文档库

最新最全的文档下载
当前位置:文档库 > 3 Semantic constraints on children's

3 Semantic constraints on children's

Semantic constraints on children's structural choices of relatives

Kerstin Nauclér

Dept of linguistics

University of G?teborg

Abstract

Children's mastery of relative clauses has been a recurrent issue in psycho-linguistic li terature. The results obtained are, however, fairly i nconsi stent and different theoreti cal clai ms have been made. Chi ldren's acquisition has been explained ei ther with reference to syntactic complexity and processing constrai nts on structural confi gurati ons, or with reference to discourse strategies and communi cati ve functi on. An analysi s that takes i nto account the i nteracti on between syntax, semantics and pragmatics, can offer a very regular description of chi ldren's producti on of relati ve clauses. The constrai nts we find on premature grammars can be traced back to previous cogni ti ve development. I will present data on Swedish relati ve clause production in natural discourse. There are two bilingual learner groups and thei r results will be compared to that of a monoli ngual Swedish learner. The results could best be explained wi thi n an integrated vi ew of language development (Berman (1984), whi ch considers different li ngui sti c as well as language i ndependent variables to be i mportant determi nants of chi ldren's language acqui si ti on.

3

1. Introduction

The answers to how, when and why chi ldren acquire relative clauses will depend on whom you ask.0 In production, Engli sh speaking chi ldren start using relati ves around the age of three, but thei r structural knowledge of different relati ves does not seem to be completed before the age of six (Bloom et al., 1980; but see Hamburger & Crain, 1982 for a somewhat di fferent picture). Swedish speaking children start well before thei r thi rd birthday and they seem to master the different structures between three and four0 (Lundi n, 1987, Plunkett & Str?mqvist, 1990). Chi ldren's first relati ves are used to specify objects and persons, and are typically specifications of objects in the mai n sentence (Limber, 1973, Menyuk, 1971, Romai ne, 1984, Plunkett & Str?mqvist, 1992).

Chi ldren's comprehensi on of di fferent relati ves has been tested in experi mental enactment tasks (Sheldon, 1978, Tavakolian, 1981, De Vi lli ers, et al, 1979, Hakuta, 1981, Goodluck & Tavakoli an, 1982). The general conclusi on is that during the pre-school age, chi ldren find most relati ves difficult to understand. There is, however, less agreement on whi ch structures are most difficult and different explanati ons have been proposed0. Generali si ng over different studies you will find two different perspectives on what mi ght explai n chi ldren's difficulties. One is to explain chi ldren's acqui si ti on with reference to syntactic strategies. Structural complexity and different processing constrai nts are claimed to account for the varying results. Another perspective is to account for children's acquisition of relati ves as a result of discourse strategies and communi cati ve functi on. Relati ves play an i mportant role as a means to handle i nformati on flow in spoken discourse (Fox & Thompson, 1990). The need of specification of new i nformati on and the frequent use of relati ves as comment specification has 0I am greatful to Anders-B?rje Andersson, Lars-Gunnar Andersson and Sven

Str?mqvist for valuable comments and suggestions.

0 For cross-linguistic differences in language acquisition, see Slobin, 1985, 1986. 0For a critique of the validity and reliability of different comprehension studies, see Hamburger & Crain (1982) and Stevenson & Sims, 1993.

4

been stressed by e.g. Romaine (1984) and Plunkett & Str?mqvist (1992).

I will argue for an integrated view of language acquisition (Berman (1984), 1986), where the i nteracti on of different aspects of the li ngui sti c and the non-li ngui sti c i nput, as well as cognitive prerequisites mi ght help us to account for children's relati ve clause grammar.

We shall first have a quick look at the li ngui sti c phenomena i n question. I will then report on an ongoing study of two groups of bi li ngual chi ldren who are acquiring Swedish as a second language. Thei r acquisition of relati ves will be compared to that of a monoli ngual learner of Swedish. The analysis will focus on the i nteracti on between semanti c features and structural choices in child spoken discourse.

