the sentence and later in other positions. Thus, in one letter the writer uses boys in both subject and object position but writes one of the student. I am fully aware that the data adduced here are of a very sketchy nat-

The development of the category of number in Tok Pisin

79

ure indeed but I am confident that future research will support the claims made. At this point, the following conclusions appear to be warranted: i) ii)

iii)

iv)

The parallelisms between first and second-language acquisition are only partial. On the whole, the parallelisms between the expansion of a pidgin and natural second-language development are greater than those between pidginization and first-language learning. Morphological plural marking is a relatively early process in second-language learning. Its absence in any known pidgin or Creole variety of English is indicative of the social withdrawal of the target language. Research that concentrates on gross superficial phenomena only is unlikely to provide the answers asked about the nature of language learning.

6. CONCLUSIONS

This paper is part of a larger project on the linguistic and extralinguistic history of pidgin and Creole varieties of English in Australia and the Pacific, carried out in collaboration with scholars from the Australian National University. The present paper constitutes an attempt to develop a framework suited to the determination of linguistic relationships among rapidly changing/developing systems. At present, a paper on the development of causative encoding (Edmondson & Mühlhäusler forthcoming) is also available and I hope to investigate the history of a selected number of lexical and syntactic aspects of these languages in the near future. As this project is still at an early stage of development many of the findings remain tentative. The conclusions arising from the present paper are: 1) Because of the enormous amount of variation along the dimensions of time, geographical and social space, any abstract 'common core' descriptions of a language such as Tok Pisin will fail to adequately characterize its grammar. 2) The only methodologically sound way to compare developing systems such as pidgins is to contrast developments, not abstract states.14 3) A comparison between present-day Tok Pisin and either English or Tolai is likely to lead to false conclusions about the role of substratum and superstratum influence in the formation of a pidgin language. Such a comparison fails to take into account universals of linguistic expansion.

80

Peter Mühlhäusler

4)

At the time of maximum contact with Tolai, Tok Pisin had not reached the structural complexity which would have enabled it to borrow syntactic or morphological constructions. 5) The development of number marking in Tok Pisin goes from a lexico-semantic to a syntactic process. Paralleling this development there is a change from context-dependency to sentence- and phraselevel redundancy. 6) The main function of lower size-level redundancy is that it creates the preconditions for an increase in speech tempo and natural phonetetological processes. 7) There is a fundamental difference between natural development and restructuring as a result of language contact. The data examined suggest that the anglicized urban varieties are not only becoming different but also more marked and hence more difficult in absolute terms. This development seems particularly dangerous in the light of the fact that Tok Pisin remains a second language for most of its users. 8) Natural linguistic development proceeds in an orderly way for all speakers undergoing a similar linguistic development. This means that the order in which new constructions enter a developing language is predetermined. In the case of developing pidgins this order appers to be motivated mainly by pragmatic considerations. For plural marking this means that plural formatives appear first with nouns that are semantically central and in prominent sentence positions. 9) There appears to be evidence that the implicational order of plural marking in a developing pidgin differs from the order in natural first-language acquisition. It is comparable, however, to secondlanguage learning. 10) Whilst it is not possible, in principle, to construct a family tree for the pidgin and Creole varieties of English in the Pacific and Australia, a dynamic comparison can demonstrate possible historical links. The evidence presented in this paper suggests: i) That Chinese Pidgin English has contributed very little grammar to other Pacific pidgins.15 ii) In the case of Tok Pisin, English was quite remote as a target language, iii) The shared grammatical properties of the present-day pidgins and Creoles are often due to employment of Pacific Islanders from many different areas in a few plantation centres. 11) If it could be shown that the souce of new grammatical categories such as number marking is universal grammar, then the develop-

The development of the category of number in Tok Pisin

81

mental data from Tok Pisin could be of considerable interest to research into linguistic universals.16 NOTES 1. An exception is a sentence quoted by Neverman (1929: 257): all fellow men he cry the women are singing'. 2. Mani 'money' is also found with numerals in givim planti tausen mani 'to give many thousand (units of) money'. 3. This reflects the more general principle that in second-language learning, perception takes precedence over production in the initial phases. It also explains the absence of phonetological rules in pidgins until late in their development. 4. This would be comparable to the development in the area of future marking (cf. Sankoff and Laberge 1974) where the original future marker baimbai has come to mean 'sometime in the indefinite future', whilst the condensed forms bai and ba refer to events in a definite future or events which are closely subsequent to other events. 5. Note that gavman in Tok Pisin usually refers to a person ('government official') and not an institution. The form gavamen illustrates the principle that mesolectal forms often combine new forms and old functions (cf. Bickerton 1979). 6. Note the variation between boys and ol boys, men and ol men. 7. Note that the form with the Tok Pisin spelling gets the -s whilst the form spelled in the English way doesn't. This is quite unexpected. 8. The rule for pluralization in enumerations is: either ol or no pluralizer followed by a string of numberless nouns. A less favoured alternative is to have ol with all nouns in an enumeration. It is possible to mark just one noun in the middle of a string for plural. Quite far-reaching violations of Tok Pisin grammar are thus involved in the urban mesolects. 9. This is a counterexample to Bickerton's generalization (1979) that in mesolectal continua old functions appear in new forms. Instead we find both new forms and a new rule (borrowed from English) which allows pluralized nouns after classifying nouns such as lain 'group'. In Rural Pidgin *lain ol man na ol men is ungrammatical. 10. Whilst speakers of Tolai and related languages formed the bulk of Tok Pisin speakers in its formative years (between, say, 1880 and 1910) and whilst the Rabaul dialect of Tok Pisin was a prestige variety up to about 1930, the proficiency in Tok Pisin of Tolais has declined ever since (e.g. remarks by Orken 1954) and the main centres of diffusion are now towns such as Madang, Lae and Goroka on the New Guinea Mainland. I think it is fair to assume that Tolai ceased to be a lexifier or grammaticalizer language after 1920. 11. The case of Tok Pisin is complicated by the fact that, apart from English, German was a superstrate language from 1884 to 1914. However, even during German colonization, English was not completely withdrawn. 12. Whether the plural ending was borrowed or not does not appear to depend on its phonological properties, as can be easily seen from the above examples. 13. It is assumed that the sequence of acquisition is by-and-large independent of the learner's first-language, a claim for which recent publications on interlanguage provide ample evidence, (cf. articles in Corder (ed.) 1977) 14. This principle also applies to contrastive linguistics in second-language learning. What should be contrasted is the learner's developing grammar and both his first and the target language and not just the learner's first language and the target language. 15. This follows from the fact that number marking is a relatively early developmental feature. If Chinese Pidgin English did not contribute to this part of grammar, it is unlikely to have been of influence in developmentally later parts of grammar. 16. An example is the claim that 'mass nouns are but one subtype of plural nominal expressions' (Gee 1972: 2).

82

Peter Mühlhäusler

BIBLIOGRAPHY Anisfeld, Moshe & Tucket, G. Richard (1967), English Pluralization Rules of six-yearold Children, Child Development 38/1,1201-1217. Bailey, Charles-James N. (1977), Variation and Linguistic Analysis, Papiere zur Linguistik 12,5-56. Bauer, Anton (1974), Das Melanesische und Chinesische Pidginenglisch, Regensburg, Verlag Hans Carl. Berko, Jean (1958), The Child's Learning of English Morphology, Word 14,150-177. Bickerton, Derek (1979), Decreolization and the Post-Creole Continuum, paper presented at the Conference on Theoretical Orientations in Creole Studies, St. Thomas. Borchardt, Karl (1926), Tok Boi Wörterbuch, Unpublished MS, Manus. Borchardt, Karl (1930), Anleitung Zur Erlernungdes Tok Boi, Mimeographed Manus. Brenninckmeyer, Leo (1924), Einführung ins Pidgin Englisch —Ein Versuch, Unpublished manuscript, Kamacham, New Britain. Burnell, F.S. (1915),Australia versus Germany, London, George Allen & Unwin. Camden, William G. (1975), Parallels in Structure of Lexicon and Syntax between New Hebrides Bislama and the South Santo Language spoken at Tangoa, paper presented at the International Conference on Pidgins and Creoles, Honolulu. Cassidy, Frederic G. (1971), Tracing the Pidgin Element in Jamaican Creole. In: Hymes, Dell (ed.), Pidginization and Creolization of Languages, Cambridge University Press. Clark, Ross (1977), In Search of Beach-La-Mar, Working Papers in Anthropology, Archeology, Linguistics, Maori Studies. No. 48, University of Auckland. Corder, S. Pit (ed.) (1977), Actes du Seme Coloque de Linquistique Appliquee de Neuchatel, Geneva, Librairie Doz. Dennys, Nicholas B. (1870), Pidgin English, The Nation 11,118-119. Detzner, H. (1921), Vier Jahre unter Kannibalen, Berlin, Verlag Scherl. Dulay, Heidi C. & Burt, Marina K. (1974), Natural Sequences in Child Second Language Acquisition, Language Learning 24/1,37-53. Dutton, Thomas E. (1970), Informal English in the Torres Straits. In: Ramson, W.S. (ed.), English Transported, Canberra, Australian National University Press, pp. 137-160. Dutton, Thomas E. (1973), Conversational New Guinea Pidgin, Canberra, Pacific Linguistics, D-12. Edmonson, Jerold A. & Mühlhäusler, Peter (forthcoming). On the Rise of Causatives and Grammatical relations in a Pidgin Language (Tok Pisin), Technische Universität Berlin, Working Papers in Linguistics 7. Ehlerding, Carl W. (1936), Pidgin-Englisch, Volk und Welt, 21-25. Fathman, A. (1977), Similarities and Simplification in the Interlanguage of Second Language Learners. In: Corder (ed.), 30-38. Fischer, Hans (1966), Cargo-Ideen,/4«r/z/Opoi 61, 49-97. Gee, James P. (1972), The Mass-Count Distinction in English, Stanford Occasional Papers in Linguistics 2, Stanford University. Guy, Jacques B.M. (1974), Handbook of Bichelamar, Canberra, Pacific Linguistics C-34. Hall, Robert A. Jr. (l943),Melanesian Pidgin English, Baltimore, Linguistic Society of America. Hall, Robert A. Jr. (1944), Chinese Pidgin English: Grammar and Texts, Journal of the American Oriental Society 64, 95-113.

The development of the category of number in Tok Pisin

83

Hall, Robert A. Jr. (1956), Innovations in Melanesian Pidgin (Neo-Melanesian), Oceania 26, 91-109. Hall, Robert A. Jr. (1966), Pidgin and Creole Languages, Ithaca, Cornell University Press. Hooley, Bruce A. (1962), Transformations in Neomelanesian, Oceania 33, 116-127. Laade, Wolfgang (1971), Oral Traditions and Written Documents on the History and Ethnography of the Northern Torres Strait Islands, Wiesbaden, F. Steiner Verlag. Landtman, Gunnar (1927), The Kiwai Papuans of British New Guinea, London, MacMillan. Laycock, Donald C. (1970), Materials in New Guinea Pidgin, Canberra, Pacific Linguistics, D-5. Madden, Carolyn et al. (1974), Acquisition of Function Words by Adult Learners of English as a Second Language, Papers from the 5th Annual Meeting of the North Eastern Linguistic Society, Harward University, 234-245. Mayerthaler, Willi (1978), Morphologische Natürlichkeit, Habilitations-Thesis, Technische Universität Berlin. Mihalic, Francis, SVD (1971), The Jacaranda Dictionary and Grammar of Melanesian Pidgin, Brisbane, Jacaranda Press. Mosel, Ulrike (1979), Tolai and Tok Pisin, typescript, Munich University, to be published in Pacific Linguistics. Mühlhäusler, Peter (1974), Pidginization and Simplification of Language, Canberra, Pacific Linguistics B-26. Mühlhäusler, Peter (1975a), Reduplication and Repetition in New Guinea Pidgin. In: McElhanon (ed.), Tok Pisin i go We?, Port Moresby, Linguistic Society of Papua New Guinea, pp. 198-214. Mühlhäusler, Peter (1975b), Sociolects in New Guinea Pidgin, ibidem, pp. 59-75. Mühlhäusler, Peter (1976a), Growth and Structure of the Lexicon of New Guinea Pidgin, PhD. Thesis, Australian National University to appear as Pacific Linguistics, C-53. Mühlhäusler, Peter (1976b), The Category of Number in New Guinea Pidgin, Linguistic Communications 13, 21-37. Mühlhäusler, Peter (1977), Creolization in New Guinea Pidgin. In: Wurm, S.A. (ed.), New Guinea Area Languages and Language Study, Vol. Ill, Canberra, Pacific Linguistics, pp. 567-576. Mühlhäusler, Peter (1979a), Structural Expansion and the Process of Creolization, Paper presented at Conference on Theoretical Orientations in Creole Studies, St. Thomas. Mühlhäusler, Peter (1979b), Papuan Pidgin English Rediscovered, in: Proceedings of the Second International Conference on Austronesian Linguistics, Canberra, Pacific Linguistics, C-61, pp. 1377-1446. Nevermann, Hans (1929), Das Melanesische Pidjin-Englisch, Englische Studien 63, 219-232. Orken, M. (1954), Pidgin-English, South Pacific 7, 863. Ray, S.H. (1907), The Jargon English of Torres Straits. In: Reports of the Cambridge Anthropological Expedition to Torres Straits, Vol. Ill, Cambridge, pp. 251-254. Reed, S.W. (1943), The Making of Modern New Guinea, Philadelphia, American Philosophical Society Memoir, No. 18. Sankoff, Gillian & Laberge, Suzanne (1972), On the Acquisition of Native Speakers by a Language, Kivung 6/1, 32-47. Sankoff, Gillian & Brown, Penelope (1976), The Origins of Syntax in Discourse, Language 52/3,631-666.

84

Peter Mühlhäusler

Schuchardt, Hugo (1881), Kreolische Studien V: über das Melaneso-Englische. Sitzungsberichte 105. Wien. Akademie der Wissenschaften. Schuchardt, Hugo (1889), Beiträge zur Kenntnis des englischen Kreolisch, II. Melaneso-Englisches.Englische Studien 13,158-162. Schuchardt, Hugo (1914), Die Sprache der Saramakkaneger in Surinam, Amsterdam, Johannes Müller. Thomas, Julian (1886), Cannibals and Convicts. London. Cassel and Co. Thomas, Howard S.M. (1969), Learning Pidgin, Sydney, The Australian Broadcasting Commission. Thurnwald, Richard C. (1913), Ethno-psychologische Studien an Südseevölkern. Leipzig, Verlag Barth. Walsh, David S. (1978), Tok Pisin Syntax - The East Austronesian Factor. In: Papers in Pidgin and Creole Linguistics, I, Canberra, Pacific Linguistics, A-54, pp. 185-197. Whinnom, Keith (1971), Linquistic Hybridization and the "Special Case" of Pidgins and Creoles. In: Hymes, D. (ed.), Pidginization and Creolization of Languages, Cambridge University Press, pp. 91-116. Young, Hugh (1976), .«4 Directory of Solomons Pidgin Idioms, Mimeo, Honiara.

William Washabaugh

4. Pursuing Creole roots

1. INTRODUCTION

This investigation began, in my mind, with what seemed to be a rather simple problem of describing a curious construction in Caribbean English (henceforth CE), namely, the use of go after verbs.1 But it soon escalated from a problem of mere description to a reconsideration of some basic notions about Creole languages. The reason for the escalation is that a description of the go construction required some comparisons of varieties of Caribbean English, such as PIC (Providence Island Creole), JC (Jamaican Creole), GC (Guyanese Creole), Belizean Creole, Saramaccan, Sranan, Gullah, etc. In this comparison of varieties of CE, I discovered that the go construction was rather unevenly distributed. That spotty distribution of the go construction raised a question, namely, whether the distribution of the construction was produced by a fading out of a primordial Creole construction or by the first glimmerings of a change on this Creole language's horizon. This question about whether the curious #o construction indicated a change on the way in or on the way out led to another question about the motivation of the change. If the change was an ending, then where did the construction come from and why was it fading from the scene? This last issue of motivation is one the answer to which affects the very notion of a Creole language and of how creolists talk about and describe Creole languages. The next two sections of the paper will explore this issue of change, motivation and the concept of a Creole language. By this exploration I will lay out the theoretical significance of what I am about here. Then, in the body of the paper, I will proceed with the task of describing the go construction and of sifting the evidence to arrive at some conclusion about the process through which it appeared in CE.

2. CHANGE, MOTIVATION, AND THE CONCEPT OF A CREOLE LANGUAGE The structural similarity of Creoles has, over the years, attracted more ex-

86

William Washabaugh

planatory hypotheses than any other phenomenon in Creole language studies. There are external historical explanations and language internal explanations. Relexification, retention of substrate characteristics, universal processes of language acquisition, each has been proposed, at one time or another, as a plausible account of Creole structural parallels. These different accounts have provoked some of the fiercest debates in Creole studies, debates which show no signs of settlement, even in the most recent literature (Washabaugh, 1979). But perhaps creolists will continue to disagree until someone reconsiders the very question which is being asked, that is, until someone doubts the validity of Creole comparisons. My point is this: structural similarities have appeared when Creole languages of different regions and of different parentages have been compared. Comparisons have been made on the prevailing assumption that all Creoles are comparable, being young languages which have not been substantially altered by processes of language change. But if the assumption, that Creoles are comparable, can be shown to be false, then the whole business of comparing Creoles to reveal structural similarities is up for reconsideration. Specifically, if different Creoles can be shown to have undergone processes of language change at different rates, then only those Creoles that are at a similar stage of change will be comparable. And the results of such limited comparisons may be altogether different from the free-for-all comparisons on which most etiological hypotheses have been based.2 The studies of language change in Creoles are few and far between. Albert Valdman's (1977) recent work in this area is most sophisticated, but Mervyne Alleyne's work (1971) is not without significance. Valdman compares varieties of French Creoles in the Atlantic and Indian Ocean areas and finds that the Indian Ocean varieties are more conservative, closer in structure to a 'creole base.' Atlantic Ocean varieties,, especially Haitian Creole, are more innovative and further removed from the 'creole base.' The evidence upon which these conclusions are drawn is not diachronic evidence gleaned from analyses of creole texts of earlier centuries, but synchronic evidence of language variation. Valdman argues that 'if, following language variationists, one assumes that language change starts out as synchronic variation, evidence for the evolution of Creole must be sought primarily in linguistic variants observable today' (Valdman 1977: 174). Phrased differently, if synchronic variation in Creoles is directional, then that variation can be projected into a temporal dimension. According to this projection the starting point of the synchronic variation is similar to the diachronic source of the construction. Valdman is not the first to suggest that Creoles are related by diachronic processes, or that creole varieties constitute stages of a language undergoing abrupt change. But his method of analysis is refreshingly free of

Pursuing Creole roots

87

assumptions about Creole origins. I am thinking of the exciting proposals of M. Alleyne which first appeared in 1971. Alleyne, like Valdman, suggests that varieties of Atlantic English Creoles are related to one another diachronically. But his account of language change is drawn directly from a hypothesis about Creole origins. Alleyne expects that, at the earliest stages, Creole varieties are prepotently influenced by African substrate languages. At later stages of development, Creoles are increasingly influenced by the European superstrate. In other words, Alleyne deduces the diachronic process which relates Creole varieties from a preconceived conviction about the significance of 'substrate' languages on Creoles. Valdman's analysis is less biased just because he discovers changes in Creole languages without relying on the notion of 'substrate' language influence. Valdman analyzed varieties of French Creole. This study will analyze varieties of CE with the aim of demonstrating, like Valdman's, that some CE constructions are conservative and closer to a 'creole base,' and that other constructions are more innovative. The focal point of the analysis will be the use of go after verbs in CE. I will present evidence to show that CE go is part of a complex and developing system of connectives,3 the full complement of which is found in some but not all CE varieties. The major objective of this investigation will be to demonstrate the existence of a connective construction which is more conservative and closer to the 'creole base.' Only that conservative construction should be used in comparisons on which etiological hypotheses are built.

3. DISTINGUISHING INTERNALLY AND EXTERNALLY MOTIVATED CHANGES

However, before this focal analysis can be begun, certain problems of method must be broached. We will see that variations and changes in the CE connective go are largely language internal variations and are not due to language borrowing. They do not result from either superstrate or substrate influence. Yet, as Alleyne has shown, Creoles are inescapably influenced by superstrate languages in their diachronic development. So, we should suspect that borrowing occurs right alongside language internal changes in the development of Creoles. But if borrowing and internal change co-occur in Creoles, and if we will be identifying a purely internal change, then we will need a principle and method for distinguishing internal change from borrowing. Internal change and borrowing are distinguished by variationists in the following fashion: In phonology and syntax both, internal language change proceeds from the more marked condition of a rule to a less marked condition. In other words, the rule is altered from a less natural condition in

88

William Washabaugh

which its application is highly specified to a more natural condition in which its application is less highly specified. Borrowing is a kind of change which usually adds markedness to a language. Borrowing proceeds by adding specifications to the application of a rule. When applied to Creole languages, this distinction between internal change and borrowing runs into a knotty problem: both sorts of processes can lead a changing Creole construction toward a state which is similar to the reflex superstrate construction. How can borrowing and internal change be distinguished when, in certain cases, they both head in the same direction, i.e. the superstrate? Perhaps the only way to answer this question and to forestall any uneasiness about distinguishing internal change from borrowing is to present two short samples of variations and changes in CE, and to demonstrate, using these examples, how borrowing can be distinguished from internal language change. Borrowing leaves a construction more highly marked, i.e. less natural and more highly specified in application, and simultaneously moves that construction closer in form to the related construction in the superstrate. Variations in the use of the infinitive marker/? in CE will illustrate the nature of this borrowing process. In older varieties of CE, fi is used to mark both finite and non-finite verbs. In the sentences of (1), for example,// is used to mark finite verbs. The evidence for such a contention is presented in (Washabaugh 1975). (1)

(2)

a.

di mahn tro di wata fi di dag dringk The man poured the water for the dog (to) drink (Providence Island Creole). b. no konim ben poti tigri diaso fo mi lere hem taki The king placed Tiger here for me (to) teach him (to) talk (Sranan, Herskovits and Herskovits 1936: 152). c. je jasa kasaba fu baakuma fe"ni tjä-go They bake casava bread for gravemen find carry-go (Saramaccan, Clock 1972: 61). d. but uh tell hunnuh wuh bes' fuh we do But I will tell you all what is best for us (to) do (Gullah, Whaleyl925: 138). i wuda nais fi jan fi go It would be nice for John to go (Jamaican Creole, Bailey 1966: 124).

