Null Subject vs. Null Object: Some Evidence from the ... · Null Subjectvs. Null Object: Some Evidence from the Acquisition ofChinese and English* Qi Wang,t Diane Lillo-Martin,tCatherineT.

Haskins Laboratories Status Report on Speech Research1992, SR-109 / 110, 225-250

Null Subject vs. Null Object: Some Evidence from theAcquisition of Chinese and English*

Qi Wang,t Diane Lillo-Martin,t Catherine T. Best,tt and Andrea Levittttt

Since young English-speaking children use null subjects systematically, it has beenproposed that they begin with the initial parameter setting allowing null arguments(NAs), and must change this setting on the basis of linguistic evidence that adult Englishprohibits NAs. A recent proposal suggests that the licensing and identification ofNAs usedby English-speaking children is like that used in adult Chinese. This predicts that youngChinese- and English-speaking children should exhibit parallel perfonnance in their use ofNAs. This study investigated this prediction using an elicited production task with bothChinese- and English-speaking children. Although the hypothesis that early Englishallows null subjects was upheld, the evidence is against the claim that early English is adiscourse-oriented language like Chinese: while the Chinese children systematically usednull objects, the American children did not. An alternative analysis of the use of nullarguments is suggested.

In the examples in (la) the subject, though notphonologically specified, has a definite referencewhich can be readily inferred from context. Sincesentences with null subjects like those in (la) co­occur with sentences like those in (lb), which dohave lexical subjects, it is not likely that themissing subjects in (1a) can be attributed to aperformance constraint on sentence length. Afurther characteristic of children's speech at thisage is illustrated by the examples in (2a). In theseexamples the unexpressed subject is an expletive,as shown by the 'translations' of these sentencesin (2b). However, according to Hyams, children atthis age do not produce sentences such as (2b).

Additional studies of children's early use ofsubjectless sentences are found with bothlanguages which do allow null subjects and thosewhich do not, such as Italian (Hyams, 1986),

(1) b. Kathryn read thisGia ride bikeI want take this off

1. INTRODUCTION

1.1 The Null Subject Phenomenon in EarlyChild Language

The null subject phenomenon, i.e., the frequentabsence of lexical subjects, is one of the mostnoticeable characteristics of early child language.The following (non-imperative) English sentences(la) and (2a), spoken by children aged from 1;8 to2;5 (cited by Hyams, 1983), are examples of thisphenomenon.

This study was supported in part by Nlli grants DC-00403to Dr. Catherine Best and DCOOO183 to Dr. Diane Lillo­Martin, both at Haskins Laboratories.

We would like to thank the Chinese children and theirfamilies, and the teachers and the American children at theChild Development Lab of the University of Connecticut, fortheir help and participation in this experiment. Without them,it would have heen impossible to have accomplished this study.We are grateful to the professors and students in LinguisticsDepartment at the University of Connecticut who have givenus valuable comments at various times. We would also like tothank Mr. Philip Chagnon at Haskins Laboratories, whohelped us greatly in making the wood dolls for the experiment.We would also like to express our gratitude to the reviewers fortheir helpful comments and suggestions.

225

(1) a. Read bear bookRide truckWant look a man

(2) a. Outside cold (2) b.No morningYes, is toys in there

('It's cold outside')('It's not morning')('Yes, there are toysin there')

226 Wangetal.

German (Clahsen, 1989; Weissenborn, in press),French (Weissenborn, in press), and AmericanSign Language (Lillo-Martin, 1986, 1991). In all ofthese studies, it has been found that at an earlyage children use subjectless sentences like theones illustrated in English above.

The search for an explanation of children's earlyuse of subjectless sentences can be related tostudies of adult languages which permit suchsentences as grammatically acceptable, bycomparison to those which do not. In the nextsection, we review some characteristics of the nullsubject phenomenon in adult languages (since weinclude null objects as well as null subjects, theterm has been generalized to 'null arguments'),and one proposal for the grammatical mechanismsunderlying this phenomenon. We will then tum toa proposal accounting for children's use of nullsubject sentences which appeals to this analysis ofadult language.

(3) a. Mangia come una bestia.

'(He/she) eats like a beast.'

b. Come como una bestia.

'(He/she) eats like a beast.'c. [e] 1m-Ie.

come-ASPl

'(He/she) came.'

(4) a. Sembra che Gianni sia matto.

'(It) seems that John is crazy.'

b. Piove oggi.'(It) rains today.'

(5) a. [e] Xiayii-le.

(It) rain-ASP

'(It) is raining.'

b. [e] Yao xiayii-le.

(It) going to rain-ASP

'(It) is going to rain.'

1.2 The Null Argument Phenomenon inAdult Languages

The null argument phenomenon is a well-knowncharacteristic of adult languages such as Spanish,Italian and Chinese. Examples from theselanguages are given in (3). The Englishcounterparts to these sentences require overtsubjects.