Swedish relative clauses

Syntactically, the relative clause is a complex noun phrase, consisting of a Head (Head NP) with an embedded relative clause, exemplified in (1).

(1)'en sak [som man s?tter t p? h?ret]'

'a thing [that you put t on the hair]'

Head NP[ relative clause ]

Semanti cally, the referent of the Head NP is specified by the i nformati on gi ven in the relati ve clause. In the restrictive relati ve clause (RRC) the domai n of possible referents of the Head NP is restricted to a smaller set by the i nformati on gi ven in the relati ve clause in order to identify a specific referent (Keenan & Comri e, 1977). In a non-restri cti ve relati ve (RC) the set of referents of the Head NP is not affected by the i nformati on gi ven in the relati ve clause. The i nformati on i n the relative clause is thus not necessary for i denti fi cati on of the referent.

5

Adding to the semanti c complexity of relati ves is the unfi lled NP posi ti on in the relati ve clause (t in ex. (1), above) whose referent may be verbally expressed on an arbitrary distance from the relati ve element ('som' ('that')) or the antecedent ('en sak' ('a thi ng')).

Relati ves serve different communi cati ve functi ons; to i ntroduce, to characterise and/or to identify referents i n discourse. A more general functi on of relati ves is, however, to make it possible for us to refer to anything we would like, even if we do not know the proper lexicalized expression. Thi s i s what Schachter (1973) has termed 'the very Rai son d'être' of relati vi zati on, exempli fi ed below, when a child is referring to a 'ski ppi ng rope' by the utterance: s?nt som man kan h o p p a ('such that you can skip').

In spoken discourse, it is notori ously difficult to ascertain whether the relati ve serves as a RRC or a RC (see e.g. Fox & Thompson, 1990). There is also good ground to believe that pre-school chi ldren pri mari ly don't use relati ves for identity function, but rather to characterize referents they introduce i nto discourse, which is also the case in adult spoken discourse (Fox, 1987). The di sti ncti on between RRC and RC will not be upheld in this paper.

Si mpli fyi ng somewhat, the di stri buti on of relati ves in di fferent li ngui sti c contexts is summari sed in the followi ng examples taken from our data (sse the following section).

RELATIVE COMPLEX

NP[rel]A:Vad ?r det h?r f?r n?t da?

B: s?nt [som man kan hoppa]

A: what is this then?

B: such [that you can skip]

CLEFT CONSTRUCTIONS

Det ?r X [rel]

Det ?r en katt [som sitter d?r]

It is a cat [that is sitting there]

'There is a cat sitting there'

6

RIGHT EMBEDDED IN A SVO-MATRIX

X V NP[rel]

Kolla den vackra bron [som Monet m?la]

Look at the beautiful bridge [that Monet painted] CENTRAL-EMBEDDED IN A SVO-MATRIX

NP[rel] V X

Dom [som har ingenting] dom ?

m a m m a

Those [who have nothing] they are

Mummy

The syntactic complexity of relati ves has been discussed with reference to embeddedness and focus. Embeddedness refers to the grammatical role of the complex noun phrase in the matri x sentence, exemplified in (2) and (3).

(2)Subject matri x Dom [som har n?nting] dom ?r pappa

Those [who have something] they are

Daddy

(3)Object matrix Ge mej den [som du har]

Give me the one [that you have]

Focus refer to the grammatical role of the head noun in the relative clause, (4) and (5).

(4)Subject focus Dom [som har n?nting] dom ?r pappa

Those [who have somethi ng] they are

daddy

(5)Object focus Ge mej den [som du har]

Give me the one [that you have] Combining the embeddedness and focus variable we end up

with structures like the following.

7

(6)SS Dom [som ?r pojkar] f?r leka med

pojkleksaker

Those [who are boys] may play with

boy's toys

(7)SO Den h?r lilla s?ngen [som ja h?rt i

morse] den var f?r l?nge l?nge sen

This little song [that I heard this

morning] it was a long long time ago (8)OO Ge mej den [som du har]

Give me the one [that you have] (9)OS Du f?r den [som st?r d?r]

You can have the one [that is

standing there]

The syntactic complexity of different relative structures can now be stated with reference to the embeddedness and/or to the focus factor. The Interrupti on hypothesis (Slobi n, 1973) and the N-V-N-strategy (de Vi lli ers et al. 1979) take the embeddedness factor to be the main cause of processing difficulties.