But in the idealized JC variety described by Bailey (1966), fi functions only as a marker of non-finite verbs as shown in (2). As a result of this specification that fi marks only non-finite verbs, all JC fi constructions parallel more closely the English for.. .to complement construction.

Pursuing Creole roots

89

In (Washabaugh 1978) I argued that this change by borrowing of the CE infinitive marker is a kind of change by decreolization. This change by decreolization might well have been overlooked if inter-language comparisons had not been made and if only intra-language analyses were used to identify borrowings - a common enough approach to decreolization. However, M. Alleyne, in lectures presented in 1971, demonstrated the power of inter-language comparisons to reveal change by decreolization. He showed that Sranan, Jamaican Creole and Gullah are each increasingly decreolized forms of a more thoroughly basilectal form similar to Sarammaccan. Evidence for this prior decreolization my be all but absent4 in the contemporary varieties of Gullah, Jamaican Creole and Sranan. So Alleyne concluded that change by borrowing in Creoles, i.e. decreolization, cannot be completely identified by analysis of just internal variation in any Creole variety. Analysis of variation within a variety must be supplemented by analysis of variation between varieties for a complete picture of decreolization. Having considered externally motivated changes, we can now turn to internally motivated change. language internal changes in any language usually render a construction less highly marked and more general in its application, though in a creole language such changes may accidentally leave constructions more similar to superstrate constructions. A CE interrogative tag, together with its forms and uses, will illustrate this process of internal change. First, Roberts (1977) pointed out that duont is one of a number of JC tags which transform a declarative sentence into a yes/no question. It differs from other tags in that it can apear both sentence-initially and sentence-finally. PIC ent is similar to duont; it can appear both sentence-finally as shown in (3) and sentence-initially (4). The phonological distinctiveness of the two forms is evidence that the ent was not recently diffused from JC to PIC. Sentence-initial tags like ent and duont have not been reported for other varieties of CE. (3) (4)

im wuda bor rru, ent? He would have shot me, right? ent i mada ga biebi agin an im kom di siem? The mother had another baby and it looked the same, right?

The interesting feature of these sentence-initial tags is their rarity. Sentence-initial yes/no tags appear in only a handful of languages, including Fante, Yoruba, and Mandinka (Moravcsik 1971: 136; Ultan 1969: 50; Holm 1978). On this evidence, sentence-initial tags can be said to be highly marked, i.e. less natural. The rare and isolated apearance of sentence-initial tags, together with the widespread attestation of less marked sentence-final tags in CE, all serve to support the hypothesis that the sentence-initial

90

William Washabaugh

tag construction is more conservative and closer to the 'creole-base' while the sentence-final tags result from innovative changes which reduce the markedness of the conservative construction. It is indeed possible that the sentence-initial tags of CE indicate influence from the Creole language substrate. But before such a hypothesis can be accepted, it should be verified by the discovery of such sentenceinitial tags in other Creoles influenced by West African languages. Since such a verification is not available, it seems reasonable to argue that the highly marked sentence-initial tag construction arose during creolization and has been subsequently on the decline due to the impact of internal change. If sentence-initial tags like ent and duont are relatively rare in CE varieties today, and if sentence-final tags are more frequent, it is primarily because of an internal change process. The fact that English makes exclusive use of sentence-final unmarked tags is incidental to this analysis.5 The internal change process in CE can be predicted independently of the existence of the English superstrate. It is important to note that this can not be said of changes in the CE complementizer system. The existence in English of the for. .. to complementizer construction is not unrelated to the change in CE which restricts fi to marking only non-finite verbs. That change cannot be predicted without taking account of the abiding influence of the English superstrate on the Creole. These examples of Creole borrowing and internal change will illustrate one other principle about internal change developed by variationists. According to Bailey's wave model of internal change (1973), the more conservative realizations of a construction are likely to appear in areas peripheral to the source of the change, while the innovative realizations of a construction are likely to appear close to the source of the change. So, it is not accidental that we find sentence-initial tags retained in varieties like JC and PIC which are peripheral to the source of change. (Presumably the source of the change should be located on the West African coast or at pivotal historical centers like Barbados or St. Kitts-Nevis.) The converse principle, that innovative realizations appear near the source of the change, is not borne out clearly by these examples. The reason for this absence of innovative forms at the source of internal changes is straightforward enough. The communities near the sources of change are just those communities which have been most intensely affected by either decreolization or other historical social upheavals. Barbados has a basilect which is not significantly distinct from its acrolect. Krio and West African Pidgin English (WAPE) on the West Coast of Africa both have had tortuous histories in which the results of internal changes have certainly been complicated by borrowings of different sorts. Summarizing the results of this digression into methods of identifying

Pursuing Creole roots

91

Creole language change, we can distinguish internal changes from borrowing by identifying, for an internal change, an internal motivation for the synchronic variations on which a change is hypothesized. On the other hand, a change by borrowing cannot be motivated independently of the target to which the change is directed. Where a language is changing internally , the conservative realizations are likely to appear in varieties furthest removed from the source of the change. Thus, the most conservative and highly marked realizations of any construction will appear in 'maroon' Creole communities like those in inland Surinam, and in outlying and isolated communities like Providence Island. However, the effects of decreolization can erase both these conservative realizations, especially in communities like Providence Island (Washabaugh 1977), and also the innovative realizations which one would expect to appear near the sources of the change. 4. ON THE WORD

This digression into methods of identifying internal language change in Creoles has been necessary and worth the pages it has consumed. We can now move to the focal issue of this investigation, which is internal change in the CE go construction. Turner (1949: 210) was, perhaps, the first to point out a peculiar use of go in CE. Later it was commented on by LePage and Cassidy (1967: 199), analyzed by Cave (1976: 12), and discussed by Bickerton (1977a). Regarding a sentence Bee (5) in Gullah, Turner says that the second go is a verb, meaning 'in order to' which connects two other verbs. Regarding sentence (6) in JC, LePage and Cassidy say that the second go is a sort of auxiliary with conjunctive force. They note that come in (6a) performs a function similar to go in (2). With respect to sentence (7) in Guyanese Creole (GC), Cave claims that the third go is an infinitive marker and is distinct from both the directional verb go and the future marker go which precedes it in the sentence. Bickerton (1977: 177), agreeing with Cave, supplies evidence for a similar use of go in Hawaiian Creole (8). (5) (6)

(6a)

yu beta go hom go si bauSa SÜan (Turner 1949: 266). You better go home see about your children. an' what yu go go buy, Miss Princess (LePage and Cassidy 1967: 116). And what did you go buy, Miss Princess? Dis naga man come come collar me de same like a say me da him sexis (LePage and Cassidy 1967: 116). This black man comes and collars me just as if I were the same sex as he is.

92 (7) (8)

(9)

William Washabaugh wi go go go fine am (Cave 1976: 12). We are going to go and find it. so a tel am wai no go mek am go aut go it raet (Bickerton 1977: 176). So I told her, 'Why don't you make it go out and eat rats?' ai tel i go get sigaret (Bickerton 1977: 24). I told him to go and get cigarettes.

The question to be asked about go in sentences (5-8) which will shed some light on changes in Creole languages is this: If go marks non-finite verbs in (7), how aid go acquire its function as marker of non-finite verbs, and how are the uses of go in (5, 6, 8,9) related to the infinitive marker go. Bickerton is the only one of the above who has tried to answer these questions. He suggests that the go infinitive marker drived from go as an imperative marker (1977: 24, 177, 293). Unfortunately, this suggestion is accompanied by no evidence or argumentation other than the observation of a certain symmetry of the use of go in embedded imperative constructions (9) and the use of go as a marker of non-finite verbs.

5. Go IN PIC Let me begin my account of the nature and derivation of go by providing more elaborate data about the uses of go in CE. My data are drawn primarily from the candid speech (PIC) of the people of Providence Island, Colombia, recorded intermittently over a period of six years (see Appendix). First, the connective appears in a number of different contexts as shown in (10). The most striking of these go sentences is (lOa) (10)

(11)

(12)

a.

firs ting im go go pap af i haan bogota The first thing (you know) he will go break his arm in Bogata. b. im go chak a dik hil go keri it She goes all the way to Dick Hill to carry it. c. shi wash aut di swet an pres it an keri it go gi im She washes out the sweat and presses it and carries it to give it to him. wi yustu go ina piipl plies go go tief dem kien We used to go into people's places, go to steal their (sugar) cane. go bring i Go bring it.

Pursuing Creole roots (13)

93

ai kuda tel wan a dem bwai go ketch it haas I could have told one of those boys to catch the horse.

wherein the go connective immediately follows a go verb. It is this immediate juxtaposition go and go which inclined Cave to claim that the second go must be a functionally different formative from the verb. Go in sentences (lOb and c) is increasingly more difficult to analyze. In (lOb) the go connective is separated from the go verb, and as a result, the possibility arises that the second go may be a repetition of the go verb. Sentence (11) shows that such a possibility is, on occasion, realized. In (lOc) the difficulty of distinguishing the verbal function of the second go from its connective function is intensified. Here the verb preceding the go complementizer is directional but it is not the go verb. Again the possibility lingers that the second go in such sentences is a conjoined or serialized verb. (Sentences (12, 13) show that both the polite imperative and the embedded polite imperative, mentioned by Bickerton (1977: 24) exist in PIC.) A second set of facts about PIC is equally important for the light it sheds on the structure and function of the go connective. Sentences (14, 15) show that kom and gan can function in much the same way as go. However, for a variety of reasons, some of which will be discussed below, the kom connective apears less frequently than go, and the^a« connective is used less frequently than kom. (14)

(15)

a.

dem kom ko saach mi hier. They came to search my hak. b. di hosban kom in ko luk biebi The husband came to to look for the baby. c. an neks de im bring sponfkiek kom gi wi And the next day he brought sponge cake to give us. a. den ah gan ga tiif aut presh pinuts bota Then I went and stole P's peanut butter. b. me tu taim ah me gan ova de gan keri im leta It was two times that I went over there to bring him a letter. c. siem taim ai tel dem, dey ron lang de gan tel R. The same time I told them, they ran along there to tell R.

A third set of facts about kom and gan will nail down the claim that kom and gan and, by implication, go, function as grammatical formatives, rather than as verbs, in PIC. The fact is that the vowel denasalization rule, presented in (Washabaugh 1978b), applies selectively to the second kom and the second gan of sentences (14, 15). In (1978), I demonstrated

94

William Washabaugh

the existence of a denasalization rule which applies variably in PIC. I noted that one of the more favorable environments for vowel denasalization is unstressed grammatical morphemes like the past tense marker men [me]. In certain phonetic contexts, the [me] undergoes denasalization and is realized as [me]. Just this same denasalization rule applies selectively to the second kom [ko] and the second gan [gä] in (14) and (15) to produce [ko] and [ga].6 But the denasalization rule does not apply to the first kom and gan of these sentences. So it must be that only the second kom and gan in (14, 15) are unstressed and available to undergo denasalization. A plausible explanation for the lack of stress on these formatives is that they serve a grammatical rather than the semantic function which is served by the stressed directional verbs. So, from three sets of facts cited above, it can be concluded that: (1) Go functions as an infinitive marker in some sentences, like (lOa), in PIC; (2) While go functions as an infinitive marker in some sentences, its use in other sentences (lOc) seems to be related to a serial verb construction: (3) The previous two statements about go apply mutatis mutandis to kom and gan. One more set of facts from PIC will establish directionality in the vatiation between go, kom and gan, and will set the stage for a comparison of CE varieties. First, it is a fact that the go connective is most frequently used after the verb go, and similarly kom and gan connectives are most frequently used after the verbs kom and gan respectively. But there are systematic deviations from this tendency. These systematic 7 deviations will reveal a directionality in the variable uses of the go, kom, and gan connectives. Sentences (16-19) demonstrate that the verbs kom and gan can be followed by the complementizer go. But the present tensed verb go, or verbs with similar directionality and temporality, cannot be followed by kom and gan. Phrased differently, the go complementizer can follow a main verb which is either go, kom, or gan. But the kom complementizer must follow a main verb with the directional characteristics of kom, and gan must follow a main verb with the temporal and directional characteristics of gan. The go connective, for its part, is not so restricted. While there are occasions, e.g. sentence (lOc) where the directionality of go is (16)

(17)

a.

ai mos kom aut go luk waif I must come out to look a wijfe. b. yu gwain kom ina mai haus go kos mai sista . . . (a hypothetical sentence, which, when submitted to a speaker's judgment, was reckoned acceptable). You're going to come into my house and cuss my sister . . . a. so ai gan batamhaus go dans wid im

Pursuing Creole roots

(18) (19)

95

So I went to Bottomhouse to dance with her. b. mi gan luk wan koknut go mek kokada (a hypothetical sentence judged acceptable). I went and looked for a coconut to make cocada. *dem se PP wudn go kom help kot kien (a hypothetical sentence). They said PP wouldn't go to help cut cane. *ai wi go batamhaus gan dans wid him (a hypothetical sentence). I will go to Bottomhouse to dance with her.

consistent with and contributes to the meaning of the main verb, still at other times, go can appear after verbs, e.g. (16, 17) which contradict its temporality and directionality. We can talk, then, of a variable retention of verbal qualities by the go, kom, and gan connectives, noting that as we move from the sentences (lOc, 14c, 15c) to (lOa, 14a, 15a) go kom, and gan all tend to lose something of their directionality and temporality. Each connective is more clearly bleached of its verbal directionality and temporality in the (a) sentences than in the (c) sentences. But the different connectives exhibit different capabilities of shedding those verbal qualities, and only go, not kom or gan, is capable of discarding completely its directionality and temporality. These facts point to the origin of the go, kom, and gan connectives in the sort of serial verb constructions from which the sentences of (c) most certainly derive. It is interesting, but not crucial to this analysis, to note that Kpelle has a consecutive verb construction which produces sentences like (20) and that Fante has a kind of serial verb construction such that 'in sentences referring to motion, verbs after the first may also be marked by derivations of [ka] 'go' or [ba] 'come' as in (21) (Welmers 1973: 373). (20) (21)

a paai Hi ijc kpalarj na etii ke He's going to go to his farm and work (Ibid., p. 364). D-ba-a ha by-to-o ndiemba He came here come bought things. He came here to do some shopping (Ibid., p. 375).

6. COMPARATIVE EVIDENCE ON Go

A look at the other varieties of CE confirms the hypothesis that the go, kom, and gan connectives are derived from a serial verb construction. Moreover, a comparison of PIC to other varieties of CE will show that the reanalysis of directional verbs into grammatical connectives involves a gradual and variable bleaching of verbal characteristics. This bleaching,

96

William Washabaugh

which accompanies reanalysis, is evidence of ongoing 'grammaticalization'; that is, the bleaching indicates a process which reduces the semantic value of formatives which are made to serve syntactic functions. First, though the data are limited, Saramaccan predictably makes use of kom and go verbs to extend and specify the meanings of main verbs of motion as shown in (22) (Clock 1972: 49). Also Saramaccan makes use of the go formative to connect a consequent verb to the main verb as shown in (23) (Ibid., p. 55). But there is no evidence that the go in Saramaccan can be immediately juxtaposed with the directional verb, nor is there evidence that the go in Saramaccan can apear with a motion verb with a different directional quality. These lacunae suggest that^o, and (22)

(23)

a.

dimujeekuleko The woman run come b. dimujeekulego The woman run go I a musu go a wosu go tei di soni You neg must go to the house go take the thing

a fortiori kom and gan, have not undergone reanalysis and 'grammaticalization' in Saramaccan. On that count, Saramaccan will be considered a conservative variety of CE, closer in structure to the 'creole base.' Other varieties of CE not only exhibit the go and kom connectives, but these connectives can be immediately juxtaposed with the main motion verb, and occasionally they can differ in directionality from the verb. The sentences of (24) together with (5-8) illustrate this innovative (24)

a.

ai e goin go pik I am not going to pick (Turner 1949: 260). b. Policeman a come fe (sic) go mek a row (LePage and Cassidy 1967: 199).

use of go and kom. These uses are labelled innovative because they are evidence that the use of a verb as a connective is not only frequent and regular, but also that that connective is undergoing 'grammaticalization' in which it is being bleached of its verbal characteristics. PIC seems to be more innovative than most other varieties in this respect. Two gaps in the evidence are worthy of some mention. First, the gan connective is conspicuously absent in some varieties of CE, e.g. Belizean Creole (G. Escure, personal communication), Jamaican Creole (LePage and Cassidy 1967). The absence of gan in these varieties may be due to its excision from the creole under pressure from the acrolect. Second, the absence of the go, kom, and gan connectives in Schneider's (1966)

97

Pursuing Creole roots

texts of WAPE, even after directional verbs (25), is curious. This absence may lend some support to Bickerton's (1977: 61) claim that WAPE is a repidginized form of Sierra Leone Krio. (25)

i don go we fi nak fut-bot He went to play football (WAPE, Schneider 1966: 185).

Two diachronic rules will formalize the complex process of language change evidenced by these facts. A first rule (see Figure 1) has it that directional serial verbs, used after verbs of motion, will function as connectives to mark relationships between consequent verbs and main verbs with directionality and temporality. A second rule will formalize the fact that the directional connectives undergo reanalysis and 'grammaticalization.' As indicated, go undergoes this 'grammaticalization' earlier and faster than kom or gan. This second rule will distinguish conservative varieties, like Saramaccan, from other more innovative varieties, and will make explicit the pattern of connective 'grammaticalization' in those innovative varieties. Figure 1. Diachronic Rules for Creole Connectives Rulel: Verb

Rule 2: Connective -*·

connective α directionality β temporality infinitive marker

verb α directionality j3 temporality /

["verb Ί j_directionalj go1 kom2 gan3

Two problems remain for this analysis of internal language change in CE. First, PIC seems to be the most innovative variety in which grammaticalization of the connective is carried further than in other varieties of CE. Yet PIC is clearly peripheral to the sources of the change and is therefore a variety which, according to Bailey's wave model, should be conservtive rather than innovative. This problem can only be solved by recognizing that internal change in CE is proceeding hand in hand with decreolization. Borrowing and decreolization have applied intensely to the central varieties resulting in the erosion of the environments for the application of internal change and the stalling of the process of change. Failing to recognize this erosion by decreolization will lead one to the specious conclusion that PIC is more innovative than centrally located varieties. The second major problem in this account is that the reanalysis of a

98

William Washabaugh

serial verb into a connective and infinitive marker is apparently unexpected and even peculiar. In his study of serial verb reanalysis in both Niger-Congo languages and in Krio, Givon (1973; 1975) found a consistent process in which serial verbs are reanalyzed as prepositions. Thus gi, tek, and go are reanalyzed as the prepositions 'for' (benefactive), 'with' (instrumental) and to' (locative) in Krio. There seems to be some discrepancy between the hypothesis that the go serial verb is reanalyzed in CE as a complementizer-like connective and Givon's hypothesis that the go serial verb in Krio is reanalyzed as a preposition. This apparent discrepancy does not weaken the claim that the go connective in CE developed from a serial verb construction, but only points up a complexity in the developmental process. That is, prepositions and infinitive markers perform similar connective functions, and it is reasonable to expect that they have evolved from a similar source (Washabaugh 1975). That similar source in the 'creole base' is most probably a serial verb construction. The CE connective go probably derived from constructions like (26) in its function was ambiguous, being either a serial verb or a preposition. From such a 'creole base' construction the go gradually acquired (26)

i de waka go na ton He is walking to town (Givon 1973: 17).

the functions of a preposition and then of a sentential connective. As the go was conscripted into service as a sentential connective (in complementary distribution with fi), na-da-a was made to serve as the semantically general prepositional connective. In Krio this developmental process was truncated. The go was reanalyzed to serve a preposition, but it was never further reanalyzed to serve as a sentential connective. The difference between the two developmental processes is a matter of a road not taken.

7. CONCLUSIONS

Now let us pause and take stock of this analysys and its significance. The hypothesis is that the go infinitive marker develops gradually in CE from a serial verb construction. That developmental process involves first the use of the serial verb as a complementizer-like connective, and later the 'bleaching' of verbal characteristics from that connective. Of the three connectives discussed, go, kom, and gan, the evidence shows that go is the first serial verb to be grammaticalized, the fastest to undergo 'bleaching.' So it is the most widespread of the three directional connectives.