The Accessibility Hypothesis (Keenan & Comri e, 1979, Keenan & Hawkins, 1986) applies to focus only and states that subject focus cause less difficulties because of the conceptual primacy of 'subject'.

The Parallel Hypothesis (Sheldon, 1974), the Conjoi ned Strategy (Tavakoli an, 1981) and the Perspective Hypothesis (MacWinney, 1978) claim that there are certain combinations of embeddedness and focus that make it difficult for chi ldren to process some of the relati ve structures. Sheldon and Tavakoli an formulate thei r constrai nts on purely syntactic grounds, whereas MacWi nney sees the Perspective Hypothesis as a more general cogni ti ve strategy, where the perspective of the 'Agent' could be taken as the pragmatic counterpart of the formal category of 'subject'.

8

None of these principles could alone account for varying results in previous studies, nor for the findings reported here. What I will try to show, however, is that we can get a very regular description of chi ldren's relati ves by an analysis of the interaction between semantics and syntax in spoken discourse.

2. Data, analyses and results

2.1 Data

Data has been taken from two bi li ngual learner groups; Early learners and Late learners. There are four chi ldren in each group. The Early learners' first exposure to Swedish was between the age of 9 and 13 months. Thei r acquisition of Swedish was followed over a period of two years. The data collection varied somewhat between the children. The youngest child was 1:4 and the oldest 2:7 at the start of the recordings. The Late learners started to learn Swedish well after the age of 3. They were also recorded during a period of two years and the ages covered in this group are 5 to 6 years. The chi ldren were audio-recorded in the nursery/pre-school setting and the data consist of natural spoken discourse in everyday activities.

The children used a total of 392 relative constructi ons; the Early learners used 214 altogether (86 NP[rel], 91 cleft constructions and 37 relati ves embedded in a SVO matrix) and the Late learners produced 178 relati ves (57 NP[rel], 79 clefts and 42 embedded in a SVO matrix).

2.2 Analysis

Thi s paper addresses the i nteracti on between syntactic and semantic features of relatives in spoken discourse. Specifically I will focus on semantic features of the Head NP, and transitivity of the predicate in the relati ve clause. For the present purpose only the following codings will be discussed.

9

HUMANNES and ANIMACY of the Head NP. Non-human heads have thus been coded for animacy, to single out the difference between non-human ani mates and i nani mates.

Depending on the transitivity of the verb, the relati ve has been coded as follows, using Dixon's case roles (1979):

1) S-RELATIVE: a subject relati ve with an i ntransi ti ve predi cate

e.g. 'som st?r d?r' ('that is standing there')

2) A-RELATIVE: a subject relative with a transitive predicate 'som bakar kakor' ('that is baking cookies')

3) P-RELATIVE:an object relati ve. The object could serve as a direct object or as an object of preposi ti on, e.g. 'som jag har gjort' ('that I have done'), 'som man kan leka med' ('that you can play with').

2.3 Results

First I will present the Early and the Late bi li ngual learners separately. We will then compare thei r use of relati ves wi th data from a monolingual speaker of Swedish.

The Early learners produced 214 relati ves altogether. The distribution of relative structures in different li ngui sti c contexts according to semanti c feature of the Head NP is presented i n Table 1.

(Insert table 1 here)

There is a striking regularity in these data. Thi s regularity is obtained by i ntroduci ng two variables not previ ously used i n acquisition studies of relative clauses, namely semanti c features of the Head NP and transi ti vi ty of the predicate in the relati ve clause.

S-relatives are used for specification of human and non-human heads alike, i.e. there is no restri cti on of the di stri buti on of i ntransi ti ve subject relati ves with respect to the semanti c

10

features of the Head NP. The di stri buti on of P- and A-relatives

i s, however, severely constrai ned.

Table 1.EARLY LEARNERS

Distribution of S-, A- and P-relatives in different linguistic contexts according to semantic feature of the Head NP.

_________________________________________________ SEMANTIC FEATURE +HUMAN -HUMAN OF THE HEAD NP

S-rel A-rel P-rel S-rel A-rel P-rel NP[rel] 9 9019 149

CLEFTS2322018 324

SVO-matrix[rel] 3 2013 118

TOTAL3533050 591

_________________________________________________ P-relati ves are never used for specification of human heads i n

early child discourse. None of the Early learners used relatives

like 'I met a guy [that you know]' whi ch would be a perfectly natural utterance in adult Swedish discourse. Every P-relative

in child discourse is used for specification of non-human heads. There is a very strong tendency to use A-relati ves to specify human referents (33 of 38 A-relati ves altogether). In the Early learners' producti on only fi ve non-human Heads are specified

by an A-relative. Of these 4 were animate heads, like in (10).

(10)Det ?r en elefant [som g?r i bajset]

11

'It's an elephant that is walking in the poo-poo'

The only exception was the following

(11)Adult: en dusch ?r det

'It's a shower'

Child: n?

vatten

[som dricker (PAUS) vatten]

'no, water [that drinks (PAUSE) water]'

An alternative interpretation of (11) is

(11')vatten

[som dricker]

vatten

i.e. a subjectless P-relati ve 'som (man) dri cker' ('that (you) dri nk') were the topic 'water' is repeated after the specification in the relative clause. If this interpretation is accepted, there are no exceptions to the strategy of using A-relati ves to specify ani mate heads only.

So, in early child discourse human referents are specified by subject relatives (S- and A-relati ves). Non-human referents are preferably specified by S- and P-relatives. The few exception to this general strategy consist of ani mate referents being specified by an A-relati ve. Inani mate referents are specified by ei ther S-relati ves or P-relati ves.

We wi ll now turn to the Late learners, Table 2. Taken together they used a total of 178 relatives.

Table http://www.wendangku.net/doc/12a13f70f242336c1eb95e89.htmlTE LEARNERS

12

Distribution of S-, A- and P-relatives in different

linguistic contexts according to semantic feature of

the Head NP .

____________________________________________

SEMANTIC FEATURE +HUMAN-HUMAN OF THE HEAD NP

S-rel A-rel P-rel S-rel A-rel P-rel

NP[rel]10 3013 229

CLEFTS16100221318

SVO-matrix[rel] 2 2016 319

TOTAL28150511866

____________________________________________ As the Early learners, the Late learners have a total restriction on the di stri buti on of P-relatives. No human referent i s specified by a P-relative, i.e. an object relative, which is reserved for non-human referents. The Late learners produce a fairly large number of A-relati ves together with non-human heads. Most of these are used to specify animate heads, (15/18).

It seems to me quite plausible to assume that chi ldren have the following relativization strategy in spoken discourse

?If the referent of the Head NP is hu man, specify it by an A-relative or a S-relative.

?If the referent of the Head NP is non-hu man, specify it by a P-relative or a S-relative.

You mi ght suspect that this strategy would be specific to bi li ngual chi ldren, acquiring Swedish as a second language. In order to test this possibility, the relati ves of a monolingual Swedish speaking child, Markus (Str?mqvi st's Corpus (1992)), have been analysed with respect to semanti c features of the Head NP and distribution of S-, A- and P-relatives. 129 relatives

13

have been analysed0. Markus' relati ves are presented in Table 3.

Table 3.L1-learner

Distribution of S-, A- and P-relatives according to

semantic features of the Head NP, irrespective of

li ngui sti c context.

____________________________________________

SEMANTIC FEATURE +HUMAN -HUMAN OF THE HEAD NP

S-rel A-rel P-rel S-rel A-rel P-rel

Total 1511062 635

____________________________________________ As the bi li ngual chi ldren, Markus does not specify Human referents with a P-relati ve. P-relati ves are used for the specification of non-human heads only. Of the 6 non-human heads, specified with an A-subject, four were ani mates. There were two i nani mate heads specified by an A-subject like in the followi ng example.