Pursuing Creole roots

99

The verbs kom and gan are grammaticalized later and slower. Moreover, the kom and gan connectives appear less frequently in PIC than go and with vanishing frequency in other CE varieties. The disappearance of especially the gan connective in CE may be due to the eroding affects of decreolization. The significance of this analysis is twofold. First, it roughly corroborates Alleyne's hypotheses about diachronic relations between CE varieties, and second it casts some doubt on the hypothesis that the structures of the 'creole-base' are products of "handicapped first language acquisition,' i.e. Bickerton's (1977a) universal creolization hypothesis. This analysis corroborates Alleyne's hypothesis. Alleyne starts out with the expectation that African substrate characteristics will be most evident in conservative varieties of CE, and increasingly less evident in innovative varieties due to the effects of decreolization. But the problems here are that Alleyne's principle for the comparison of CE varieties is drawn straightforwardly from Creole studies itself, and that Alleyne fails to distinguish internal change from decreolization. The analysis presented here is founded on methods and principles developed outside of creole studies. So, when I conclude that the most conservative varieties of CE are also similar in structure to African substrate languages, that conclusion is a happy result of the analysis, rather than a principle from which the analysis begins. Second, by distinguishing change from decreolization, this analysis shows that CE is traveling its own road of development besides also traveling a road dictated by superstrate languages. By conflating the two roads, Alleyne detracts from the independence and autonomy of the creole language and, by implication, of the Creole culture. A second result of the analysis is that it weakens the universal creolization hypothesis. The similarity of the connectives in the CE 'creolebase' to serial verbs in West African languages is hardly accidental. Nor is it likely that such a 'creole-base,' indicated in this analysis, is dictated by an innate and universal 'blueprint' for language acquisition. Only a specific historical explanation seems to account for the structure of connectives in the CE 'creole-base.' True, the consideration of just one feature of the 'creole-base' is not sufficient to cast aside the whole hypothesis which has been formulated on a consideration of multiple features of Creoles. But this one specific analysis should temper the enthusiasm of those who expect to find universal processes at every turn in creole language development.

100

William Washabaugh

APPENDIX

Hansel and Gretel (O.H., 12 years, female, recorded 1972). wan taim a man had a waif and hi ded an i marid tu a neks wan an i had tu pikninL an e niem hansel an gretel. wan taim di uman and di man se de gwain out in di wudz. an dey gwain out in di wudz - and di uman dem e neva laik di tu piknini — an di man drop bred skrumz fi de fala di trak wen de komin. dem gan huom lef dem. an de drop bred skrumz. nau di bord dem pik dem op an de neva no we fi go. an de gan raun a uol liedi haus wik - di haus nutn els bot kiek; an nutn els bot kiek, priti. an de, de Staat go pik af di kiek, pik afdi kiek, nau wid dot den di uol liedi kom out; im äks dem if dem waan stan wid im. dem tel im yeh. an nau im put di likl bwai ina i kief an im tek di gyal nau fi kuk and work nau. dot taim nau de gi di likl bwai a rat tiel and him put im ina i kief se wen im get fat im gwain kil im. dot de nau bwai, di uol liedi — evri die im go - di likl bwai shub out di - evri die i sen di likl gyal go, go go sii if di bwai fat nuf. evri die im go di likl bwai shub out di, di rat tiel. di nada die agin im lef di gval agin, im shub out di rat tiel agin, dot die im gan im shub out di finga. di uol liedi se 'fat nuf den.' aarait den, im put di gyal fi hat di wata nau se dot de gwain kil di bwai. im put di gyal fi hat di wata. wen im gan, wen im gan, gan gan piip dong ina di pat, im tel, im tel di gyal go go sii ifdi wata hat nuf yet. di gyal tel im, him no no wen taim di wata hat nuf, bot him fi kom kom luk. an iz him kuda kom fi piip dong in de di gyal jos push i hed rait dong ina di hat wata an slam in di stuov pa im. nau di likl gyal tek out di bwai kwik an de tek nuf moni we de kuda liv and de gan. an de Staat tu travl. wen de get, de sii wan riva, and wen de luk de sii di man likl dag komin rait raun and im jos kieri dem rait huom. wen dey gan di man de rait huom de and di uman we im tek ded and jos him wan de huom. so wen im get di man did de muon an iz de get di man stop muon bekaz em de de. di dag gan rait de an luk fi dem.

NOTES 1. A version of this paper was presented at the NWAVE Conference, Washington, D.C., 1978. I appreciate, first and foremost, the warmth and patient assistance of the people of Providence Island. In addition, I thank D. Bickerton, F. Eckman, and P. Muysken for their comments and criticisms. 2. Subsequent to the first presentation of this argument at NWAVE 1978, Derek Bickerton presented a similar argument at the Conference on Theoretical Orientations in Creole Studies, 1979. 3. 'Connective' will be used throughout to refer to a kind of infinitive marker-in-

Pursuing Creole roots

101

development. The term 'connective' is intended to be just as weasily and ambiguous as the functions of the formatives which 'connective' designates. 4. Alleyne, in his lectures, has cited a handful of residues of a less decreolized JC basilect, e.g. sinek for 'snake' from the more basilectal sineki, and her for 'carry', which is hypercorrected on analogy with sinek. 5. Some might wish to argue that the sentence-initial tag construction in CE is modeled on superficially similar constructions in English, e.g. Ain't she sweet. However, even aside from the fact that ain't in such a sentence does not function as a tag, I can see no mechanism through which such a model could be carried so selectively to varieties ofCE. 6. A most striking bit of evidence for this denasalization process came to my attention as I read narrative transcripts written for me by O.H., now seventeen years of age. She regularly deleted the orthographic V from words in which I independently detected denasalization to have occurred. So, for example, her written transcription of sentence (15a) is 'a drink of water den a gan ga thief out presh penuts butta.' 7. Besides these systematic deviations, there are unsystematic deviations from the tendency for the go, kom, and gan connectives to follow go, kom, and gan verbs. First, go and kom are both polysemic verbs, and where a particular sentence calls for a nondirectional reading of the verbs go or kom, speakers do not employ the go, or kom connective. So in so gi mi nau hau yu an him kom fi fait, kom is followed by fi rather than by kom. Second, occasionally the go and kom connectives are found in conjunction with a fi infinitive marker as in mieri neva kom fi go bring fish (a hypothetical sentence). That is, in PIC a non-finite verb may be redundantly marked by two connectives, i.e.β + kom/ go. This conclusion is corroborated by Bickerton (1977: 177) who argues that such redundancy is characteristic of a language in decreolization.

BIBLIOGRAPHY Alleyne, Mervyne (1971), Acculturation and the Cultural Matrix of Creolization. In: Hymes, Dell (ed.), Pidginization and Creolization of Languages, New York, Cambridge University Press. Bailey, Bl. (1966), Jamaican Creole Syntax, London, Cambridge University Press. Bailey, C.J. (1973), Variation and Linguistic Theory, Arlington, Virginia, Center for Applied Linguistics. Bickerton, Derek (1976),Change and Variation in Hawaiian English, Vol. I. University of Hawaii, mimeo. Bickerton, Derek (1977), Change and Variation in Hawaiian English, Vol. II. University of Hawaii, mimeo. Bickerton, Derek (1977a), Pidginization and Creolization: Language Acquisition and Language Universals. In: Valdman, Albert (ed.), Pidgin and Creole Linguistics, Bloomington, Indiana University Press. Cave, George (1976), Sociolinguistics and Child Play in Guyana, University of Georgetown, mimeo. Givon, Talmy (1973), Prolegomena to any sane Creology, University of California at Los Angeles, mimeo. Givon, Talmy (1975), Serial Verbs and Syntactic Change: Niger-Congo. In: Li, Charles N. (ed.), Word Order and Word Order Change, Austin, University of Texas Press. Clock, Naomi (1972), Clause and Sentence in Saramaccan, Journal of African Linguistics 11,45-61.

102

William Washabaugh

Herskovits, Melville and Herskovits, F. (1936), Surinam Folklore, New York, AMS Press. Holm, John (1978), The Creole English of Nicaragua's Miskito Coast, Unpublished dissertation, London, University College. LePage, Robert and Cassidy, F. (1967), Dictionary of Jamaican English, New York, Cambridge University Press. Moravcsik, E. (1971), Some cross-linguistic generalizations about yes-no questions and their answers, Working Papers in Language Universals 7,45-193. Roberts, P. (1977), Duont: a case for spontaneous development, Journal of Creole Studies 1(1), 101-108. Schneider, Gilbert (1966), West African Pidgin English, Athens, Ohio University Press. Turner, Lorenzo (1949), Africanism in the Gullah Dialect, Ann Arbor, University of Michigan Press. Ultan, R. (1969), Some general characteristics of interrogative systems, Working Papers in Language Universals 1,41-63. Valdman, Albert (1977), Creolization: Elaboration in the Development of Creole French Dialects. In: Valman, Albert (ed.), Pidgin and Creole Linguistics, Eloomington, Indiana University Press. Washabaugh, William (1975), On the Development of Complementizers in Creolization, Working Papers in Language Universals 17,109-140. Washabaugh, William (1977), Constraining variation in decreolization, Language 53 (2), 329-352. Washabaugh, William (1978), Complexities in Creole Continua, Lingua 46, 245-261. Washabaugh, William (1979), Review of Pidgin and Creole Linguistics, Albert Valdman (ed.), Journal of Linguistics 15(1), 154-157. Welmers, William (1973), African Language Structures, Berkeley, University of California. Whaley, Marcellus (1925), The Old Types Pass, Boston, Christopher Press.

Chns Come

5. A re-evaluation of the predicate in Ile-de-France Creole

1. For the purposes of this paper,1 the Indian Ocean Creole French dialects of Mauritius (M), Rodrigues (Ro), and Seychelles (S) may be considered as a single language; despite relatively minor lexical, phonological, and syntactic differences, mutual intelligibility is high. We shall label this language Ile-de-France Creole (IF), since it may be postulated that all three modern dialects derive from a variety of Creole French which evolved in Mauritius (once known as Ile-de-France) in the Eighteenth century.2 2. Recent descriptions of IF3 have in common, explicitly or implicitly, a set of assumptions concerning the structure of the predicate, and in particular, an underlying Copula is either implied or explicitly postulated. We may represent these assumptions along the following lines: VP VG

-*

Aux + VG

->

NP Copula + AP PP V+ (NP) + (PP)

(Adverbs are included as PPs)

The constituent Aux(iliary) corresponds approximately to the tense/ aspect markers. These are, with approximate golsses: φ 'present', ft' 'past', ova, va, a 'future', pu 'future (definite)',//«, in, η 'completive\fek 'immediate completion, to have just', ape, pe (and an archaic/obsolescent formapre) 'progressive'. These PS rules allow a description, of sorts, of the IF data. Four classes of verbs follow from these rules:4 I—

[+ transitive, + attributive] Moris i n don so kuto Teof 'Maurice has given/gave his knife to Thdoph' II — [- transitive, + attributive] zot ova al lekol 'they will go to school'

104 III — IV— There VVI VII -

Chris Come [+ transitive, - attributive] ι η suiy latab 'he has wiped the table' [-transitive,-attributive] i ape rofle 'he is snoring' are also three kinds of equational (copulative) sentences: Copula + NP mo papa ti sarp tie 'my father was a carpenter' Copula + AP lerua i bet 'the king is stupid' banan i n mir the bananas are/have become ripe' Copula + PP mo tilekol Ί was at school'5

2.1. As can be seen, this formulation postulates a deletion of Copula, since the surface form of the copula in V-VII is zero. In certain contexts (Interrogatives, Comparatives, Locative Relatives with (a)kot 'where'), a form ete occurs. Ete is optional in the present.6 Examples: kot u liv (i ete)? 'where is your book?' ki Z tiete? Vhat was John?, what did John used to be?' ki nasio u (ete)? 'what nationality are you?' ki u (ete)? *who are you?' i pli kuy ki u (ete) 'he is more stupid than you (are)' mo pa kone lekel (i ete) Ί don't know which one it is' i n al kot lerua ti ete 'he went where the king was' Such sentences do indeed suggest that IF has an underlying copula ete which has been 'exposed' clause-finally as the result of the application of movement rules (or, in the case of comparatives, by co-referential deletion rules). Thus, the reasoning goes, its non-appearance elsewhere is due to relatively low-level deletion rules (cf. Valdman 1973, 1978: 231-241, and the recent descriptions of IF). Note however that the recoverability criterion for deletion rules is not met in the sentences V-VII above. There is another surface form se, which corresponds roughly to French c'est 'it is': we return to this in section 3.22 below. The final point that needs to be made here is that copulative sentences are by definition semantically statives, a fact which leads to some sleightof-hand in the explanation of various co-occurrence phenomena, as we shall see. 2.2. These PS rules have other drawbacks. They do not account for the close semantic and syntactic relation between adjectives and verbs, and they give a rather distorted picture of the IF predicate. Since they place

A re-evaluation of the predicate in Ile-de-France Creole

105

verbs and adjectives in different categories, they do not allow a unified account of the semantic and distributional facts. For the same reason, they are directly reponsible for an erroneous analysis not only of the passive in IF but also of the verbal classes (Corne 1977a, 1977b: 70-71,153158). In short, the basic assumptions represented by these PS rules are wrong. 2.3. In this section, some facts for which a coherent analysis is required are briefly presented and the inadequacy of the postulated PS rules to deal with them is shown. 2.3.1. The aspect marker (a)pe usually confers on the predicate a clear progressive (or continuative) sense: zot ti ape s te 'they were singing' mo pe oblize fer sa Ί am being obliged to do it' zot ova pe peye par zot burzua 'they will be being paid by their boss' Compare however the following: ι pe malad 'he is getting sick' mo pe kota li Ί am beginning to love her' mo pe k ta li (M) Tm crazy about her (at this very moment)' //' pe fatige tap to laport (M) 'he is getting tired of knocking at your door' mo pe lafe Ί am getting hungry' mo pe fe (Μ) Ί am starving' As these examples show, (a)pe may also produce what appears to be an inchoative sense, or the predicate may emphasize the 'now-ness' of the event. With certain predicates, (a)pe is excluded: */ pe anda 'he + (a)pe + inside' *i pe bet 'he + (a)pe + stupid' *i pe sarp tie 'he + (a)pe + carpenter' Within the postulated framework, these facts can be handled only in an ad hoc manner. There is no problem with ape + Verb, since ape has always a clearly progressive sense. It still has this sense with oblize '(to be) obliged', for example, but not -with fatige '(to be) tired' where it has what can be reasonably faithfully glossed as inchoative '. .. is getting . ..' in English; see the sentences above. Now within the framework of the postulated PS

106

Chris Corne

rules, one would surely wish to have oblize and fatige in the same, nonVerb, category (non-Verb because the final vowel truncation rule has not applied, v. 2.33 below). One is thus led to postulate an ad hoc semantic feature [± action] to account for the different senses of ape (Corne 1977b: 65-66). The problems do not end there, however, since the non-occurrence of ape also needs explanation. Another feature [± durative] can be postulated; [+ durative] does not allow ape, [- durative] does. Thus, in the examples above, bet 'stupid* is [+ durative], malad 'sick' is [- durative]. This feature is slightly less arbitrary than [± action] or whatever else one might postulate to explain the different senses that ape confers on predicates, since it can be used to explain the distribution of fin as well. 2.3.2. The aspect marker fin (variants in, n) always indicates completion, with respect to present time: mo n maze Ί have eaten'

(cf. mo maze Ί eat')

or to past time: mo ti n maze Ί had eaten'

(cf. mo ti maze Ί ate')

or to future, future-in-the-past: ler u ava n repar u loto *when you (will) have repaired your car' motion bez li Ί would have hit him' Note that in all these examples, the predicate contains a verb. In order to account for the distribution of fin, predicates containing Copula (i.e. which are statives, as already noted) must be assigned to one of two categories: those which are 'generally' true, [+ durative], and those which have come to be so, [- durative]. Durative statives do not require the presence of the completive marker: tu le zur, maze i pare ver midi 'every day, the food is ready towards midday' normalma, loto i kuver asuar 'usually, the car is covered at night' Non-durative statives (i.e. those which have come to be true) do require fin: nu maze i n pare komela Our food is ready now' loto i n kuver, koma u n dir mua the car is covered, as you told

me'

A re- evaluation of the predicate in Ile-de- France Creole

107

Copula + NP is usually [+ durative], but [- durative] also occurs: / n ler pur nu ale 'it is time for us to go' Copula + PP is usually [+ durative], although [- durative] also occurs: s , ui, i n lor la! Ves> THAT is perfect!, Right on!' (literally, that yes it completive on there) Copula + AP is [± durative]; some adjectives (e.g. bet 'stupid') are usually [+ durative], others such as su 'drunk' are usually [- durative], while yet others (eg. pare 'ready') may be either. 23.3. Final vowel truncation. IF has a phonological rule of final vowel truncation; we may represent this as:

# ]Verb+X Where X is specified in syntactic terms 7 and where the verb belongs to a large sub-class of verbs having, at some point in their derivation, the phonological shape (X)VCe.8 The fundamental postulate here is that this rule applies only to verbs. Some examples: zot pa kon koz fr se they don't know how to speak French' ι ga$ li έ sok lie gets a shock' mo pu vin ris e zur 1 shall be (come) rich one day* sa kopozer i apel Amed that composer's name is Ahmed' έ Ion i kut de rupi One ell costs 2 rupees' so valiz pez ve liv (M) ' his suitcase weighs 20 pounds' labutik iferm siz-er the shop closes at 6 o'clock' However, there are cases where this rule can be seen to have applied to items whose sense seems clearly adjectival. Examples: eski nu va met zot tu melaz asam? 'are we going to put them all mixed up together?' (li) ti aloz lor kanape (M) 'he was stretched out on a sofa' (Baissac 1888: 371) Further examples will be seen below.9 2.3.4. Passives. Within the framework of the PS rules given above, 'pas-

108

Chris Come

sive'-type sentences in IF can be analysed as deriving from a rewrite of AP (cf.Corne 1977a, 1977b: 149): AP -» (Adverb degree ) + AG + (PP) AG -> Adjectif+CPP) The PP constituent of AG is the traditional 'complement of the adjective': zot eterese ek lopes 'they are interested in fishing' zot pros pur (zot) ale they are on the point of leaving' It may be postulated that this complement depends on a feature [+ attributive]; adjectives with [- attributive], such as esiyifiä 'stupid, insupportable, intolerable (of a person)', do not allow such complements. Full passive sentences, with the agent expressed in a PP, may be considered as consisting of an underlying Copula + an Adjective [+ attributive] + PP: zot in fek peye par zot burzua they have just been paid by their boss' Compare the active version of this sentence: zot burzua in fek pey zot their boss has just paid them' Here, pey(e) is a verb. It has the features [+ transitive, + attributive] (although the direct object their money', their pay', etc. remains unexpressed), and the final vowel truncation rule has applied. In the passive version, peye is an adjective derived from the corresponding verb. It has the features [+ attributive, - durative], and the truncation rule has not applied. This non-application of the truncation rule is (within this analysis) a constant feature of such sentences. Let us now take two sentences: (1) (2)

so lipie i n tase da koltar 'his feet are stuck in the tar' dun vade da labutik (M) 'rice is sold in shops'

Here, tase and vade must be analysed as adjectives: the truncation rule has not applied, and in (1) we have the characteristic non-durative stative use of fin. Compare now: (3) (4)

so lipie i n tas da koltar liis feet have stuck in the tar' dun van da labutik (M) 'rice sells in shops'

A re-evaluation of the predicate in Ile-de-France Creole

109

We note here that the truncation rule has applied, and that the completive marker in (3) has a perfective sense. Thus, tas and van1 ° are verbs. There are many other such pairs: ban bras i n käse ek diva 'the branches are/have been broken by the wind' ban bras i n käs ek diva the branches broke with/because of the wind' nuvel pe fane da lari the news is being spread in the streets' nuvel pe fan da lari the news is spreading in the streets' It is evident that this analysis depends crucially on the truncation rule: if it applies solely to verbs, then forms such as tase and vade in (1) and (2) above must be something other than verbs. We arrive thus at our 'adjectives derived from verbs' on the one hand, and transitive verbs used intransitively' (tas and van in (3) and (4) above) on the other (cf. Corne 1977a). The somewhat over-subtle distinctions made in the glosses given for all the above pairs are, at least to some extent, artefacts designed to support the analysis.11 While the distribution of fin in 'passive'-type sentences appears to support the passives-as-adjectives analysis, a major difficulty arises with ape, as we have seen: its co-occurrence with fatige (example in 2.31 above) gives an inchoative sense to the predicate, whereas its co-occurrence with oblize (231 above) does not. And yet both these items occur in passivetype sentences and must (since the vowel truncation rule has not applied) be analysed as adjectives derived from verbs. Such analyses are almost inevitable within the confines of the postulated PS rules. They can be made to Vork' in most cases, but at a certain cost. 235. The '#y>-passive' (Corne 1977a, 1977b: 158-169). In S, there is a structure consisting of the verb gay(e) 'get' (always truncated,gay) which corresponds roughly to the so-called 'gef-passive' in English. Examples: zot pu gap peye par guvernma they will be paid by the government' zot ova pe gay peye they will be getting paid' pagar u a gay tape 'careful you don't get hit' / ti n gay pardone 'he was pardoned' pe gay servi par plizier serviler 'he is being served by several servants' Pier i n gay elekte minis 'Peter was/got elected Minister' lisiegay morde ekpis 'dogs get bitten by fleas'