(12) Markus: e mafin

i s machi ne

Mother: e de en maskin d?r ja de e de

is this a machine there yes it is

Markus:[som klippe gj?set]

that mows the lawn

Mother:tror du den g?r de PAUS en gr?sklippare

You think it does PAUSE a lawn mower0

0 Data on Markus' relatives has been presented in Andersson & Richthoff (1990, Appendix). The database consists of 135 relatives. 6 of them has been discounted in this analysis. The reason for this is that they were either repetitions or not headed.

14

Markus' production of relati ves was recorded between 1;11 and

3;10, i.e. he and the Early learners are in the same age. The Late learners were 5 to 6, when recorded. The Early Learners adhered

to the strategy of not specifying Non-humans with A-subjects to 98%, The Late learners to 87%. Markus comes in between with 94%. Being, a monoli ngual speaker of Swedish, he can be expected to be a little ahead of his peers. There is, however, n o reason to believe that the relati vi zati on strategies accounted for should be something specific to second language acquisition.

The analysis of the different learners' producti on of relati ves i s summarised in Table 4.

Table 4.Distribution of S-, A- and Prelatives according to semantic features of the Head NP, irrespective of linguistic

context.First and second language learners.

________________________________________________ SEMANTIC +HUMAN -HUMAN FEATURES OF

THE HEAD NP

Total No. S-rel A-rel P-rel No.S-rel A-rel P-rel No. EARLY

LEARNERS (214) 35 33 0(68) 50 591(146) LATE

LEARNERS (178) 28 15 0(43) 51 18 66(135) L1-LEARNER (129) 15 11 0(26) 62 6 35(103) TOTAL (521) 7859 0(137)163 29192(384)

________________________________________________

0This is in fact a very typical example of another function of relatives in early

child discourse, namely to use a relative instead of the lexicalized compound

in adult vocabulary (Hamburger, 1980).

15

The following hierarchy captures these children's relati vi zati on strategies:

S-rel > P-rel > A-rel

241 192 88

1.S-relati ves are the most frequent. There is no restri cti on o n

thei r di stri buti on depending on semanti c features of the Head NP.

2.P-relati ves are also frequent. Thei r di stri buti on is restricted

to Non-human heads, speci fi cally Inani mates.

3.A-relati ves are the least frequent. Wi th very few exceptions

their distribution is restricted to Animates and specifically to Human heads.

3. Discussion

How shall we account for these fi ndi ngs? As pointed out earlier, chi ldren's acquisition of relati ves has been explained mai nly from two different perspectives, namely with reference to syntactic or to discourse strategies. I will argue for an integrated view of language acquisition whi ch consi ders different linguistic as well as language i ndependent variables to be i mportant determi nants of chi ldren's language acquisition (Berman (1984). I will argue that an analysis that takes into account the i nteracti on of pragmatic, semanti c as well as syntactic variables has much to offer in the description of chi ldren's relati ve grammar.

The hierarchy presented above cuts across the Accessibility Hierarchy (AH) formulated on cross-linguistic grounds by Keenan & Comrie (1978):

SU > DO > IO > OBL > GEN > COMP

The AH has been claimed to be uni versally valid. It has also been proposed as an explanati on for psycho-linguistic data

16

(Keenan & Hawkins, 1986). It is a common finding in child data that subject focus is the most frequent relati vi zati on strategy, whi ch have lent support to the psycho-linguistic relevance of the AH. And indeed, if you treat transi ti ve and intransitive subjects as one category 'subject' it would also account for my data. However, i mportant generali sati ons on chi ldren's relativization strategies would then be missed.