110

Chris Come i ti gap avoye Sesel par guvernma 'he was sent to Seychelles by the government' sez i n gay repare 'the chair got repaired' tu nu kalu i n gap buar 'all our toddy has been drunk'

Within the framework of our PS rules, this structure may be analysed as NPi gaye Np[NPi Copula AP]. The NPj tends to be animate rather than inanimate, although both occur. This seems to be due to the semantics of the construction. The subject of the main clause is an Agent (or an Instrument in the case of a subject with inanimate reference), the embedded co-referential subject is a Patient or Beneficiary. To some extent, the gap construction often implies sympathy for, agency of, intentionality or responsibility on the part of, or some more or less direct involvement on the part of, the surface subject; however, for at least some speakers, there seems to be a tendency to use the gay construction in lieu of the Ordinary' passive, and the whole matter requires further investigation. The gay construction tends not to occur with certain 'adjectival' kind of item, such as fatige 'tired', araze 'very angry', oblize Obliged', forse 'forced', sipoze 'supposed', okipe 'busy', a-koler 'angry', zalu 'jealous', zoli 'pretty' .. .12 While the truncation rule does not normally apply to the embedded item, it may occasionally do so: u pu gay kup u liku 'you will get your head (neck) cut off i n gap met da prizo 'he got put in prison' tu ban zil ki ti n gay apel Trua-Frer 'all the islands which had been named Trois-Freres' Now IF has a well-attested rule of Indefinite Subject deletion (more clearly seen in M and Ro than in S, where it applies only in embedded sentences): lota, ti degrad karo ar pios (M) 'in the past, they used to clear fields with picks' The deleted subject dimun 'they, one, people' or zot 'they' is recoverable. Thus, where the truncation rule has applied, the structure may be analysed as N?! gaye NP[dimun V NPi ], with subsequent co-referential NP deletion of some kind. 3. It is clear then that the postulated PS rules are inadequate, and that some generalisation has been overlooked. Part of the problem, as far as the recent descriptions of IF are concerned, is purely practical, in that

A re-evaluation of the predicate in Ile-de-France Creole

111

none of the authors of these statements are native speakers of any dialect of IF and thus do not have full control of the data.13 But an equally large share of the blame for the current unsatisfactory state of affairs must be ascribed to the descriptive framework(s) adopted, or, in other terms, to the implicit assumptions about IF structure. To account satisfactorily for the facts noted in the preceding sections, we need an approach which goes beyond a simple Stative vs. non-stative dichotomy. Let us postulate that IF predicates contain abstract semantic units which are represented in surface structure by lexemes such as malad 'sick', aroze 'angry, anger', sate 'sing', etc.14 We shall label all such items 'Verbals' (since the term 'predicate* as used here applies to surface forms including accompanying tense/aspect markers). Our next postulate is that there are three categories of 'Verbals', based on their semantics. These are (i) state, (ii) change-of-state, or process, and (iii) action. Action predicates may be defined as having an Agent as the subject. Process predicates imply a change in the state or condition of the subject, while state predicates establish an equational relation between the subject and themselves. State and process predicates have subjects of several kinds (Patients, Experiencers, Beneficiaries). Within this framework, it is not the case that a given lexical item is always either active, or processive, or Stative. Action, process, and state are semantic concepts; a given lexeme may be active in one predicate, processive in another, and Stative in another, interacting with the surface subject (as well as with underlying embedded subjects) on the one hand and with semantic units of tense, aspect, and mode (present,15 past,(in)definite anticipation, completion, progression, immediate completion) on the other to give rise to the surface meanings. 3.1. We can now give a preliminary sketch of how this approach deals with the facts presented in the preceding sections, facts which, as we have seen, can be handled only imperfectly by the underlying Copula hypothesis. 3.1.1. 'Progression' is represented by (a)pe and indicated extended duration of the event or series of events. When the 'Verbal' represents action, the sense is clearly progressive: zot ti ape plis koko 'they were husking coconuts' With process 'Verbals', the sense is not always as clearly indicated by the English gloss. In some cases, a 'be ... ing' gloss is appropriate:

112

Chris Corne mo pe deperi (Μ) Ί am losing weight' li pe mor (M) 'he is dying' mo pe malere Ί am being unhappy (now)' mo pe okipefer sa (Μ) Ί am busy doing it' mo pe oblize fer sa (Μ) Ί am being obliged to do it'

In others, the sense is better rendered by 'be getting': mo pe dakor ek li Ί am beginning/coming/getting to agree with him' mo pe ά-koler Ί am getting angry' mo pe fatige tan sa topaz Ί am getting tired of hearing that noise' While in yet others, a paraphrase is necessary: mo pe kota li (M) 'I'm crazy about her (at this very moment)' mo pe fe (M) 'I'm starving' mo pe araze (M) 'I'm very angry (now)' In all cases, ape still has its basic sense 'progression'. With a state 'Verbal', (a)pe does not occur, since a state by definition has duration and is completive, i.e. Progression and State are semantically incompatible. In the above examples, the nature of the subject should be noted. Action predicates have an Agent as their subject (since that is how they are defined), but process predicates are more complex. With a single (oneitem) 'Verbal', the subject is a Patient (the predicate tells us what is happening to the subject, as in mo pe deperi for example) or an Experiencer (the predicate says something about the subject's mental disposition, as in mo pe dakor ek li). With 'compound' predicates (i.e. derived by rule, such as fatige fer etc.), the subject is the Patient/Experiencer of the predicate as a whole, but is the Agent of the embedded clause. In such cases, the whole predicate is processive, not solely the main 'Verbal'. Thus *mo pe fatige alone is not an acceptable way of saying Ί am getting tired'. Indeed, a back-translation of processive concepts such as 'she is getting prettier' usually yields sentences containing vin(i) 'to become': li pe vin pli zoli (M) 'she is getting prettier' (*// pe pli zoli) mo pe vin fatige Ί am getting tired' mo pe vin araze (Μ) Ί am becoming/getting furious' or

komas(e)/kumas(e) to begin':

A re-evaluation of the predicate in Ile-de-France Creole mo pe komas -koler

113

am beginning to get angry'

We now have a principled account of the distribution of ape which also accounts for the apparent semantic shifts. There no longer subsists any need for an ad hoc feature to distinguish between oblize and fatige, and the postulated feature [± durative] mentioned in 2.3.1. simply corresponds to process vs. state. Figure 1 illustrates the distribution of ape.

Action Process State Figure 1. Occurrence of ape 3.1.2. 'Completion' is represented by fin. With action 'Verbals', the sense is perfective with respect to present, past, future, or future-in-thepast (see examples at 2.32.). With process 'Verbals', the sense is 'has become'. Examples: ian tro tar 'it will be too late' (will have come to be) banan i n mir komela 'the bananas are ripe now' / n ler pur nu ale 'it is time for us to go' With state 'Verbals' fin does not occur, since a state has by definition already been attained.16 Examples: lerua i bet 'the king is stupid' normalma, tapisri kole lor miray ar lakol (me sä kuyö la, li fin servi kulu) (M) 'wallpaper is usually stuck to the wall with glue (but that dope has used nails)' Figure 2 illustrates.

fin Action Process State Figure 2. Occurrence of fin Again, the feature [± durative] corresponds to process vs. state, and we no longer require to make the distinction between states which are generally true (= state) and those which have come to be so (= completive + process), since this distinction is now inherent. 3.1.3. Final vowel truncation. The formulation of this rule turns out to

114

Chris Come

be rather complex, since truncation applies to 'Verbals' in sentences that are active, processive, or stative (see the sentences given in 2.33.; these sentences were chosen precisely to illustrate the point). We have also seen that the 'transitive verbs used intransitively' (2.3.4.) are subject to the rule. The key seems to lie in the nature of the surface subject.17 Final vowel truncation applies only when the subject is the Agent: divä pe kos ban bras 'the wind is breaking the branches' It applies similarly when the (underlying) subject of an embedded clause is agentive: / deman Za sat sa 'he asks John to sing it' zot pa kon koz fräse 'they don't know how to speak French' (both satfe) and koz(e) have underlying agentive subjects). In a sentence such as: / gap li sok 'he gets a shock' there is co-reference between the subject (Agent) i and the (indirect object) pronoun li 'himself; this co-reference (or reflexivity) produces, for the sentence as a whole, a Patient reading for the subject, but its basic agentive nature is clear. I would suggest, as a tentative hypothesis, that in a sentence such as: labutik iferm siz-er the shop closes at 6 o'clock' the subject labutik is to be considered in the same light: labutik is both the Agent and the Patient offennfej, whereas in: labutik iferme ozordi 'the shop is closed today' it is the Patient (the Agent remaining unexpressed).18 Similar considerations lead me to conclude that items such as kut(e) 'to cost\pez(e) 'to weigh' have in effect agentive subjects: e Ion i kut de rupi One ell costs two rupees' although the net result of such sentences is clearly stative. An even clearer case of a stative sentence is: sa bonom i apel Lik 'that man is called Luc'

A re- evaluation of the predicate in lie- de-France Creole

115

where sa bonom is both Agent and Patient. The case of vinfi) 'to become' may be analysed differently. The presence of this item always (in the sense of 'become') produces a processive predicate: mo pu vin ris

shall become rich'

I would argue that as the subject ofvini, mo is the Agent, while as the subject of m it is the Patient (i.e. mo and ris stand in an equational relation and the sentence is derived by the embedding of mo ris into mo vim"). The rule of final vowel truncation may be (rather crudely) formulated as follows: Verbal' /Agent Figure 3 illustrates. Action Process State Figure 3. Domain of application of truncation rule It is important to note that this formulation of the rule (as can be clearly seen in Figure 3) claims that the 'Verbal' (in a given sentence) which has undergone final vowel truncation represents Action, without prejudice to the ultimate sense of the predicate as a whole (this being the result of the interplay of the subject, the 'Verbal', the predicate markers, underlying subjects, etc.). In other words, this rule is seen as applying consistently, with no exceptions whatsoever, within the appropriate syntactic environments represented here by X (bearing in mind that different speakers may well have slightly different specifications for X), whenever the subject of the 'Verbal' is an Agent. Thus, in the sentence given as an example in 233: eski nu va met zot tu meläz asam? zot tu is the object of metfe) but it is the agentive subject of melaz(e) which is therefore perceived as an action 'Verbal'. This sentence is therefore to be glossed as 'are we going to put them all mixing up together' or as "... them all, such that they mix up together'. 3.1.4. Passives. The following sentences are by now familiar:

116 (a) (b) (c)

Chris Come diva i n käs ban bras the wind broke the branches' ban bräs i n kose ek diva 'the branches are/have been broken by the wind' ban bras i n kos ek diva the branches have broken in the wind'

In (a), kasfej is an action 'Verbal'. In (c) it is also an action 'Verbal' in the sense described above, while in (b) it is a process 'Verbal'. The facts adduced in favour of the analysis of passives as 'adjectives derived from verbs' are accounted for, and the difficulty concerning ape is resolved, as has been shown. The transitive verbs used intransitively' (as in sentence (c) here) are now seen as action 'Verbals' by virtue of the role of the subject as both Agent and Patient. Note in particular that the rule of final vowel truncation is no longer crucial to the analysis, but is simply one piece of evidence among others which support it. 3.1.5. The data concerning the construction with gape are also accounted for. The tendency for the subject to be animate seems to be due simply to the fact that gape is (in this construction) an action 'Verbal' and thus has an Agent as its subject. The reason for the non-occurrence of the construction with 'Verbals' such as fatige, oblize, etc., is not entirely clear. Note that process and state predicates already have Patient-type subjects. In the gape construction, the (»referential subject of the embedded clause has always a Patient as subject. The construction may then be viewed as a strategy for using as state or process 'Verbals', lexemes which are more usually action 'Verbals'. Since fatige etc. more usually occur in process or state predicates than in action predicates, the gape construction is superflous. The major, immediately apparent, surface difference between process and action 'Verbals' is the application of the truncation rule to the latter. The use of the gape construction provides a more marked procedure for differentiating between action and process, and it is perhaps for this reason that some speakers tend to use it more than others. The data available to me suggest that, unlike ape, fin, and the e -> rule, the gape construction does not have a neat cut-off point (see Figure 4). gape + Verbal Action Process

Figure 4. The gape construction

Process State

A re-evaluation of the predicate in Ile-de-France Creole

117

The cases where the 'Verbal' following gay has been subject to the truncation rule fit less well into this perspective. On the one hand, the use of the gaye construction converts the action 'Verbal' into a process 'Verbal'; on the other, the application of the truncation rule shows that the action 'Verbal' has not been so converted. But note that our analysis of an underlying subordinate agentive subject dimun (235.) is completely compatible with the operation of the truncation rule. 3.2. There are a number of consequences to this approach, but we shall here consider only one of them, the surface forms of 'Copula'. 3.2.1. As we have already seen (2.1.), ete occurs in exposed clause-final position (or, more precisely, as the final element of VP) in Interrogatives, Comparatives, and Locative Relatives. It is optional in the present tense. Since we no longer have an underlying copula, the question of whether ete is the surface form of a copula which is deleted in other contexts, or whether it is inserted in these three contexts as a kind of 'support' verb for the preverbal markers, simply does not arise.19 As a state 'Verbal', ete represents the semantic units of location, existence, or equivalence in just those cases where movement rules have shifted other lexical items capable of representing these units. We may represent this crudely, using interrogation to illustrate: Interrogation + Patient + State + Location -> Location + Patient + State + ? Interrog.l ProformJ (e.g. kot u liv i ete? 'where is your book?'). Let us adopt the suggestion (note 15 above) that the temporal semantic units are [± past], since ti is in fact the only marker — apart from φ — which seems to co-occur 'naturally' with ete (cf. the discussion in Baissac 1880: 32n.). Now in this context, [- past] maybe represented on the surface or not, since φ (i.e. the absence of ti) already indicates non-past: kot u liv? 'where your book (non-past)?' kot u liv i ete? 'where your book it non-past locate?' If non-past is to be overtly represented, ete is required and (in the case of comparative clauses) the subordinate clause subject NP must have the appropriate shape:

118

Chris Come me

mua

ipliriski

'he is richer than

mo ete

I am

On the other hand, [+ past] does require to be represented as ti, which is itself a preverbal particle, and so the Stative 'Verbal' ete must occur. 322. The form se. 'Focussed' sentences involve left-dislocation of a consituent and embedding by relativisation. Thus, from: divä i n käs ban bras the wind has broken the branches' by left-dislocation of the subject we get: (sa) diva (,) ki n käs ban bras 'it is the wind that broke the branches' and of the object: (sa) ban bras (,) ki diva i n käse 'it is the branches that the wind broke' The focussed element represents always a state 'Verbal', marked for past if appropriate: pa ti e katiolo ki n disparet, ti e kanot lopes 'it wasn't a dinghy that had disappeared, it was a fishing boat' While non-past is unmarked (the sa in parentheses in the examples above is an article and is irrelevant to the present discussion). Non-past is optionally marked by the subject pronoun / in affirmative 'impersonal' sentences except with clock times (where i is obligatory), but is unmarked with negatives: (i) eposib pur tuy ti 'it is impossible to kill him' u kone i truaz-er e dmi? 'do you realise it is three-thirty?' pa fasil pur koz ek Man 'it is hard to talk to Mary' The use of se appears to correspond to non-past in affirmative sentences (*se pa, *pa se, *se ti, *ti se).This usage is well attested in S, but is considered as French by some informants. The following illustrate the patterns observed so far:

A re-evaluation of the predicate in Ile-de-France Creole

119

se ki mo ule fer u plezir 'it is because I want to please you' (/ akoz is 'more Creole' than se ki here) larazma, se larazma 'a deal is a deal' sä voyaz, se Lina ki n plere 'this time it is Lina who cried' u napa none pur dir akor? no, se tu 'you have nothing further to say? No, that's all' pa so toro, se so zenis ki n mor (Ro) 'it is not his bull, (but) it is his heifer that is dead' While se manifestly derives from French c'est 'it is', its distribution in IF suggests that it is a surface representation of the semantic uints [- past] and [- negative] in the context of focussing, rather than a surface form of copula. It is not inconceivable that the status of se is different from speaker to speaker. Let us suppose that se is indeed a French 'influence'. Nonpast is generally unmarked, while the past is marked by ft. The introduction of c'est/se by French/Creole bilinguals, either now or at some time (s) in the past, does not distort the Creole system at all, and can be adopted by monolingual Creole speakers as an optional, possibly prestigious, nonpast marker. For bilinguals, meanwhile, it quite possibly retains its French structure of subject + copula. Further, since ft' is an overt past marker, the introduction of c'etait 'it was' (so far unattested in my files) would clearly introduce a competing form. 4. This semantics-first analysis accounts not only for all the facts that the PS rules with underlying Copula can handle, but also for some which they cannot. It also gives a very different picture of the way in which the IF predicate system works. I advance it as a tentative hypothesis only, which will require testing against further data;20 in its present form it is no more than a crude approximation, and it remains to be seen whether all surface facts can be accounted for. To name just one area which requires detailed re-examination in the light of this hypothesis, the preverbal markers and the combinations thereof in each dialect need careful reassessment. Another question which is raised, is whether there is any group of speakers for whose grammar art underlying copula represents a valid abstraction. There is a pseudo-problem of a terminological nature with the designation 'Verbal' in the case of some statives, since our state 'Verbals' are often represented on the surface by NPs. For example, in / ft' e bo sarpätie 'he was a good carpenter', the state 'Verbal' is represented by the NP e bo sarpatie. There is an obvious relation between nouns and states; similarly between verbs and actions or processes; adjectives have a less clear-cut relation, representing processes or states, and they seem to fall

120

Chris Corne

somewhere between nouns and verbs as to their nouniness or verbiness (cf. Ross 1972 — his syntactic squish seems compatible with the kind of approach adopted here). That semantic units of the kind suggested here are fundamental in a way that 'parts of speech' categories such as nouns, adjectives and verbs are not, is shown by the use of etymolgical nouns as actions verbs (e.g. lasas 'hunt(ing), to hunt', lapes 'fishing, to fish', kok to steal', etc.; cf. also Bentolila's 1978 analysis of Haitian Creole). The use of a given lexical item in a variety of grammatical functions is a widespread phenomenon in Creole languages (as it is in pidgins) and is presumably explicable, not only in terms of the dynamic processes involved in what Mühlhäusler (1979) calls developmental continua, but also by substratum (substrata?) semantic influences (cf. Huttar 1975). Thus, for IF, the semantic structure of the predicate system indicates substratum (by which I mean 'non-French') influence, while superstratum influence is more marked in the areas of lexical form, segmental phonology, and (surface) syntactic structures (cf. note 13 above). Beyond the scope of this paper is the question of the relation between IF on the one hand, and Reunion Creole (R) on the other. Since R contains a number of features peculiar to itself, such as surface copula, past participles, infinitives, and relatively complex morphophonemic affixation rules (cf. Corne n.d.), an investigation needs to be made into whether the predicate system of any variety of Creole in the R continuum resembles that of IF in a semantics-first approach. If such resemblances are found, the Bourbonnais hypothesis as to the genetic affiliation of the Indian Ocean Creole dialects (note 2 above) will be strengthened. If not, as I suspect may turn out to be the case, it will be seriously weakened. Equally beyond the scope of this paper is the question of the relation between IF and the Creole French dialects of the Americas. As indicated above, Christie's 1976 analysis of predicate marking in Dominican has inspired the present re-evaluation of IF, but there remain enough significant differences between Dominican and IF to render suspect any conclusions as to a common origin (monogenesis).

NOTES 1. An early version of this paper was read to a meeting of the Auckland Branch of the Linguistic Society of New Zealand on 21 March 1979. I wish to thank Ross Clark, KJ. Hollyman, Andrew Pawley and Roger Peddie for their helpful comments at that time. Thanks are also due to P.-M. Moorghen for his written comments on that early version. I am especially grateful to Philip Baker for his detailed and insightful criticism; many of his suggestions have been incorporated here, and in particular his comments as to the importance of the nature of the subject of the sentence (see section 3.0). I must nonetheless accept total responsibility for the analysis offered here, all inadequacies being, as always, all my own work. This paper is based on research funded in part by the Re-

A re-evaluation of the predicate in fle-de-France Creole

121

search Committee of the University of Auckland, to the members of which I wish to express here my sincere thanks for their support. 2. The question of whether this variety came from Bourbon (= Reunion) as maintained by Chaudenson 1974, or whether it is a (partially) independent development as suggested by Baker 1976 (cf. also Βοΐΐέβ 1977b), is irrelevant for our present purposes. Cf. also Corne and Moorghen 1978. The term 'Ile-de-France' Creole is taken from Papen 1975a. 3. For M, v. Baker 1972, Corne 1973; for S, v. Papen 1975b, Bollee 1977a, Corne 1974-75, 1977a, 1977b; for Ro, v. Corne and Stein, in press. See also Papen 1978b fora unified treatment of all three dialects. However, I do not wish to exculpate myself by implicating others: the main thrust of this paper is directed against my own previous work. 4. Examples are all from S unless otherwise indicated. 5. As Teo/and lekol show, a PP may occur with no surface preposition. 6. In the examples given here, / is the subject-reprise peculiar to S (cf. Corne 197475). In M, ete appears to be obligatory in most contexts, but this does not affect the arguments in section 3.2. below. In (S) oli? 'where is?', the -//' may be seen as a fossilised form of the present tense of the copula (Corne 1977b: 63). 7. Papen 1975a, 1978a; McKibbin and Corne, ms. While some fuzzy areas remain, it can be stated that this rule applies when the element following the verb is a direct constituent of the verbal group (object, dative, locative, an 'infinitive', etc.), or, in other terms, when the following element falls within the scope of the verb. When the following element is not a direct constituent of the verbal group (e.g. a sentential complement), the rule does not generally apply. To illustrate, compare mo kon dase Ί know how to dance' with mo kone (ki) mo dase Ί know that I dance'. This formulation may be taken to represent roughly the previous statements concerning final vowel truncation: the phenomenon has been hitherto seen in terms of (i) verbs (ii) constituent structure (i.e. the element following the verb) (iii) optionality or obligatoriness. We show in 3.1.3. below that this overlooks crucially the role of the subject. (Although the matter is not pursued in this paper, I would suggest that the role of constituent structure as well as the question of optionality needs re-examination in the light of the hypotheses advanced here.) Finally, the presence of a sentence constituent (or semantic unit) [+ emphasis] seems to block the rule. 8. Also -/" with two verbs, vinfi) 'to come, become' and sortfi)/surt(i) 'to go out, leave, come fTom'.Surtfi) is the M and Ro form. 9. An exhaustive analysis was made of the distribution of short vs. long forms of verbs in the two works by Baissac (1880, 1888), v. McKibbin and Corne, ms. In the process, it became quickly apparent that the truncation rule was applying to elements which seemed to us to have an incontrovertibly passive (i.e. adjectival, v. below), rather than an active, sense. A total of 37 such cases were noted, including 16 before a locative (vs. 21 long forms) and 13 before a manner PP or adverb (vs. 15 long forms). Having considered (and rejected) the possibility that the truncation rule should be reformulated so as to apply to adjectival forms, we sought an explanation at the level of the basic verb/adjective distinction made by the PS rules given above. We suggested a line of approach very close to the one adopted in this paper. 10. The truncation rule applies simultaneously with two other rules in the case of a subset of verbs containing a nasal vowel. These rules are: 0)

( )

Γ+cons" +voice L+stop Γ-cons" +nas

Γ+consl [_+nas J r-cons~| -nas

/ / / /

-cons |+nas J

# ("+cons"l Unas

122

Chris Come

Thus, v de 'sell' -> να«:

v

d e

(ii) 0 ) J1. \ v α n Φ r-

11. A subtle distinction does exist, however. It involves some kind of contrastiveness or emphasis, the exact nature of which is not quite clear to me. When the short form occurs, the following PP (which may or may not fall within the scope of the verb, cf. note 7 above) is given slightly more emphasis. This emphasis seems to be similar to that involved in 'focussed' sentences (Come 1977b: 196-197). This may be illustrated by: (a)

lam van da labutik, may fren, pa da lafarmasi (M) 'blades sell in shops but not at the chemist's, my dear'

Where labutik is explicitly constrasted with lafarmasi. Compare this with the following, where 'normal' emphasis is involved: (b)

e ta makro, to pa kapav raze? lam vade da labutik. mo siyal tua ki bato lam pa n kule sa ku la (M) 'you pig, can't you shave? Blades are selling in shops. I'll point out that the ship carrying the blades hasn't sunk this time'

Both sentences, with their glosses, were provided by P. -M. Moorghen. He comments thus: 'Given the context, van would be unacceptable in (b). The speaker wants to emphasize that blades are really selling in shops; his friend has no excuse for not getting shaved. In (a) vade would be unacceptable since the speaker is making a general statement: blades sell in shops but not at the chemist's. Now in this particular instance, it is clear that the blocking of final vowel truncation in (b) is of the same nature as in (c): (c)

to kone, mo n rode to zafer la (M) 'you know, I did look for your whatsit'

Here, [+emphasis] requires rode, [-emphasis] would require rod. Thus vade in (b) and van in (a) may be seen as being the 'same' (cf. Corne 1977a: 38). The presence of emphasis in (b) results in an accentuation of vade and an intonation contour (which may include a very brief pause) different from the normal contour of (a) or of (d): (d)

tu ban lam in vade 'all the blades are sold'

12. That is, I have not yet come across any examples (nor do I have any Seychellois speakers to hand). With other adjectival items, it does occur: zotgay fre asuar 'they get cold at night' k kan gay gro, nu kup li (Ro) 'when the cane gets big, we cut it' It also occurs with nominal items, as in: pol i n gay prezida 'Paul got to be president' (has been elected president) 13. As far as my own work is concerned, the fact that I am a native speaker of English with a reasonable knowledge of French has undoubtedly influenced my views about the structure of IF. Nonetheless, it appears to be the case that a very large proportion of the

A re-evaluation of the predicate in Ile-de-France Creole

123

surface phenomena of IF, at least in the varieties that I am most familiar with, do indeed have parallels in French (standard, popular, regional, dialectal, archaic). 14. This approach is inspired directly by Christie 1976 on Dominican Creole French. I am grateful to Pauline Christie for making a copy of her paper available to me. Although a detailed comparison remains to be done, there appear to be both significant parallels as well as significant differences between Dominican and IF. 15. It is not completely clear that 'present' is a valid category here (although most descriptions - Baker 1972 is an exception - seem to assume that it is). A zero marker in IF (i.e. no other preverbal marker occurs) does indeed often indicate the present tense, but this is often the permanent (universal, habitual) present: larivier i kul da lamer 'rivers flow into the sea' But the marker ti 'past' may be omitted once past time has been established, for example in story-telling (Baker 1972: 106) and in subordinate clauses: mo ti kuar i pit al -pie

thought he would/was going to walk'

It seems distinctly possible that tense in IF should in fact be seen solely in terms of past/ non-past (cf. Baker 1972: 107). 16. Philip Baker has pointed out to me a problem in M: fin co-occurs with ena when there is no overt subject: fin ena en gra topaz da Ian 'there was a lot of noise in the street' 17. I owe this observation to Philip Baker (personal communication). Cf. also Baker 1972: 97-105. 18. The parallel with the reflexive verbs and with the 'symmetrical' verbs of French is clear (cf. the discussion in Corne 1977a: 42-43). Other examples are ban bras i n käs ek diva 'the branches have broken in the wind' (ban bras is both Agent and Patient) and ban bras i n kose ek diva 'the branches have been broken in/by the wind' (ban bras is Patient only). 19. Valdman (1978: 235-236) argues that the American Creole French forms ye, se, and zero are contextually conditioned allomorphs of an underlying copula, i.e. Copula -*· in certain contexts. This notion and that of a low-level deletion of copula seem to be equivalent. 20. Philip Baker offers (personal communication) a further piece of evidence for this hypothesis. He had been using the prefix re- as one of several criteria for distinguishing verbs from adjectives (re-ekrir 'to re-write, write again'). He recently discovered an unusual form re-malad in a written text: täsio li al remalad äkor 'lest he fall ill again'. Even if this use of re-malad turns out to be as idiosyncratic as Baker feels it to be, the fact that it can occur at all seems to me to reinforce the hypothesis. Since actions and processes (but not states) can repeat, malad is here being used as a process 'Verbal'. I would suggest, as a tentative postulate, the following: re-

Action Process! State

124

Chris Come

BIBLIOGRAPHY Baissac, C. (1880), Etude sur le patois Creole mauricien. Nancy, Imprimerie BergerLevrault. Reprinted 1976, Geneva, Slatkine. Baissac, C. (1888), Le Folklore de I'ile Maurice. Reprinted 1967, Paris, Maisonneuve et Larose. Baker, P. (1972),Kreol. A description of Mauritian Creole. London, Hurst. Baker, P. (1976), Towards a social history of Mauritian Creole. B.Phil. dissertation, University of York. Bentolila, A. (1978), Creole d'Ha'iti: nature et fonction - function naturelle, Etudes Creoles 1,65-15. Bollee, A. (1977a), Le craole francais des Seychelles. Esquisse d'une grammaire — text es — vocabulaire. Tübingen, Niemeyer. Bollee, A. (1977b), Zur Entstehung der französischen Kreolendialekte im Indischen Ozean. Kreolisierung ohne Pidginisierung. Geneva, Droz. Chaudenson, R. (1974), Le lexique du parier Creole de la Reunion. Paris, Champion. Christie, P. (1976), A re-examination of predicate marking in Dominican Creole. Paper presented to the conference on New directions in Creole studies, University of Guyana, August 1976. Corne, C. (1973), Tense and aspect in Mauritian Creole, Te Reo, 16,45-59. Corne, C. (1974-75), Tense, aspect and the mysterious i in Seychelles and Reunion Creole, TeReo, 17-18,53-93. Corne, C. (1977a), A note on "passives" in Indian Ocean Creole dialects, Journal of Creole Studies, 1(1), 33-57. Corne, C. (1977b), Seychelles Creole Grammar. Tübingen, Narr. Corne, C. (n.d.), Creole French verb morphology: a re-examination. Paper presented to the Second New Zealand Linguistics Conference, Victoria University of Wellington, August 1978. Corne, C. and Moorghen, P.-M. (1978), Proto-creole et liens genetiques dans 1'Ocean Indien, Langue francaise 37,60-75. Corne, C. and Stein, P. (in press), Pour une etude du Creole rodriguais, .Crudes Creoles 2. Huttar, G. (1975), Sources of Creole semantic structures, Language 51,684-695. McKibbin, J. and Corne, C. Ms. La morphologic verbale dans le mauricien du XIX6: implications pour la nature du predicat? To appear in Etudes Creoles. Mühlhäusler, P. (1979), Structural expansion and the process of creolization. Paper presented to the Conference on Theoretical orientations in Creole studies, St. Thomas, Virgin Islands, March 1979. Papen, R. (1975a), 'Nana k nana, nana k napa', or, the strange case of 'e-Deletion' verbs in Indian Ocean Creole. Paper presented to the International Conference on Pidgins and Creoles, University of Hawaii, January 1975. Papen, R. (1975b),,4 short grammar of Seychellois Creole. Typescript, 84 pp. Papen, R. (1978a), Etat present des etudes en phonologic des Creoles de 1'Ocean in dien, Etudes Creoles 1, 35-63. Papen, R. (1978b), The French-base Creoles of the Indian Ocean: an analysis and comparison. Ph.D. dissertation, University of California, San Diego, ui + 617 pp. Ross, J.R. (1972), The category squish: endstation Hauptwort, Papers from the 8th regional meeting, Chicago Linguistic Society, 316-328. Valdman, A. (1973), La copule dans les pailers francais Creoles. Canadian Journal of Romance Linguistics 1,93-111. Valdman, A. (1978), Le Creole: structure, Statut et origine, Paris, Klincksieck.

Ellen Woolford*

6. The developing complementizer system of Tok Pisin 1. INTRODUCTION

Tok Pisin is the major lingua franca of Papua New Guinea. It originated as a pidgin, but it now has more than 10,000 young native speakers (Sankoff and Laberge 1973). Because these native speakers comprise less then 5% of the total number of Tok Pisin speakers and because the language of these native speakers differs little from that of their parents (Sankoff and Laberge 1973), it is difficult to say whether Tok Pisin is currently a pidgin or a creole. Todd (1974) classifies it as an extended pidgin and it could also be called an incipient Creole. Of much interest is the question of whether the changes that are occurring in Tok Pisin during this transition period from pidgin to creole are unique to creolization or whether they can be identified as ordinary processes of language change. The claim will be made here that the mechanism of change by which Tok Pisin is developing a complementizer system is a quite ordinary process of language change, syntactic reanalysis. No special processes unique to creolization are involved.1 If Tok Pisin can be said to be undergoing creolization, then creolization is governed by the same principles that govern other language change. If there does exist some unique kind of language change that should be set apart by the label 'creolization', it is the case argued for by Bickerton (1975). According to Bickerton, in situations where children do not have sufficient exposure to any existing language (including a stable pidgin) to learn it as their native language, the children create a new language out of the fragments of the languages they hear plus their knowledge of universal grammar. This case will not be considered here. In cases such as Tok Pisin in which a stable pidgin is created before * This research was supported in part by NIMH Fellowship number 5F32 MH0724402. I would like to thank Derek Bickerton, Joan Bresnan, John McCarthy, Norm Mundhenk, Pieter Muysken, Haj Ross, Gillian Sankoff and William Washabaugh for valuable comments and discussions. I would also like to thank the people of Papua New Guinea for their help in my research on Tok Pisin. Only actual utterances are used as example sentences in this paper. This article appeared also in a slightly different form in Kenneth C. Hill (ed.), The Genesis of language. Karome Publishers, Inc.' Ann Arbor. 1979.

126

Ellen Woolford

native speakers are acquired,2 the only difference between creolization and other language change (besides perhaps the rate of change) is that the normal balance between changes that lead to simplification of the grammar and changes that lead to increased grammatical complexity is upset. The net result is increasing elaboration of the grammar. (See Bever and Langendoen, 1972 on the normal balance of language change.) The rapid ongoing change in Tok Pisin gives us an unusual chance to study syntactic change in progress. From the investigation of the Tok Pisin complementizer system that follows, it is concluded that, at least in this case, the progress of syntactic change is not unlike that described by Labov and others for sound change. (See Weinreich, Labov and Herzog, 1968 and Labov, 1972.) What appears to be abrupt and discontinuous change in historical perspective is shown, after close examination of the change in progress, to be a smooth progression of small steps joined by the mortar of variation and ambiguity. If we could look back on the present from several hundred years in the future, it might appear that a new phrase structure rule, S -» COMP S, had suddenly been added to the grammar of Tok Pisin. What we see now, however, is a period in which two grammatical systems coexist. This is made possible by the fact that the two different grammatical systems can produce the same surface strings. Thus, during the transition period, the underlying structure of complement sentences in Tok Pisin is ambiguous. The new system comes in gradually by means of step by step changes in the lexicon. Individual speakers vary a great deal with respect to which verbs in their lexicons are marked to take the new type of complement. This process by which syntactic change can occur gradually, without any gap in communication, is called syntactic reanalysis. Langacker (1977) describes syntactic reanalysis as a process in which a sentence is reanalyzed by some language learners as having a different underlying structure. This reanalysis of the underlying structure does not produce any immediate change in the surface string and thus communication is not impaired. Later, however, subsequent changes have surface structure consequences that resolve the ambiguity and verify that reanalysis has occurred. Three different instances of syntactic reanalysis in Tok Pisin have created three different complementizers: one from a general preposition, long to, of, for, etc.', one from an adverb, olsem 'thusly', and one from a conjunction, na 'and'. In their original roles as prepositions and adverbs, these words appear in positions just preceding complement clauses. Since we are observing change in progress, some speakers still use these words in their original capacity, while others have reanalyzed them as complementizers. Because such ambiguity is a necessary part of the change process, it is very difficult for the linguist to find hard evidence that reanalysis

The developing complementizer system of Tok Pisin

127

has actually occurred in any particular sentence. Nevertheless, there is evidence which, although not entirely conclusive, indicates that syntactic reanalysis is in progress in Tok Pisin and a complementizer system is being created. In the absence of movement rules in Tok Pisin that might provide syntactic tests that would clearly indicate the correct underlying structure of particular sentences,3 the best evidence available that Tok Pisin now has S^-type complements and complimentizers is the following: A grammatical system that includes S-type complements and complementizers predicts tha actual behavior of long, olsem, and na much better than does a system without them. Moreover, a fourth complementizer, we, has recently begun to appear in a surface position that did not exist in Tok Pisin before the S structure was introduced. In sections 2 through 5, the details of the behavior of long, olsem, na and we will be presented and the old and the new complementation systems will be laid out and constrated. Evidence and argumentation will be given to support the hypothesis that Tok Pisin is developing a complementizer system by the mechanism of syntactic reanalysis and that that system has the following properties: A. B.

C.

In the lexicon, verbs can be subcategorized to take one of three complement types, PP[P S], s[COMP S], and S.4 There is no sub categorization for particular complementizers. COMP may be filled by any one of the complementizers long, olsem, or na in VP complements and by we in relative clauses. There is an optional rule of complementizer deletion. The frequency of application of this rule increases with decreasing age of the speaker and with increasingly urban environments.

This system correctly predicts the following facts about Tok Pisin: 1. Long is obligatorily present before the complement clauses of certain verbs, but olsem, na and we are never obligatory. This is predicted by the fact that prepositions are not deletable, but complementizers are. When long is a preposition it is obligatorily present, but when it is a complementizer it is optional. The deletability of long is determined by the preceding verb because the selection for PP or S complements is marked on individual verbs in the lexicon. 2. If long is not obligatory, long, olsem and na are interchangeable. This is predicted by the fact that if the structure of the complement is~s[COMP S], any one of the complementizers may be chosen to fill the complementizer slot (except the relative clause complementizer we). If, on the other hand, the structure of the complement is pp[P S], then only the preposition long may appear and as a preposition it is obligatory. 3. There is a great deal of variation between individual speakers as to

128

Ellen Woolford

which verbs take which complement types. This is predicted if verbs are subcategorized in the lexicon for complement type because things marked in the lexicon are notoriously variable between speakers.

2. Long Long is the general purpose preposition in Tok Pisin and it has a wide variety of meanings in different contexts. Some of these are illustrated in the examples below: (1) (2)

(3) (4)

(5 )

Long moning ol i paitim garamut. In the morning they beat the drum. Yupela i go antap long pies. You(pl) go up to the village.

Yumitupela sindaun long dispela arere long wara. We (incl) (dual) sit on this side of river Let's sit on the river bank. Yu ken i kam bek long lukim mipela. you can come back again for see us You can come back again (for) to visit us. Nogat stori long tokim yu. (neg) have story for tell you (I) have no stories (for) to tell you.

I will argue that Tok Pisin has a phrase structure rule that expands or rewrites PP as either P NP or P S. (A similar claim has been made for English by Emonds (1976) and others.) Examples 1 through 3 above have prepositional phrases of the [P NP] type while examples 4 and 5 have prepositional phrases of the [P S] type. Prepositions can never be deleted in either construction. Besides the purpose clauses in examples 4 and 5, there are certain verbs whose sentential complements must always be preceded by long:

(6) (7)

Mi amamas long bekim pas yu bin raitim long mi bipo. I am pleased (for) to answer the letter you wrote to me before. Tambu tru long stil, bipo. taboo real for steal before It was absolutely taboo (for) to steal before.

In contrast to these verbs, there is another set of verbs whose complement clauses are only optionally preceded by long for most speakers:

The developing complementizer system of Tok Pisin

129

(8)

a. / no inap long baibaiyumi ken bekim bek. (neg) possible for (Put) we can repay back It is not possible that we will be able to repay it. b. Bifo i no inap man i stil olsem. before (neg) possible person steal thus Before, it was not possible for people to steal like that.

(9)

a. Ol i no save long öl i mekim singsing. they (neg) know that they make ritual They did not know that they had performed a ritual, b. Ol i no save samting bai kukim ol. they (neg) know thing (Put) burn them They did not know the thing would burn them.

I claim that complements obligatorily preceded by long have the structure Pp[P S], whereas complements only optionally preceded by long have the structure s[COMPS]. A longitudinal study of the frequency of long deletion before complement clauses indicates that the possibility of deleting long in this context is a recent development. Consider the verb inap 'possible or able' since it occurs more frequently with complements than other verbs in the data. From the text in Hall (1943) we find that all complement clauses of inap are preceded by long, but unfortunately there were only two such complements in the data. It is a problem to find clauses strung together at all in this earlier data because people tended to confine themselves to simple sentences, especially in narrative. Nevertheless, the hypothesis that previously all complements of inap had to be preceded by long is supported by data on contemporary speakers. In the data that I collected in 1975, there is a definite correlation between increasing age and a greater percentage of complements of inap preceded by long.5 This data is shown in Table 1. Percent occurrence 6 Speaker Sex Area of Origin Age oflong/inap—S Actual count S. J. K. L. P. G. F.

male male female female

Bogia (Madang) 22 Finschhafen 20 Chimbu 21 Wabag& 35-40 Kainantu female Pindiu 35-40 male Sepik 50 male Finschhafen 45-50

0 0 4 37

0/4 0/5 1/24 21/56

67 83 86

4/6 10/12 12/14

Table L Percentage of complement clauses of the verb inap 'possible, able' that are preceded by long in data collected in Papua New Guinea in 1975.

130

Ellen Woolford

It appears that, prior to World War II, all complement clauses had the structure pp[P S] as in (10).

(10)

öl i mekim singsing they performed a ritual At some point in time following World War II, language learners began to reanalyze sentences like (10) as having the structure in (11).