In a study of relative clauses in adult spoken English discourse, Barbara Fox (1987) found S-relati ves and P-relati ves to be far more frequent (90%) than A-relati ves (10%). She also pointed out that a reformulati on of the AH along these li nes, could solve the problem that West-Austronesi an ergative languages, like Tagalog and Batak, pose for the AH. In Batak and Tagalog relativization is possible on intransitive subjects and on objects, but not on transitive subjects. Fox argues for a reformulati on of the leftside of the AH, i.e. that the category of 'subject' be split into S- and A-subjects in order to capture these facts (frequency in adult discourse and restri cti ons on relati vi zati on in ergati ve languages):

Absolutive0 (S-rel and P-rel) > A-rel

The 'Absoluti ve Hypothesi s' states:

"that a langu age mu st be able to relativize on S and P, if

it has a strategy for relativization at all. The reasons for

this constraint on relativization were fou nd to arise

from constraints on conversationally appropriate

strategies for introdu cing referents into the discou rse.

In particu lar, the assu mption that 'su bjects' (S and A)

are most easily relativized-on becau se of a special

cognitive statu s is challenged on two cou nts: (a) it seems

to be the category ABSOLUTIVE, rather than SUBJECT,

which occu pies the leftmost position on the accessibility 0The category 'absolutive' consists of intransitive subjects and transitive direct objects. In ergative languages, 'absolutive' constrasts with the category

'ergative', marking transitive subjects.

17

hierarchy; and (b) it is the discou rse fu nction of S- a n d

P-relatives, rather than a special cognitive statu s, w h i c h

gives them the prominence in langu ages across t h e

world."

Fox, 1987, p. 869.

Di scourse-level explanati ons could help to explain different aspects of the distribution of relatives in chi ldren's producti on. For all chi ldren, relati vi zati on on Non-human heads is by far more frequent (74%) than on Human heads (26%).

Thi s is an i nteresti ng finding. In studies of children's producti on of relati ve clauses it has been found that object relati ves (OS- and OO-structures) are the most frequent in all age groups, and that they are the first ones to appear (Limber, 1973, Menyuk, 1971). From a structural point of vi ew, it seems as if it is easier for chi ldren to add a clause at the end of a sentence rather than in the middle, since this mi ni mi ses constrai nts on processing, the Interrupti on hypothesi s formulated by Slobin (1973).

Also in some of the comprehensi on studies, object embedding seemed to cause chi ldren less problem than subject embedding (Hakuta, De Vi lli ers & Tager Flusberg, 1982; Hamburger & Crain, 1982 ). The surface ordering of major consti tuents (not their grammatical roles) is taken by children to signal the major grammatical relations in the sentence, referred to as the N-V-N-strategy (Hakuta, De Vi lli ers & Tager Flusberg, 1982). This strategy captures the fact that chi ldren tend to parse surface strings of nouns and verbs as SVO-structures, reflecting what Slobi n (1985) refers to as 'the prototypical scene'. These processing constraints might very well explain the way children cope with on-li ne-processi ng of central embedded relati ves. In producti on the i nserti on of a pronomi nal copy after the embedded clause in the mai n sentence, as in example (13), could be seen as a reflection of this strategy.

(13) dom [som har i ngenti ng] dom ?r mamma

'those [who have nothi ng] they are mummy

18

However, these principles could hardly explain the constraints on chi ldren's structural choices in spoken discourse. It is also a fact that chi ldren produce central-embedded structures fairly early on (2;2 in the Str?mqvist Corpus, 3;2 in my own data).

In a discourse perspective it is the communi cati ve functi on of relati ves as comment specification that has been stressed (e.g. Romaine, 1984, Plunkett & Str?mqvist, 1992). The great number of object relati ves should mi rror the communi cati ve need to specify new i nformati on, whi ch conforms to a more general pattern (in languages like English and Swedish) where new i nformati on tends to come at the end of an utterance (Gi vón, 1979).

Persons we talk about tend to be known from mutual knowledge (specifically in early child discourse) and they are comparatively few. Things we talk about, on the other hand, are less li mi ted and they tend not to be known beforehand. Consequently, non-human enti ti es need to be specified far more often than human ones.0Thi s holds true also for adult discourse, where 65% of all relati ves had non-human heads (Fox, 1987). Fox also found that in spoken discourse in general, the relati on subject:object is 3:1, whereas in relati ves thi s relation is 1:1.