(H)

ol i mekim singsing they performed a ritual This reanalysis resulted in the addition of a phrase structure rule, S -* COMP S, to the grammar. Subsequently, a rule of complementizer deletion, COMP ->· , was also added to the grammar and the difference between the underlying PP and S structures became detectable on the surface. At the present time, there is a set of verbs that are still sub categorized by all speakers to take pp[P S] complements. Two of these were given in examples 6 and 7. There is also a class of verbs that are sub categorized to take s[COMP S] complements by all speakers. Sentences 8 and 9 were examples of this class. In between these two poles, however, there is a large group of verbs that are sub categorized to take different complements by different speakers. This is what we expect in any situation where change is in progress. If the choice of complement type is specified in the lexical entries of individual verbs (as it is in English, cf. Grimshaw (1979)), it is not surprising that the substitution of S complements for PP complements should proceed through the lexical system at different rates for different speakers.

The developing complem entizer system of Tok Pisin

131

Despite this great amount of variation in the speech community, individuals may have fairly simple and regular grammars. For example, let us consider three speakers in Table 2 who have very different frequencies for the occurrence of long preceding the complements of different verbs. Speaker

Age

G. P. K.

50 35-40 21

Percent use of long preceding the complements of laik laikim giaman tokim 'want'(intr) Vant^trans) 'pretend' 'tell' 0(0/17) 100 (3/3) 50 (1/2) 0 (0/4) 0(0/13) 100(1/1) 100(1/1) 50 (1/2) 0 (0/53) 0 (0/5) 0 (0/3) 0 (0/2)

Table 2. This sort of variation between speakers is just what we would predict if these speakers had the following lexical sub categorization for complement types: G. (age 50) laik[_S] laikim [_S] giaman [_P S] tokim [_S]

P. (age 35-40) laik[_S] laikim [_P S] giaman [_P S] tokim [_S]

K. (age 21) laik[_S]_ laikim [_S] giaman [_S] tokim [_S]

Table 3. Verbs such as laik Vant (intransitive)' whose complements have never been preceded by long, either in the old data from Hall (1943) or in the recent data, are sub categorized to take bare S complements. Verbs such as laikim *want (transitive)' are still sub categorized to take bare S complements by the oldest speakers, but they are now being changed to take more complex complements by younger speakers. If the small number of tokens represented in Table 2 are at all representative, there seems to be an implicational progression such that a verb is sub categorized to take a PP[P S] complement before it is sub categorized to take an s[COMP S] complement. This is further support for the hypothesis that the S complement is created by syntactic reanalysis of the pp[P S] complement. The only thing that needs to be added to this system to completely predict the pattern of variation in Tok Pisin speakers' use of long preceding complement clauses is a variable rule of COMP deletion. As we saw in Table 1, complementizer deletion correlates with speakers' age and probably with the rural-urban continuum. In the introduction to this paper, the fact was mentioned that since the

132

Ellen Woolford

process of syntactic reanalysis necessarily involves structural ambiguity, it is difficult to prove that reanalysis has actually taken place. With the evidence presented up to this point, one might argue that a seemingly simpler model that does not involve syntactic reanalysis at all fits the facts as well as this one does. One might argue that Tok Pisin never had a Pp[P S] complement type at all. Suppose, instead, that Tok Pisin has always had the choice of subcategorizing verbs to take either S or S complements and the only change that has occurred is the introduction of a variable rule of complementizer deletion and a change in the sub categorization of some verbs to take IS instead of S complements. Under this model, all the verbs in Table 3 that are subcategorized to take P S complements would be subcategorized to take S complements, but they would be marked to indicate that the variable rule of complementizer deletion could not apply to them. Evidence supporting the original model involving a pp[P S] complement type will be presented at the end of the next section on the behavior of the complementizer olsem.

3. Olsem Olsem has a wide variety of functions and meanings in Tok Pisin. Some of these are illustrated in the examples below:

(12)

Em i kamap yangpela boi olsem James. He grow young boy like James He grew up to be a young boy like James (i.e. James' size).

(13)

(14)

(15) (16)

(17)

Na insait long bus i nogat ol olsem ol pikinini bilong diwai nabaut and inside of forest pi like pi child of tree around And in the forest weren't there like7 fruit of trees around na olsem laulau? and like Malay apples and like Malay apples? Em i go i stop olsem wan mun samting. he go stay like one month approximately He went and stayed about one month.

Binatan tasol i mekim nais olsem. insects only vibrate thus It was just insects that were vibrating like that. Olsem nau bai yumi tok wanem ? thus now (Put), we say what So now what will we say? Em tingting em behain bai olsem yet.

The developing complementizer system of Tok Pisin

(18) (19) (20)

133

he think it later (Put), same still He thinks it will still be the same in the future. 'Yu lukaut,' em i tokim mi olsem. ΎΟΜ watch out,' he said to me thusly. Elizabeth i tok olsem, "yumi mas kisim ol samting pastaim'. Elizabeth spoke thusly, 'We must get things first'. Yu no ken ting olsem mipela i lusim tingting longyu pinis. you Neg. can think I thusly 1 we lose thought of you complete jthat I You must not think!f like 1I we have forgotten you. [that J

In this last example, 20, olsem as an adverb is in position to be reanalyzed as a complementizer. This sentence could have either of the following underlying structures (ignoring the negative and modal): (21)

yu

In example 23 from a young urban speaker, it is fairly clear from the intonation that olsem is being used as a complementizer. (23)

Na yupela i no save olsem em i matmat? and you(pl) neg know that it cemetery And you did not know that it was a cemetery?

134

Ellen Woolford

The use of olsem as a complementizer is quite similar to the use of that as a complementizer in English, but it is unlikely that borrowing was involved. More likely, the similarity arises from a similarity of origin. Although there is no actual documentation for this process in Old English, it is generally assumed that the demonstrative daet was reanalyzed as a complementizer (Lehmann 1972 and Allen 1977) in much the same way that olsem was reanalyzed in Tok Pisin. (24)

a. 'John left.' He said that. b. He said that: 'John left.' c. He said that John left.

(from Allen 1977:126)

The reanalysis of a lexical item that precedes a direct quote as a complementizer is apparently a fairly common occurrence in language change. Lord (1976) lists a large number of languages including the Kwa languages and other African and Asian languages that have complementizers identical to or virtually identical to the verb meaning say in those languages. She presents detailed arguments for the case in Ewe, including the fact that be 'say' has lost semantic, syntactic and morphological properties of the verb in its role as a complementizer. One example of the sort of structurally ambiguous string that led to reanalysis of be as a complementizer in Ewe is 25, taken from Lord (1976:182): (25)

fia gbe be womagava chief refuse (say) they-PROH-come NEC The chief refused, said they should not come. The chief forbade that they should come.

Further evidence that olsem has been reanalyzed as a complementizer in Tok Pisin comes from the fact that there is phonological reduction of olsem to olse and even to se for some speakers. Mühlhäusler (forthcoming) gives examples of se introducing indirect speech, although he has found it only after the verbs ting 'think' and tok 'say'. Let us now return to the question we left at the end of section two. Does the grammar of Tok Pisin allow a choice between three complement types, S, PP, and S, — the third one being a recent innovation — or does the grammar have only two complement types, S and S? Although one might favor the two choice model on the grounds of simplicity, it turns out that this model leaves many facts unexplained and requires some arbitrary stipulations. The two choice model predicts that when verbs are subcategorized to take S type complements, any one of the complementizers should be able to fill the COMP slot. Nevertheless, an attempt to substitute olsem for long

The developing complementizer system of Tok Pisin

135

in sentences such as 6 and 7 fails, as we see in examples 26 and 27 below:

(26) (27)

*Mi amamas olsem bekim pas yu bin raitim long mi bipo. I am pleased that (I) answer the letter you wrote to me before. *Tambu tru olsem stil, bipo. It was absolutely forbidden that (one) steal before.

In contrast, the three choice model predicts that there will be some complements — the pp[P S] variety - in which only long may precede the complement clause. Moreover, the three choice model correctly predicts that in just those cases, long cannot be deleted. The two choice model could be shored up with a means of subcategorizing for specific complementizers and a means of specifying that complementizers following certain verbs are exempt from the COMP deletion rule. Nevertheless, it would remain an unexplained accident that the set of verbs that is subcategorized to take long is precisely the same set of verbs whose complements are exempt from the COMP deletion rule. There are no instances of COMP nodes that must simply be filled by any complementizer. The three choice model also correctly predicts that whenever long is optional in a complement, the complement is an S type complement and olsem can fill the COMP slot as well as long can. For example, in sentences 8 and 9, the substitution of olsem for long is perfectly grammatical.

(28)

/ no inap olsem boi'- yumi ken bekim bek. It is not possible- that we will be able to repay it.

(29)

Ol i no save olyem samting bai kukim ol. They did not know that the thing would burn them.

4.Na Na 'and' in Tok Pisin is an ordinary coordinate conjunction that joins words, phrases or sentences, just as and does in English. Since na appears in a position just preceding clauses, it is also in a position to be reanalyzed as a complementizer. (30)

Husat tokim yu na yu kam ? who told you and you come Who told you to come?

(31)

Ol papamama na ol hetman i mas tok na ol i marit. (Mühlhäusler (forthcoming)) pi. parent and pi. elder must say and they marry The parents and elders must tell them to get married.

136

Ellen Woolford

Unlike the case of long and olsem, the reanalysis of na as a complementizer actually changes the meaning slightly in some sentences and this makes it possible to find examples in which na must be a complementizer instead of a conjunction. If the clauses in (32) were merely conjoined, the sentence would make no sense. (32)

Gutpela nayu kam. good and you come It is good that you come.

S.We

The three complementizers discussed so far only appear in complements in the verb phrase. They are never used in S constructions in relative clauses. A separate complementizer, we, is used by a very small percentage of Tok Pisin speakers, but only in relative clauses.8 As with the other Tok Pisin complementizers, we is always optional. Although there is no separation between relative clause complementizers and other complementizers in modern English, this may be the exceptional case. Besides Tok Pisin, both Irish and Biblical Hebrew have separate complementizers for relative clauses9 and this was the case in Old English as well (Allen 1977:139). It is highly probable that relative clause complementizers and VP complement complementizers always enter languages by separate routes, althouth they may later merge as they have done in English. Elsewhere in the grammar of Tok Pisin, we is used as the wh word Vhere' by all speakers, as in 33. (33)

Dispela tupela man i stop we? this two man be located where Where are those two men?

As a complementizer, we is simply generated in COMP position in relative clauses and there are no obvious intermediate steps by which this change came about. (34)

(35)

Dispela man i kolim stret man we em i poisinim long en. This man name exact man that he did-black-magic on him This man named precisely the man that (he) performed black magic on him. Mipela ol we i save kaikai saksak em i putim long mipela tasol. we pi that know eat sago he put to us only We who are used to eating sago, they gave it to us only.

The developing complementizer system of Tok Pisin

137

It cannot be argued that we gets into the COMP node by wA-Movement because Tok Pisin relatives have resumptive pronouns which occupy the site from which we would have been moved.10 Moreover, I have presented extensive arguments in earlier work (Woolford 1978a and 1978b) that Tok Pisin has no rule of wA-Movement. Finally, there is evidence that we in this context is a complementizer and not a wA-word in the fact that it loses the semantic information that it carried as the wA-word 'where'. In both example 34 and 35, we has lost all of its locative meaning and is being used with human subjects. Although we did not become a complementizer by the process of syntactic reanalysis as in the case of long, olsem and na, these previous instances of syntactic reanalysis established the S structure and the COMP node. This created a place for we to be generated where none previously existed.

6.CONCLUSION

The data that have been presented in sections 2 through 5 fit well with the hypothesis that Tok Pisin has developed a complementizer system by the process of syntactic reanalysis. A preposition, an adverb, and a conjunction that occur in positions immediately preceding complement clauses have been reanalyzed as complementizers. As a result, a new phrase structure rule, S ->· COMP S, has been added to the grammar. Once this S construction was added, a relative clause complementizer appeared in a position that did not previously exist, i.e., the COMP node. The S type complement is slowly spreading through the verb system as more and more verbs are subcategorized in the lexicon to take this new type of complement. In addition to the variation produced by differences in verb subcategorization for complement type in different speakers, further variation is produced by a variable rule of complementizer deletion. Complementizer deletion is increasing rapidly in younger speakers and in urban areas. All of the changes involved in the development of a complementizer system in Tok Pisin are quite ordinary processes of language change. There is nothing involved that is unique to creolization.

NOTES 1. Washabaugh (1975) makes a similar claim in his paper on the development of the )1 complementizer from a preposition in Providence Island Creole in Colombia. 'The development of complementizers, .. . , is not a unique process for each distinct language, but

138

Ellen Woolford

rather this development in creolization, and perhaps in all cases of grammar elaboration, proceeds generally along the same route.' 2. Whether or not a stable pidgin is created before creolization begins is determined by the social context. In a situation in which speakers from many linguistic groups are mixed together and cut off from contact with their home groups, but not immersed in any new language, creolization begins immediately. This was the case with slaves in English and French colonies in the Caribbean. In a situation such as Papua New Guinea in which speakers of many languages come into contact with each other, but are not cut off from their groups for any great length of time, a stable pidgin is likely to form. A forthcoming volume of papers on this subject, edited by William Washabaugh and myself, will be entitled The Social Context of Creolization. 3. If Tok Pisin had movement rules, it would probably be easier to distinguish between the preposition long and the complementizer long. For example, in ongoing work on Sranan Tongo, Jan Voorhoeve has had some success in distinguishing verbs from prepositions in a similar situation of syntactic reanalysis by the fact that prepositions are fronted with the rest of the phrase while verbs remain behind. 4. This model of lexical subcategorization for complement type owes much to Grimshaw's (1979) account of lexical subcategorization for complement type in English, although the Tok Pisin system is much different from the English one. 5. One old woman about fifty years old from Manus did not fit this pattern at all. There were five instances of the verb inap plus a complement clause in her speech sample and none of them were marked with long. It is likely that she speaks an even earlier dialect in which inap is subcategorized to take only bare S complements. See the discussion following Table 3. 6. All of these speakers were born in rural villages and all of them now live in urban areas, except for K. and P. who live in semi-urban areas. If there is a rural-urban continuum that parallels the age correlation with long deletion frequencies, this would explain why K. and P. are a bit conservative for their age groups in terms of long deletion. 7. This use of olsem is very similar to the use of like in the current dialect of English heard especially in California. Both are used as hesitation fillers and as a means of putting distance between the speaker and his statements. In this example, the speaker is trying to be polite while questioning the ridiculous statement that a Papua New Guinean army platoon was starving in the forest because their rations were late in arriving. 8. Most Tok Pisin speakers optionally bracket relative clauses with the ia marker at one or both ends of the clause, as described in Sankoff and Brown (1976). 9. I want to thank Ken Hale and John McCarthy for supplying this information about Irish and Biblical Hebrew. 10. In 35, an independently motivated rule of free pronoun deletion has deleted the resumptive pronoun in the relative clause as well as the object pronoun in the main clause.

BIBLIOGRAPHY Allen, Cynthia (1977), Topics in Diachronie English Syntax. Unpublished doctoral dissertation, University of Massachusetts, Amherst. Bever, T.G. and Langendoen, D.T. (1972), The interaction of speech perception and grammatical structure in the evoluation of language. In: Stockwell, R. and Macaulay, K.S. (eds), Linguistic Change and Generative Theory, Bloomington, Indiana University Press. Bickerton, Derek (1975), Creolization, linguistic universals, natural semantax and the brain. Paper presented at the International Conference on Pidgins and Creoles, Honolulu, Hawaii.

The developing complementizer system of Tok Pisin

139

Emonds, Joseph (1976), A Transformational Approach to English Syntax. New York, Academic Press. Grimshaw, Jane (1979), Complement selection and the lexicon, Linguistic Inquiry 10,279-326. Hall, Robert A. Jr. (l943),Melanesian Pidgin English. Baltimore, Linguistic Society of America. Labov, William (1972), The internal evolution of linguistic rules. In: Stockwell, R. and Macaulay, K.S. (eds), Linguistic Change and Generative Theory, Bloomington Indiana University Press. Langacker, Ronald (1977), Syntactic reanalysis. In: Li, C. (ed.), Mechanisms of Syntactic Change, Austin, University of Texas Press. Lehmann, Winfred (1972), Proto-Germanic syntax. In: Goetsem, F. van and Kufner, H.L. (eds), Towards a Grammar of Proto-Germanic, Tübingen, Max Niemeyer Verlag. Lord, Carol (1976), Evidence for syntactic reanalysis: from verb to complementizer in Kwa. In: Steever, S, et al. (eds), Papers from the Parasession on Diachronie Syntax, Chicago, Chicago Linguistic Society. Mühlhäusler, Peter (forthcoming), Complementation. In: Wurm, S.A. (ed.), Handbook of New Guinea Pidgin, Canberra, Australia, Pacific Linguistics. Sankoff, Gillian and Laberge, Susan (1973), On the acquisition of native speakers by a language, Kivung 6, 32-47. Sankoff, Gillian and Brown, Penelope (1976), The origins of syntax in discourse: a case study of Tok Pisin relatives, Language 52, 631-666. Todd, Loreto (197'4),Pidgins and Creoles, London, Routledge and Kegan Paul. Washabaugh, William (1975), On the development of complementizers in creolization, Working Papers on Language Universals 17,109-140. Stanford University. Weinreich, Uriel, Labov, William and Herzog, Marvin (1968), Empirical foundations for a theory of language change. In: Lehmann W. and Malkiel, Y. (eds), Directions for historical Linguistics, Austin, University of Texas Press. Woolford, Ellen (1978a), Topicalization and clefting without wh movement. In: Stein, M. (ed.), Proceedings of the eighth annual meeting of the North Eastern Linguistic Society, Amherst, Massachusetts. Woolford, Ellen (1978b), Free relatives and other base generated wh constructions. In: Farcas, D., et al. (eds.), Papers from the fourteenth regional meeting of the Chicago Linguistic Society, Chicago, Chicago Linguistic Society.

Hans den Besten*

7. Marking WH-Movement in Afrikaans 1.INTRODUCTION

The subject matter of the present paper is the syntax of the formative wat in Afrikaans. Wat plays an important role in the syntax of Afrikaans relatives, comparatives and other constructions. Traditionally it is conceived of as one of the relative pronouns. The present paper purports to show that such an analysis is wrong and that in fact wat is a marker for noninterrogative WH-Movement in finite clauses. Consider the following examples: (1)

(2)

(3) (4) (5)

a. die boek wat daar staan the book that there stands b. die geld watjy verdien het the money that you earned have a. die vrou wat so kwaad is the woman that so angry is b. die man wat sy tiefhet the man that she loves die nooi oor wie ons gesels het the girl about whom we talked have die gedig waaroor ons so lank gepraat het the poem which-about we so long talked have hierdie mense wie se afkoms onbekend is these people whose origins unknown are

* A first version of this paper was presented at the annual meeting of the Dutch Society for Linguistics (Algemene Vereniging voor Taalwetenschap) January 21 1978. More or less the same version was incorporated in a paper on the syntax of Afrikaans and creolization (Den Besten 1978a). My interest for wat in Afrikaans dates back to those memorable days in the Fall of 1976 when Henk van Riemsdijk, Jan Koster, and I discussed the syntax of wat in Afrikaans comparatives with Hans du Plessis. I would like to thank the latter one as well as Jan Vorster, Mrs G. Lijphart-Bezuidenhout, and Mrs Elizabeth Eybers for their willingness to be my informants.