Like adults, chi ldren use different li ngui sti c means to specify persons and things. Disregarding the unrestricted di stri buti on of the S-relati ve for the moment, it is a fact that Human referents are specified by A-relati ves, whi le Non-humans pri mari ly are specified by P-relatives. The i nteresti ng fact here is not that non-humans don't get specified by an A-relati ve, i.e. by thei r actions. It's quite natural that i nani mate objects don't perform actions to any larger extent. The exception to this 0Another reason of the high frequency of relatives that specify non-human heads is the possiblity to refer to things whose lexicalized expression you don't know. This is indeed a very efficient way for language learners to enlarge a restricted vocabulary.

19

general fact of the world are different kinds of machi nes li ke: 'this damn compu ter drives me crazy'.0

What needs to be explained, however, is the fact that children, in contrast to adults, never specify Humans with a P-relative. Pre-school chi ldren never say thi ngs li ke 'The boy [that I talked to yesterday] comes from Somalia'0 or It's a boy [that I know]. Where do chi ldren get this restri cti on from? It mi ght be a strong tendency in adult discourse to specify human referents by A- and S-relati ves, but it is also a fact that P-relati ves could do the job. One plausible explanati on would be that this tendency in the li ngui sti c i nput by underextensi on may be taken to be an absolute constrai nt on child grammar0. The reason for handli ng P-relati ves in this way may be found i n chi ldren's perception and categorisation of the non-li ngui sti c context. I thi nk we have good reasons to beli eve that prior cogni ti ve development, still at this stage, has a role to play i n the way children's grammar is shaped.

Consequently, I don't agree with Fox's conclusi on, namely that 'it is the discou rse fu nction, not cognitive reasons that can explain relativization strategies in the world's langu ages'. You might see it the other way around; that it is the cogni ti ve status of what may serve as Agents and Patients that i nfluence the way discourse is structured.

As human beings we interact with people and act upon physical objects in the real world. It is plausible to assume that our perception and categorisation of the world will i nfluence the way we (learn to) talk about persons, typical Agents, and about things, typical Patients. These cognitive concepts are established on language i ndependent grounds during the sensori-motor 0Inanimate objects get 'personified' (Lyons, 1977).

0This example is an authentic utterance of an adult, talking to one of the Early learners.

0Overextension has been a common explanation of premature linguistic rules in child language (Clark,1973), whereas the importance of underextension may have been overlooked (but see Griffiths, 1986).

20

stage in the chi ld's cogni ti ve development (Piaget, 1954), but may still in later stages of language development restrict the way chi ldren organise coherent and cohesi ve text (Bowerman, 1982, Stevenson & Sims, 1993).

The conceptual i mportance of 'Agent' and Pati ent' is supported by recent studies on the acquisition of passives (Lempert, 1990) and verb transi ti vi ty (Ingham, 1993) . Cogni ti ve i nput in the physical si tuati on, rather than syntactic i nput, has been shown to have a critical i nfluence on chi ldren's acquisition of transitivity. Saliency in the physical situation is also taken to be the explanati on of chi ldren's use of definite expressi ons (Stevenson & Sims, 1993). Children between 3 and 8 years were tested and compared to adults. Stevenson & Si ms found that chi ldren only describe features of a referent that are sali ent to them. They fail to use non-sali ent features of the referent that are discriminatory, and persist in using sali ent features that are redundant.

In child discourse, relati ves, like other definite expressions, are primarily used to specify referents according to sali ent features in the physical si tuati on. The perspective of the 'Agent', typically a human, has been shown to constrai n chi ldren's structural choices, whi ch lends support to the Perspective hypothesis (MacWinney, 1977). That P-relati ves are not used to specify human referents could be seen as a reflection not only of the conceptual primacy of the 'Agent', but also as a consequence of the prototypical 'Pati ent'. Chi ldren's relati vi zati on strategi es could be summarised as follows.

?If you talk abou t persons, specify them by their actions or by the way you perceive them.

?If you talk abou t physical objects, specify them by the way people u se them or perceive them.

4. Summary and conclusion

21