142

Hans den Besten

On the basis of the above examples one might conclude that Afrikaans makes use of three relative pronouns: wat, wie and waar. And in fact this has been and still is the interpretation that can be found in traditional analyses of Afrikaans.1 According to these analyses the set of relative pronouns can be divided into a set of dependent forms containing wie and waar and a set of independent forms consisting of only one member, e.g. wat. The set of independent forms is indifferent with respect to the feature HUMAN, whereas the set of dependent forms distinguishes the [+HUMAN] pronoun wie form the [-HUMAN] pronoun waar, about whose syntax more will be said in section 2. and section 4.. Thus, by making use of the (admittedly ad hoc) syntactic features DEPENDENT and PO (= Prepositional Object), one can devise the following paradigm for relative pronouns:2 +DEPENDENT

(6)

-DEPENDENT

+ PO

-PO

+HUM

wie

wie

wat

-HUM

waar

—

wat

The picture gets complicated is one takes free relatives into account. In free relatives [-DEPENDENT] pronouns are sensitive to the feature HUMAN, witness the following two examples: (7) (8)

Wie so praat, is mal Who that way speaks is silly Wat hy se, kan nie waar wees nie What he says cannot true be not

Furthermore, in so far as it is possible to get free relatives headed by relativized PPs at all, the same distribution of dependent forms is found as in normal relative clauses.3 Therefore, it is clear that the [-DEPENDENT] relative pronouns in relative clauses headed by antecedents are somewhat exceptional. And one of the purposes of the present paper is to present arguments that the relative marker wat is not a pronoun but a pronoun but a complementizer in the disguise of a relative pronoun. Now consider the following examples: (9) (10)

Jan het meer geld verdien as wat ek verdien het John has more money earned than what I earned have Jan koop meer boeke as wat Piet plate koop John buys more books than what Pete records buys

Marking WH-movement in Afrikaans

143

The presence of wat in comparative clauses is required. Thus it seems to be the case that the same morpheme that is used in Afrikaans relative clauses shows up in comparative clauses to which — in other languages — rules such as Comparative Deletion and Sub Deletion would be applied. Since wat in relative clauses is usually conceived of as a relative pronoun, the conclusion that Afrikaans comparative clauses are derived by means of an application of WH-Movement seems to be reasonable. The objection that wat ek verdien het in (9) may be a free relative may be right in so far as example (9) is concerned, but it has no force. In section 6.3. we will encounter comparative clauses headed by wat that cannot possibly be interpreted as free relatives. Similarly, there is no way to interpret Sub Deletion comparatives such as (10) as containing free relatives. Therefore, the assumption that WH-Movement is applied to Afrikaans comparatives and Sub comparatives is plausible. Furthermore, one may conclude that the pertinent w/z-element wat may not delete, because the presence of wat in comparative clauses is obligatory. These data are interesting from a theoretical point of view because they constitute potential evidence for the hypothesis put forward by Chomsky (1977) to the effect that unbounded deletions, such as Relative Deletion and Comparative Deletion, must be described in terms of successive cyclic WH-Movement and subsequent deletion of the pertinent whelement in COMP. Thus the derivation of an English comparative construction such as in (11) is supposed to be as indicated in (12): (11) (12)

John is taller than I am a. John is taller than [s COMP I am [.. +WH ..] ] WH-Movement: a => b b. John is taller than [s [ COMP [· · +WH ..] ] I am ] Deletion in COMP: b **· c c. John is taller than [5 [COMP e ] I am]

Chomsky concludes that a WH-analysis may be assumed whenever a construction under consideration has the following properties which are also observed by overt WH-Movement in relative and interrogative clauses: (13)

a. there is a gap b. where the appropriate 'bridge' conditions are met, there is an apparent violation of the Subjacency Condition, the Tensed S Constraint, and the Specified Subject Constraint c. the Complex Noun Phrase Contraint is observed d. the pertinent constructions observe wh-island constraints

Du Plessis (1977) has shown for Afrikaans comparatives that they satisfy

144

Hans den Besten

Chomsky's diagnostic for WH-Movement (e.g. (13)). And Du Plessis (1977) and Van Riemsdijk (1978a) conclude that Afrikaans presents us with overt evidence for Chomsky's WH-hypothesis in that the comparative wh -element — which must be postulated as an abstract entity in the description of languages such as English and Dutch - shows up in Afrikaans in the guise of the morpheme wat (compare (9) and (10)). Van Riemsdijk (1978a) relates this occurrence of wat to the occurrence of the wh -pronoun waar in Dutch comparatives such as the following one: (14)

Jan heeft meer geld verdiend dan waar zijn vrouw op John has more money earned than what his wife on gerekend had (Van Riemsdijk (9)b) counted had

in (14) may not be deleted and Van Riemsdijk concludes that such examples are overt evidence for Chomsky's WH-hypothesis. However, in Den Besten (1978b) it is shown that these Dutch wh -pronouns belong to free relatives. The same paper contains an argument for WH-Movement in Dutch comparative clauses which is based upon the following contrast: (1 5)

(16)

a. Ik heb meer boeken gekocht dan (*dat) Piet gekocht heeft I have more books bought than (*that) Pete bought has b. Deze tafel is twee meter langer dan (*dat) die tafel breed is This table is two meter longer than (*that) that table wide is Hij is eerder lang dan *(dat) hij sterk is He is rather tall than *(that) he strong is 'He is tall rather than strong'

As can be seen the absence of the complementizer dat 'that' correlates with the application of Comparative Deletion or Sub Deletion. It is argued that this regularity must be related to the rule which deletes the lexical complementizer in Dutch relative clauses: (17)

de mensen die (*dat) ik ontmoet heb the persons who (*that) I met have

It is claimed that the same rule applies to the comparative clauses of Dutch and that therefore an example such as (15) a must be derived from an intermediate structure that may be rendered as follows: (18)

a . Ik heb meer boeken gekocht dan ts [cOMP [ . . + WH . .] dat] Piet gekocht heeft]

Marking WH-movement in Afrikaans

145

The general rule deleting the lexical complementizer in (17) will also delete dat in (18)a, yielding (18)b: (18)

b . Ik heb meer boeken gekoch t dan ts [COMP [ · · + WH . . ] e ] Piet gekocht heeft]

At the higher cycle dan 'than' will trigger the deletion of the postulated wA-element, yielding (18)c (= (15)a): (18)

c. Ik heb meer boeken gekocht dan [s tcOMPe l Piet gekocht heeft]

The rule deleting the lexical complementizer in (17), (15)a, and (15)b cannot be applied to the complements of eerder- comparatives. This can be related to the fact that it is not easy to imagine an application of Comparative Deletion to such complements. We will see below that Afrikaans eenfer-comparatives realize a different option (Section 6.3.). Thus it is assumed that prior to the application of deletion rules there signals WH-Movement if the appropriate conditions (i.e. diagnostic (13)) are met. Finally it is noted that those nonstandard dialects of Dutch which have an optional complementizer deletion rule in relative clauses also have an optional lexical complementizer in comparative clauses — which follows from the analysis. Turning back to Afrikaans now, the central claim of the present paper will be that the morpheme wat in Afrikaans serves a similar function in that it signals noninterrogative WH-Movement in finite clauses if and only if the pertinent w/z-element has been deleted. More precisely, it will be claimed that wat is the complementizer dat (also: lot) that' underlyingly which has acquired its deviating phonetic shape via the following rule:

(19)

X - [+WH] - dat - Υ 1 2 3 4 = 1 2 wat 4

»

This rule presupposes the existence of two positions in COMP. Evidence for this assumption will be presented in section 2.. Thus it is assumed that prior to the application of deletion rules there will be an intermediate structure which contains the following COMP: (20)

[COMP [ x " ' . . + W H . . ] war]

In this COMP either wat (or may be dat/lat) will be deleted — which will yield (3) - (5), (7), and maybe (8) - or the wh -element will be deleted —

146

Hans den Besten

which will yield (1) - (2), (9) - (10), and maybe (8). Therefore, the present paper will support Du Plessis (1977) and Van Riemsdijk (1978a) in so far as they claim that the occurrence of wat in (9) and (10) is overt evidence for the WH-analysis of Comparative Deletion and Sub Deletion. However, the interpretation of the syntactic status of wat will be different. The pertinent analysis has been suggested in Den Besten(1978b,fn. 15). A description of wat along the lines of this analysis can be found in Den Bester (1978a). In the latter paper the description of WH-Movement in Afrikaans is embedded in a discussion regarding possible syntactic interference in the development of Afrikaans. The present paper will present a more complete version of the relevant linguistic parts of Den Besten (1978a, section 3.) without further discussion of the historical development of Afrikaans, possible interference, creolization, etc.. The argumentation will proceed as follows: Section 2. deals with some relevant properties of the syntax of Afrikaans, whereas section 3. considers the hypothesis concerning the syntactic status of wat in some more detail. Section 4. presents an argument for the complementizer status of wat in relative clauses. Section 5. will explore the predictions that are made with respect to the behaviour of the object marker vir in relative clauses. Section 6., finally, contains a discussion of the occurrence of wat in a variety of constructions: relative clauses, temporal relatives, cleft sentences, and, last but not least, comparative clauses. A final section 7. will conclude this study.

2. SOME REMARKS CONCERNING THE SYNTAX OF AFRIKAANS

Afrikaans is a language with an underlying SOV word order that makes use of the rule of Verb Second, well-known from the syntax of German and Dutch, to which languages Afrikaans is closely related. Verb Second is a somewhat unfortunate name for a transformation that — depending upon other factors — puts the finite verb in second position in declarative sentences and interrogative questions and in first position in yes/no-questions and imperatives. For the description of the phenomena that interest us here we will assume that the internal structure of the COMP in Afrikaans is as follows: (21)

COMP -> ([+WH]) [±Tense]

[±Tense] designates the position for the lexical complementizer. [+Tense] will lexicalize as dat (or lot) 'that' or of Vhether, if. The complementizer om 'for' will lexicate the position [-Tense]. In root sentences [±Tense]

Marking WH-movement in Afrikaans

147

serves a different function in that [+Tense] will be the receptacle for the finite verb proposed by Verb Second: (22)

X - [+Tense] - Υ - [v +Tense] - Ζ 1 2 3 4 5 => 1 4 3 e 5

(root transformation)

Thus, this rule relates inter alea (23)a and (23)b: (23)

a.. a. —,ofjy —, uj jy vir vii ιιυπι horn %ΚΛΙΚΓΙ gesien rici het

—, whether you him seen have b. Het jy vir horn gesien ? Have you him seen? For further discussion of Verb Second see Den Besten (in prep.). There is ample evidence for the existence of two positions inside COMP in the case of interrogative clauses, witness the following examples: (24)

(25)

a. Ek weet nie hoedat (also : hoelat) hy dit gedoen het nie I know not how-that he it done has not b . Ek wonder wat dot (also : wat lot) hulle doen I wonder what that they do a. Ek weet nie watter klere om aan te trek nie I know not which clothes for on to put not b. Ons weet eintlik nie waarheen om te gaan nie We know actually not where -to for to go not

It is assumed that the w/z-phrases (hoe, wat, watter klere, waarheen, etc.) are put into the [+WH] position with a rule similar to rule (22): (26)

W t - [+WH] -W 2 -[x-'+WH] - W3 1 2 3 4 5 1 4 3 e 5

=>

Note that the occurrence of dot that' in (24) requires a revision of rule (19). WH-Movement and Verb Second will derive interrogative questions such as the following one which is related to the subordinate clause in (24)b: (27)

Wat doen hulle? What do they?

148

Hans den Besten

We will assume furthermore that declarative sentences are also derived by means of WH-Movement, the pertinent wh -elements being demonstrative pronouns. Usually these pronouns are deleted, as is indicated in (28): (28)

Jan (??die) het ek nie gesien nie John (??wA) have I not seen not

According to Bouman and Pienaar (1924) this usage, which is well-known from Dutch and German, is grammatical, but it is my impression that the left Dislocation surface structure which is quite common in Dutch and German is much weaker in Afrikaans. Example (28) is grammatical, though, if the antecedent and the resumptive demonstrative pronoun are seperated by comma intonation. The position of the finite verb in declarative sentences such as (28), (9) - (10), and (24) - (25) can be accounted for in the fashion described above for yes-no-questions and interrogatives. A second phenomenon that is of some importance for the description of WH-Movement in Afrikaans is the existence of the so-called R-pronouns. PPs in Afrikaans are prepositional for the most part and in general one may say that prepositions may not be stranded. However, there is an exception to this statement that is related to an interesting case of syntactic suppletion. The corresponding phenomena in Dutch (and German) have been described in Van Riemsdijk (1978b) and we will take over this analysis for the description of Afrikaans PPs. Let us assume the following (over)simplified base rules for the expansion of the P" that is contained in the PP (i.e. the P'"): (29)

P" -> [+R]P' P' -*· PNP

All pronouns except the [-HUMAN] ones may occur in the NP slot. Preprepositional R-pornouns (locative preforms) must be used instead: (30)

a.oorhom b.*oordat c.daaroor about him about that thereabout (= about that)

The following suppletion rule introduces the feature [+R], which is the central feature designating locative pro-PPs like hier 'here' and daar there': (31)

[+PRO, -HUM] -» [+R] / [P- P _ ]

A simple rule of R-Movement will now move the [+R] pronoun into the

Marking WH-movement in Afrikaans

149

[+R] slot to the left of the P', yielding the example under (16)c. Let us furthermore assume — again simplifying matters — that the base rule expanding S in Afrikaans looks as follows: (32)

S+NP[+R]VP

Now the rule of R-Movement may move the R-pronoun out of the PPinternal position [+R] into the [+R] slot between the subject NP and the VP (see (33)b), from where it may move - by WH-Movement - into COMP (see (33)c):4 (33)

a. Ek wil nie daaroor praat nie I want not thereabout talk not

b. Ek wil door nie gor praat nie I want there not about talk not c. Paar wil ek nie oor praat nie There want I not about talk not The same rules apply whether the R-pronoun is demonstrative (as in (33)) or interrogative or relative. Thus in wh-questions we can find pairs such as thefollowingone:

(34)

a. Waar het hulle oor gesels Where have they about talked? b. Waaroor het hulle gesels? Whereabout have they talked?

An R-pronoun does not have to strand its preposition, witness (34)b (and also (4)). Therefore, a good variant for (33)c is (35): (35)

Daaroor wil ek nie praat nie

Now note that the feature [-HUM] used in rule (31) is a morphological feature and can not always be interpreted as meaning 'nonhuman'. Thus, whereas interrogative waaroor always means 'what about, about what' (compare (34)), relative waaroor may refer either to human or to nonhuman antecedents, although normative grammarians object against such constructions and prefer paradigm (6) as exemplified in (3) and (4). Compare the following example: (36)

die nooi waaroor ons gesels het (=(3)) the girl whereabout we talked have

150

Hans den Besten

This construction will play an important role in section 4.. As an introduction the hypothesis concerning wat expounded in section 1. will be considered.

3. THE STATUS OF WAT' - A HYPOTHESIS

Let us briefly review the tentative argument against the pronominal status of the relative marker wat that was presented in section 1.. It was pointed out that in genitive phrases and in PPs the [+HUMAN] relative pronoun is wie whereas the [-HUMAN] relative pronoun in PPs is waar. On the other hand the [-DEPENDENT] relative pronoun is invariantly wat. This contrasts with the obligatory use of wie as a [-DEPENDENT] relative pronoun in free relatives. Compare the following examples: (37)

a. Hy wat so praat, is mal He who that way speaks is silly b. Wie so praat, is mal (= (7)) Who that way speaks is silly

The [-HUMAN, -DEPEDENT] pronoun in free relatives is wat (compare example (8)). The conclusion that the relative pronoun in headed relatives (relatives headed by lexical antecedents) may be deleted, thereby obliterating the difference between [+HUMAN] and [-HUMAN] pronouns which shows up elsewhere, suggests itself immediately. This implies that wat in headed relatives (and maybe also in certain free relatives, for instance in (8)) cannot be a pronoun and must be the complementizer dot flat which has acquired a different shape. This conclusion is not unreasonable in view of the fact that more languages possess such specialized complementizers. We can think of the relative marker wo in German dialects, the invariable relative marker what in certain English dialects, or the relative marker som in the Scandinavian languages (compare Taraldsen 1978). On the other hand, one may doubt the interpretation of the data on Afrikaans free relatives I have given above. Under my interpretation wie in (7) / (37)b and wat in (8) (provided it is a pronoun) are in COMP. Recently it has been proposed by Bresnan and Grimshaw (1978) to analyze free relatives - at least in English — as generating their wh-elements in antecedent position and not in COMP. A counterproposal can be found in Groos and Van Riemsdijk (to app.). The authors note for Dutch and German that free relatives behave like relative clauses in that they may extrapose freely — at no costs —, whereas Heavy

Marking WH-movement in Afrikaans

151

NP Shift has a marginal status in these languages. The same observations can be made for Afrikaans. Whereas Heavy NP Shift is stylistically marked (compare (38)) the extraposition of free relatives is not marked at all and can be compared with the extraposition of headed relative clauses (compare (39) and (40)): (38)

(39) (40)

a. Ek het die vleis wat ek in die yskas gevind het opgeeet I have the meat that I in the fridge found have eaten-up b. ??Ek het opgeeet die vleis wat ek in die yskas gevind het Ek het die vleis opgeeet wat ek in die yskas gevind het I have the meat eaten-up that I in the fridge found have z.Ekhet wat ek in die yskas gevind het opgeeet I have what I in the fridge found have eaten-up b. Ek het opgeeet wat ek in die yskas gevind het

Therefore, since (40)b corresponds to (39) and not to (38)b we may conclude that the relative pronouns wie and wat in free relatives are in COMP and not in antecedent position. And consequently my doubts as to the pronominal status of wat in headed relatives are justified. Let us now review the alternative analysis I hinted at above. The description of Afrikaans relatives (and comparatives) I would like to propose runs as follows: Relatives and comparatives are derived via the rule of WH-Movement. This rule (i.e. rule (26)) puts the w/z-phrase into the [+WH] slot at S-level. Under the assumption that there is a two-place COMP in Afrikaans (compare section 2.) WH-Movement will yield a COMP of the following kind: (41)

[cOMp[x"'..+WH..]def]

To this structure is applied a rule of Wat Formation, a specified substitution rule. In the previous section we have seen that in doubly filled COMPs containing interrogative wh -phrases the accompanying lexical complementizer dot does not become wat (compare (24)). Therefore, the Wat Formation rule suggested in section 1. (e.g. (19)) must be slightly changed, so that the rule will pick out only noninterrogative wh -phrases. The relevant feature distinguishing relative pronouns from interrogative pronouns is the feature for definiteness. Therefore, rule (19) can be rewritten as follows: (42)

X _ [χ-- +WH,+DEF ] - dat - Υ 1 2 3 4 = 1 2 wat 4

*

152

Hans den Besten

Note that this rule is in fact underdetermined by the data (i.e. (1) - (5)). One might as well propose the following rule: (43)

χ _ [ N -"+WH,+DEF] - dot - Υ 1 2 3 4 1 2 wat 4

=>

We will see in section 62. that this distinction may play a marginal role in modern Afrikaans. To the output of this rule of Wat Formation the following obligatory deletion rules are applied: (44)

(45)

WH-Deletion Wi - [χ·" . . + W H . . ] - wat - W2 1 2 3 4 ^ 1 e 3 4 Complementizer Deletion W! - [ X " · .. +WH .. ] - dat/wat - W2 1 2 3 4 = 1 2 e 4

*

We will assume for the time being that (44) and (45) apply in this order. Note that (45) may be generalized so that it will also optionally apply to wA-clauses, since the presence of dot in wA-clauses — contrary to the presence of om in such clauses — is not obligatory. Finally, let us assume that rule (44) is applied up to recoverability, where recoverability must be defined such that the rule can apply to headed relatives (including (37)a) but not to free relatives. This issue will be reconsidered below. It will be clear that rule (44) in cooperation with WH-Movement and Wat Formation yields the examples under (1) - (2), (9) - (10), (37)a, and (38) - (39). On the other hand rule (45) in cooperation with WH-Movement and maybe Wat Formation will yield (3) - (5), (7), (36), and maybe (8). Now the question is whether there is any other justification for the assumption that wat is a complementizer besides the occurrence of the wh-pronoun wie in free relatives. Section 4. will present an argument for the complementizer status of wat in relative clauses. The complementizer status of wat having been proven, this result can be used to solve a problem — first noticed by Raidt (1969), I think — concerning the distribution of the object marker vir in relative clauses (section 5.). And finally it can be shown that the interpretation of wat as a complementizer facilitates the analysis of certain other constructions such as temporal relatives, clefts, and comparatives (section 6.).

Marking WH-movement in Afrikaans

153

4. R-PRONOUNS AND RELATIVIZATION

We have seen above that WH-Movement may strand prepositions if the Prepositional Object is an R-pronoun (compare (34) in section 2.). Now what about R-pronouns in relative clauses? Up to now we have considered only cases of R-pronouns with Pied-Piped prepositions (i.e. (4) and (36)). Now consider the following example: (46)

Eksamens is goed waarvoor ek bang is Exams are stuff of which I afraid am

Given the distribution of R-pornouns one might expect (47)a to be grammatical and (47)b ungrammatical. However, it is the other way around: (47)

a. *Eksamens is goed waar ek bang voor is b. Eksamens is goed wat ek bang voor is

(47)a is said to be possible in archaic style. However, it is more plausible to interpret such examples as belonging to a netherlandicizing style which never had any basis in the Afrikaans spoken by those whites and nonwhites whose native language is Afrikaans.5 Thus the relative marker in headed relatives is wat whenever we may postulate a single pronoun in COMP. However, this requirement conflicts with the requirement that a Prepositional Object be [+R] if its preposition is stranded. The resolution for this problem is easy if we assume that the invariable element wat is not a pronoun but a complementizer. The derivation of wat in (47)b will proceed as follows: After WH-Movement the COMP in (47)b has the following form:

(48)

a. [COMP [N " ' waar] t[+T] dot ] ]

Wat Formation will change dot that' into wat 'what' — which is in fact the form of the interrogative pronoun wat? Vhat?' as well as one the forms of the neuter relative pronoun in Dutch:6 (48)

b. [COMP [N " ' waar] [ t+T] wat ] ]

Finally, wat will trigger the deletion of waar (rule (44)): (48)

C.[COMP e

[[+T]Wat]]

Since we need these rules to account for the irregular behavior of R-pronouns under relativization (compare (4), (46) and (47)) it is predicted

154

Hans den Besten

that in the derivation of an example such as (2)a (repeated below) there must be an intermediate structure containing a doubly filled COMP as in (49) a. The derivation of wat in (2)a will then proceed along the usual lines: (2)

a. die vrou wat so kwaad is the woman that so angry is

(49)

a.[COMP [N'" wie][{+i}dat ] ] who

that

b. [COMP [N' '' wfe ] [ [+T ] war ] ] c. [COMP

e

[ [+T] wat ] ]

The postulation of wie in the derivational history of wat in (2)a is a reconstruction based upon the occurrence of wie as a [+DEPENDENT] and therefore undeletable - form in (3) and (5) and as a [-DEPENDENT] form in free relatives (see (7)/(37)b). In section 5.2. we will come back to this point. This way we have established the complementizer status of the relative marker wat. Now, one minor issue remains to be answered: What happens to an R-pronoun in free relatives? On the basis of the free relatives in (7)/ (37)b and (8) (repeated below) it was concluded in section 3. that the WHDeletion rule (44) may be applied up to recoverability.7

(7)

Wie so praat, is mal

(8)

Who that way speaks is silly Wat hy se, kan nie waar wees nie What he says can not true be not

It seems reasonable to define recoverability so as to disallow the deletion of wie, because wie is not recoverable in the -absence of a lexicalized head. What is more, the zero head acquires its meaning by the relative clause. Argueing along the same lines one may interpret wat in (8) as being a relative pronoun. However, again R-pronouns behave in unexpected ways: (50) (51)

Ek het gekoop \ waar ions gister qor gesels het \ vvci ] I have bought what we yesterday about talked have Hy will speel met \*WM*r\Sy pa me^ ^erk \ wat J He wants play with what his dad with works

Only in subject phrases does it seem to be possible to use waar:

Marking WH-movement in Afrikaans sc ) wie ] [[+τ j dat ] ]

Since the NP is recoverable, the derivation of wat in (70) (first reading) will follow the usual course:

Marking WH-movement in Afrikaans (71)

161

b.fcoMP [N"'(vilr)wfe][ [ + T , wa/]](byrule(42)/(43)) c- [COMP e [[+T ] wat ] ] (by rule (44))

That is all that needs to be said about the absence of objective vir in relative COMPs. It is interesting to note that the Indirect Object marker vir behaves almost totally the opposite way. Let us recall the pattern of vir with double Object constructions: (59)

a. Ons het (vir) Kowie 'n plaat gegee We have (for) Jimmy a record given

b . Ons het 'n brief *(vir) Piet geskrywe We have a letter *(to) Pete written In section 5 .1 . it was concluded that there are two analysis for vir Kowie in (59)a. Either it is an NP introduced by objective vir or it is a PP. Vir Piet in (59)b, though, is structurally unambiguous: It is a PP. This analysis predicts the existence of two different COMPs when Indirect Objects are relativized:

(72)

a. die man wat jy 'n boek gegee het the man that you (subj .) a book given have b . die man vir wie j'y 'n boek gegee het the man to whom you (subj .) a book given have

The derivation of wat in (72)a is as in (71). The derivation of vir wie in (73)b is as in (73): (73)

a. [COMP [p" ' vir wie ] [ [+T] dot ] ] b. [COMP [p" ' vir wie ] [ Ϊ + Τ ] wat ] ] (by rule (42)) c - [COMP [p " ' vir wie ] e ] (by rule (45))

PPs cannot be deleted — which accounts for the occurrence of vir wie in (72)b as against its absence in (70) (first reading). Now note that speakers of Afrikaans will almost exclusively opt for the PP version of Indirect Object relatives. This corresponds with the fact that the same option is made under interrogative WH-Movement and Topicalization:

(74)

a. Virwie het/y 'n brief geskrywe? To whom have you a letter written? b . Vir Jan het ek 'n boek gegee To John have I a book given

162

Hans den Besten

(Compare (59) above.) Note that in fact it cannot be decided whether vir in (74) is the preposition vir or the Object marker vir. However, that does not matter. It suffices to know that for 'garden path' reasons speakers prefer to use vir. For the same reasons the prepositional option is chosen in relative clauses. This contrasts strongly with the absence of 'garden path' considerations with relativized Direct Objects. As was said above, the latter statement deserves further qualification. In fact not all speakers of Afrikaans reject vir wie in Direct Object relatives. Compare my example (66) a which I marked ungrammatical and the example below: (75)

die man vir wie ek gister raakgeloop het the man whom I yesterday met have

This observation seems to imply that objective vir can be a preposition. However, this conflicts with the evidence I presented in section 5.1. against such an analysis. In fact example (75) can be easily reconciled with the claim that objective vir is a case-prefix on the N'' and not a preposition. There is an interesting correlation between the grammatically judgements concerning (75) and those concerning (76): (76)

die man wie ek gister raakgeloop het the man whom I yesterday met have

Those that do accept (75) will accept (76) as well. Those that don't accept (75) will not accept (76) either. Furthermore, those that allow examples such as (75) - (76) will also accept (77): (77)

die ou man wie vir my hierdie verhaal vertel het the old man who me this story told has

Apparently, for some the WH-Deletion rule (44) is not completely obligatory. Since wie in (76) and (77) is an NP, we may stick to the conclusion of section 5.1. that objective vir is not a preposition and interpret vir wie in (75) as an NP. Normativists decry the use of wie in (76) and (77) as an aglicism — not noting that the same applies to (75). However, it is more probable that the use of wie is truly Afrikaans, since it naturally fits in with the Afrikaans system of relativization as we have reconstructed it in this paper. Those who use constructions such as (75) - (77) apply a system of rules that is only minimally different from the (probably more general) system discussed in this paper. The latter system applies an obligatory rule of WH-

Marking WH-movement in Afrikaans

163

Deletion, the former one applies an optional rule of WH-Deletion. Both rules can be formalized as in (44). Thus, the relative pronoun wie that I postulated in the derivation of example (2)a (see (49) and also compare (71)) on the basis of the occurrence of wie in free relatives and — as a dependent form — in headed relatives can be said to have a basis in linguistic reality. Most probably speakers of Afrikaans will use vir wie (DO) especially for reasons of disambiguation. We know (compare Raidt 1969) that this is the case in the Afrikaans Bible translation. Raidt (1969), a philological study on the use of vir, presents evidence that the occurrence of vir wie (DO) in relative clauses is a minor option in Afrikaans. Raidt discusses the occurrence of vir in four different environments (p. 45-46). Relativization is the sole environment where only a minority of the ten writers whose work she perused uses objective vir, whereas the use of vir is almost universal elsewhere. At least one author, Van Wyk Louw, writes vir wie to create a personal style. Given the above considerations it is clear that the behavior of Direct Objects under relativization follows from a theory that postulates a rule of WH-Deletion (rule (44)). Note that the free deletability of vir wie as a Direct Object as against its undeletability as an Indirect Object is (a) independent evidence for the existence of a rule of WH-Deletion in Afrikaans and for the complementizer status of wat, and (b) independent evidence for the nonprepositional status of objective vir. The facts concerning objective vir in Afrikaans are strongly reminiscent of the behavior of objective a in Spanish (and Portuguese). In Spanish relative a quien(es) (litterally *1ο whom' (sing./pi.)) may delete if it is a Direct Object, whereas it may not if it is an Indirect Object. This does not come as a surprise if we consider that Afrikaans vir must be related to the Iberian a via Creole Portuguese (and maybe Lingua Franca) per.10 With the data presented above the complementizer status of wat seems to me to be sufficiently established. In the next section we will show that wat is a marker for WH-Movement and WH-Deletion in a variety of constructions. First, we will reconsider the data on relative clauses and add a couple of observations. Then we will discuss the use of wat in cleft sentences and finally we will return to the use of wat in comparatives we hinted at in section 1..

6. 'WAT' AS Λ MARKER FOR WH-MOVEMENT AND WH-DELETION

6.1. 'Wat' under relativization In section 5. evidence was presented that the relative marker wat is a com-

164

Hans den Besten

plementizer. This evidence is based upon the cooccurrence of wat with stranded prepositions, the noncooccurrence of vir with wat in Direct Object relatives, the deletability of accusative vir wie in relative clauses, and we may add to this the dialectal use of wie (S) and (vir) wie (DO) as well as the opposition between free relatives and headed relatives as regards wie (see section 1. and section 3.). The evidence for WH-Movement in Afrikaans relatives is based upon the occurrence of relative PPs and genetive phrases in COMP as well as upon the occurrence of wie in the COMP of free relatives and the dialectal of wie (S) and (vir) wie (DO) in the COMP of headed relatives. Thus wat in the following examples is a special complementizer the phonetic appearance of which has been triggered by a wh -element that has subsequently been deleted: (78 )

a. die mense wat (vir) my huh-toe gebring he t the folks that me home brought have b. die mense wat hy huis-toe gebring het the folks that he home brought has c. die probleme wat onsoor gepraat het the problems that we about talked have

To these observations can be added some observations on the occurrence of wat in temporal relatives. These observations are interesting for independent reasons because they provide us with more data bearing on the problem of recoverability discussed in section 4.. The observation on the use of wat in temporal relatives is due to Kempen (1962). Kempen notes that there are three options for the COMP of a temporal relative:

(79)

toe die eerste keer dot hy hier was wat when the first time that he here was that

Toe is a relative PP. In the second and third variant of (79) toe has been deleted and the complementizer dot optionally becomes wat. The use of dot in (79) may belong to a netherlandicizing style, because the first and the second option in (79) are possible in Dutch (de eerste keer toen/dat hij hier was) and because of the overall popularity of wat in Afrikaans (see below). The use of wat can be related to the optional absence of that in English temporal relatives (the first time (that) he was here) —

Marking WH-movement in Afrikaans

165

which is an indication for the syntactic presence of a phonologically absent wh-pronoun. The deletability of toe in (79) must be related to the fact that in Afrikaans NPs can be used as temporal adverbials. Thus, semantically speaking, the deletion of toe is recoverable on account of the semantics of the full NP. However, there may be individual differences as to when temporal w/z-elements are recoverable. One of my informants was very definite about the following paradigm: (80)

(81)

a. Die dag toe ek moes promo veer, was ek sieh The day when I had-to defend-my-thesis was I ill b. *Die dag wat/dat ek moes promoveer was ek siek a. *Op die dag toe ek moes promoveer, was ek siek At the day when I had-to defend-my-thesis was I ill b. Op die dag wat/dat ek moes promoveer, was ek siek

Thus, for some toe is recoverable if the NP antecedent has an adverbial function, whereas for others toe is recoverable only if the antecedent is preceded by a temporal preposition. Note the obligatoriness of the deletion in (81).11 It will be obvious that wat in (79) - (81) cannot be a pronoun — which tallies with the conclusion we arrived at in the preceding section. In the following two subsections we will encounter other environments where wat has a general 'relativizing' function. 6.2. 'Wat'in clefts It will not come as a surprise that wat shows up in Afrikaans cleft sentences with an NP in topic position, given the presence of ννΛ-phrases in the COMP of the corresponding construction in other languages such as Dutch and German. Compare the following examples: (82)

a. Dis (vir) Jan wat ons gesien he t It's (OM) John that we seen have bl. Dis nie Gnl. Van den Bergh vir wie hy bang is nie It's not Gnl. Van den Bergh of whom he afraid is not b2. Dis nie Gnl. Van den Bergh wat hy bang voor is nie It's not Gnl. Van den Bergh that he afraid of is not

It is interesting to note, though, that in PP clefts the complementizer wat is strongly preferred over dot 'that' (the correct form in Dutch). Consider the following examples:

1 66 (83)

(84)

Hans den Besten a. Dit was toe wat/dat my voder so moesja (Scholtz 1963) It was then that my father so had-to hurry b . Dis van bangigheid watjdat hy so bewe (Scholtz 1963) It's out of fear that he so trembles a. Dis oor Biko wat/dat ons gepraat het It's about Biko that we talked have b . Dis met mnr. Woods van Oos-Londen wat/dat ons gesels het It's with mr. Woods of East-London that we chatted have

The derivation of wat in (83) - (84) is straightforward. Thus, the complementizer wat in (84)b may be derived the following way: (85)

a. [COMP [P' ' ' ™et wie ] [ (+T1 dot ] ] with whom that b · [COMP [p' ' ' net wie ] [[+T ] wat ] ] (by rule (42)) c-tcOMP e [[+T]wai]] (by rule (44))

Note that no new rules are needed for the derivation of wat in PP clefts. The same rules that were postulated for the analysis of relative clauses can be used for clefts. Wat Formation will change dot into wat and wat will trigger WH-Deletion because the pertinent w/z-PPs are recoverable, on account of the fact that their antecedents are PPs and not NPs. That wat is used in the examples in (84) is not really surprising. The clefted PPs in (84) must be linked to a VP-internal position in the subordinate clause and WH-Movement is an obvious candidate to perform that task, as was pointed out in Chomsky (1977). No such linkage is required in the case of the clefted PPs in (83)a and b. The pertinent PPs in topic position are 'circumstantial' adverbials and the corresponding embedded clauses might be conceived of as Subject complements. No such interpretation is possible for the sentences in (84). Scholtz (1963) was the first one, I think, to point out that in PP clefts dat is — as he terms it — increasingly being replaced by wat. It may not be without significance that the examples he adduces to illustrate his point (i.e. those in (83)) are of the 'circumstantial' type. Therefore, Afrikaans has generalized WH-Movement for the two types of PP clefts that have to be kept apart for English (see Chomsky 1977). In so far as the spoken language admits dat in PP clefts at all, dat may be expected to be best in 'circumstantial' PP clefts.12 Finally, a small remark about styles in Afrikaans. The use of wat in PP clefts is pervasive in spoken Afrikaans, whereas the written netherlandicizing style, allows both wat and dat in such constructions. On the other hand wat (or overt wA-phrases) are required in NP clefts for both styles. Now, given what we know about the syntactic analysis of objective vir (see section 5.1.) we can predict that objective vir + NP will require wat if it shows up in topic position in clefts.

Marking WH-movement in Afrikaans

167

In fact, such constructions do exist. First note that horn in example (86) is in the oblique (accusative) Case. Therefore, it is not surprising that objective vir may be used as well. Compare (87)a and b: (86) (87)

Dis horn wot ons gesien het It's him that we seen have a. Dis vir horn wot ons gesien het b. Dis (vir) Pie t wat ons gesien het It's (OM) Pete that we seen have

And the prediction that the formal style will not allow dat 'that' in COMP is borne out: (88)

a. Dis vir Jan wat ons raakgeloop het b. *Dis vir Jan dat_ ons raakgeloop het It's OM John that we met have

These data constitute independent evidence for the nonprepositional status of objective vir (compare section 5.). 6.3. 'Wat'in comparatives We may now return to the analysis of Afrikaans comparatives. In section 1. it was noted that wat shows up in Afrikaans comparative clauses both under Comparative Deletion and under Sub Deletion. The analytical problem we posed ourselves as regards comparative wat has in fact been solved. As early as 1924 it was noted (by Bouman and Pienaar) that in comparative clauses the subordinator dat is replaced by wat — 'eintlik die betrekHke voornaamwoord' as Bouman and Pienaar put it hesitantly ('actually the relative pronoun', p. 212). Du Plessis (1977) and Van Riemsdijk (1978a) conclude that wat is the relative pronoun and that therefore Afrikaans comparatives present us with overt evidence in favor of Chomsky's WH-analysis of Comparative Deletion. My doubts as to the pronominal status of wat have been proven to be right. However, since the surface form wat is triggered by the presence of a w/z-phrase that is subsequently deleted, the central claim in Du Plessis (1977) and Van Riemsdijk (1978) can be upheld: The presence of wat is overt evidence for Chomsky's WH-hypothesis concerning comparative clauses. The analysis of wat as a complementizer facilitates the analysis of Afrikaans comparatives. Whatever constituent has been comparativized, or, to put it differently, whatever constituent has been wiped out by Comparative Deletion, wat will show up. Compare the following examples:13

168

(89)

Hans den Besten

a. Hy het meer boeke geskrywe äs wat ek gelees het He has more books written than what I read have b. Sy het vinniger geloop as wat ek gehardloop het She has faster walked than what I run have c. Ons moet nog versigtiger skrywe as wat ons praat We must still more cautiously write than what we speak d. Dit is nou makliker as wot dit ooit was It is now easier than what it ever was

The invariability of wat may be a problem for those who analyze wat as a pronoun, but it follows as a natural consequence under an analysis that assigns wat to the class of complementizers. Thus, the description of Afrikaans comparatives will not essentially differ from the description that is proposed in Den Besten (1978b) for the analysis of Dutch (and German) comparatives (compare section 1.). As than, as' will be analyzed as a sentence-introducing preposition and the derivation of wat in for instance (89)b will proceed as follows: WHMovement will yield a doubly filled COMP as in (90)a: (90)

a. [COM? [A'" hoe vinnig ] [ [+T] dot ] ] how fast that

To this COMP the usual rules are applied: (90)

b. [COMP [A' ' ' hoe vinnig ] [ [+T] wat ] ] c. [COMP e [[+T]Wflf]]

(by rule (42)) (by rule (44))

The rule of WH-Deletion (rule (44)) will obligatorily delete comparative whphrases in COMP. This will ensure the absence of wh-phrases in comparative COMPs in the simple cases where APs and [-DEPENDENT] NPs are involved. However, an extra filter of a very simple type is needed in case a [+DEPENDENT] NP is comparativized. First note that comparativized NPs are quantified expressions by definition (compare Bresnan 1973 and Den Besten 1978b). Now, such NPs in Afrikaans may not be extracted out of PPs because there are no [+R] quantifiers. There are only [+R] pronouns. Therefore, under our analysis one would expect that example (91)b — which corresponds to (91)a — would be grammatical, since the quantified wh -phrase must Pied-Pipe its preposition whereas the corresponding preposition in the main clause is not included in the comparative antecedent. Nevertheless, (91)b is ungrammatical — as is (91)c:

Marking WH-movement in Afrikaans (91)

169

a. Hy het met meer mense gepraat as ek He has to more people talked than I (S) b. *Hy het met meer mense gepraat as met hoeveel mense ek He has to more people talked than to how-many people I gepraat het talked have c. *Hy het met meer mense gepraat as wot ek gepraat het He has to more people talked than what I talked have

The ungrammaticality of (91)c follows from the theory. The ungrammaticality of (91)b does not. Therefore, in order to account for the ungrammaticality of sentences such as (91)b — which, by the way, are not totally outrageous — we need a simple filter to exclude the occurrence of whphrases in comparative COMPs. Note that no such filter is needed in the case of relatives or clefts:

(92)

*as -,[X' - . · +WH .. ]

This filter does not contain an unless-clause (compare Chomsky and Lasnik 1977) nor does it overlap with the deletion rules (44) and (45). Hopefully an independent explanation will be found that will obviate the need for a filter. Filter (92) seems to be contradicted by examples quoted in Du Plessis (1977) and Van Riemsdijk (1978a). Consider the following sentences taken from Du Plessis (1977): (93)

a. Ons werk vir heelwat meer geld as waarvoor jullie werk We work for much more money than for-which you (pi.) work b. Jan hou van meer soorte vrugte as waarvqn sy vrou hou John likes more types of fruit than (of)-which his wife likes

There is no reason to despair, though, since the constituents introduced by waarvoor and waarvan respectively are free relatives. As I said above quantified NPs do not have a [+R] variant. Therefore, waar in (93)a and b is a relative pronoun. The free relatives in (93) are shared nominals. This means that in (93)b the types of fruit John's wife likes are subsumed under the types of fruit John Likes himself. For more particulars I refer to the discussion in Den Besten (1978b) of similar constructions in Dutch that were dealt with in Van Riemsdijk (1978a). In the former paper it is suggested that all filters (and all specified deletion rules) are subject to the Subjacency Condition (see Chomsky 1973). Therefore, the sentences in (93) do not pose any problem for filter (92). Filter (92) does not 'see' as waarvoor or as waarvan.

170

Hans den Besten

As expected the following variants of (93)a and b are grammatical as well: (93)

a'. Ons werk vir heelwat meer geld as wot julle voor werk b'.Jan hou van meer soorte vrugte as wot sy vrou van hou

These examples superficially resemble the following sentence quoted from Verhage (1967): (94)

a. Die aantal werkwoorde is v eel groter as wat ons bewus van is The number of verbs is much larger than that we aware of are

The free relative in (94) may be interpreted as die (aantal) werkwoorde wat ons bewus van is 'the number of verbs we are aware of, but in that case the free relative is not a shared nominal. If we undo Comparative Ellipsis in (94)a we get the following admittedly horrible sentence with two wats in a row. The two wats and the respective extraction sites have been coindexed to make parsing easier: (94)

b.Die aantal werkwoorde is veel groter as wati [Ν" ' Δ [jj wat2 ons bewus 2 van is ] l is ]

Since wat is used in many different constructions it will not come as a surprise that some comparatives may be structurally ambiguous. Consider again example (9): (9)

Jan net meer geld verdien as wat ek verdien het John has more money earned than what I earned have

The wat·clause in (9) may be either a comparative clauses or a free relative that is a shared nominal. However, both syntactic analysis will yield the same semantic reading (compare Den Besten 1978b). Before this subsection is closed two more types of comperative constructions must be considered. First, Sub Deletion comparatives. In view of the conclusion we have reached concerning the syntax of wat we may conclude that Sub Deletion in Afrikaans is also derived via WH-Movement and WH-Deletion witness the following examples: (95)

a. Hy het meer boeke gekoop as watjy plate gekoop het He has more books bought than what you (sing.) records bought have b. Die vrou is nes vet as wat haar ou mannetjie lank is That woman is just as fat as what her husband tall is

Marking WH- movement in Afrikaans

171

Such examples were also noticed by Du Plessis (1977) and Van Riemsdijk (1978a). Second, eerier-comparatives. In section 1. the following Dutch sentence was presented, and it was noted that in such constructions dot 'that' may not delete: (16)

Hij is eerder lang dan *(dat) hij sterk is He is rather tall than *(that) he strong is 'He is tall rather than strong'

The undeletability of dot in (16) was explained under reference to the semantic structure of that sentence. Sentence (16) may be conceived of as a comparison between two clauses as indicated in (96): (96)

eerder [