In these so-called 'pro-drop' languages, the ex­pletive elements equivalent to English it andthere are also phonologically null, as illustratedin (4) (Italian, from Hyams, 1983), and (5)(Chinese). 2

In adult Chinese, the expletive elementequivalent to English it can be phonologically nullas in Spanish or Italian, as illustrated above (Sa,b, c).3 Alternatively, a non-expletive subject can befound in any of these sentence types, illustrated in(6a, b, c).

(Italian; Hyams, 1983)

(Spanish; Hyams, 1986)

(Chinese; Huang, 1982)

c. [ei] Kimshangqu [ej] yao xiayii-le.

(It) seem (it) going to rain-ASP

'(It) seems that (it) is going to rain.'

Null Subject tIS. Null Object: Some Evidencefrom the Acquisition ofChinese and English

(6) a. Tiim xiayii-le.

sky rain-ASP

Lit., 'The sky is raining.'

b. Tiim yao xiayii-le.

sky going to rain-ASP

Lit., 'The sky is going to rain.'

c. Tiani kanshangqu rei] yao xiayii-le.

sky seem going to rain-ASPLit., 'The sky seems to be going to rain.'

227

How can one account for the occurrence of nullarguments in these languages, compared tolanguages which prohibit null arguments, such asEnglish? Jaeggli and Safir (1989) proposed thefollowing Null Subject Parameter, stated in (7), asa principle of Universal Grammar (UG) to makethis distinction.

(7) The Null Subject ParameterNull subjects are permitted in all and onlylanguages with morphologically uniforminflectional paradigms.(Jaeggli and Safir, 1989, p. 29).

According to Jaeggli and Safir, a morphologicalparadigm is uniform if all its forms aremorphologically complex or none of them are. Forexample, the Italian inflectional paradigmconsists entirely of morphologically complexforms, hence null subjects are allowed; in Chinese,no forms are morphologically complex, hence nullsubjects are allowed here too. In the case ofEnglish, however, morphologically complex formssuch as walks, walked, walking, coexist withmorphologically simple forms, such as walk. ThusEnglish is a 'mixed' system and null subjects areprohibited.

The Null Subject Parameter stated in (7) tells uswhen a null subject is possible. However, Jaeggliand Safir (following others such as Rizzi, 1986)also propose that a null subject can occur onlywhen its referential value can be recovered. Theypropose three mechanisms for the identification ofnull arguments: (i) local AG(reement) including atense feature, (ii) a c-commanding nominal, or (iii)a Topic. Failure to satisfy either of the twonecessary and sufficient conditions, namely, a

morphologically uniform paradigm and arecoverable referential value for the thematic nullsubject, will result in the prohibition of nullsubjects in a language. Although the use of nullarguments thus requires two conditions to be met,for ease of exposition we will refer to a NullSubject (or Argument) parameter with settings[+/-pro-dropl. (This also enables us to be neutralwith respect to other analyses of the nullargument phenomenon.)

The use oflocal AG to identify the reference of anull argument follows from numerous reports inthe literature linking null arguments with 'rich'agreement. Early reports were confined tolanguages with only subject-verb agreement (suchas Italian, discussed in Rizzi, 1982); theselanguages allow null arguments identified byagreement only in subject position. Later studies(such as McCloskey and Hale's 1984 work onIrish) have demonstrated that languages withother types of agreement often display nullarguments in other positions. Jaeggli and Safiradd the condition that a tense feature must bepresent in order to account for the lack of nullarguments in German and other V2 (verb-second)languages. The null arguments which areidentified by AG are considered to be members ofthe empty category pro, [+pronominal,-anaphoricl.

The use of a Topic to identify null subjectsfollows from Huang's (1984; 1989) proposal.Huang distinguishes "discourse-oriented"languages from "sentence-oriented" languages.The "discourse-oriented" languages, like Chinese,have a rule of "topic-chaining" by which thediscourse topic is grammatically linked to a nullsentence topic which in turn identifies a null

228 Wangetal.

argument. This null argument is a variable leftfrom the movement of the empty topic to sentence­topic position. According to Huang, a topic maybind a variable in either subject or object position.These two kinds of null arguments are illustratedin (8).

In addition, there is a third method of identify­ing null arguments which results in a sub­ject/object asymmetry. Because a c-commandingNP can also be an identifier, in languages likeChinese a null pronominal (pro) may be found inembedded subject position, as in (9a), but not inobject position, as in (9b). This distinction is found

because the empty embedded subject can be iden­tified by the matrix subject; it functions grammat­ically like a pronominal rather than a variable.However, the empty object cannot be identified bythe matrix subject, since identification has to beby the closest nominal element.• Thus, empty ob­jects can only be identified by an empty topic, in­dicated by OP in (10).

To summarize, Jaeggli and 8afir proposed thatthe difference betweeJl the grammar of pro droplanguages such as Italian versus those such asChinese is the method of identification of the nullargument. This is illustrated in (11).

(8) a. Discourse Topicj [s'topici [s [ei] !NFL liri-Ie ]]

come-ASP'(He) came.'(Huang, 1984)

b. Discourse Topicj ~'topiCi [s wo INFL [ mei kAnjiim [ ei]] ]]I not see (himi)

'I did not see (him).'

(9) a. Zhangsani, tai shu o[ei] mei kanjian LIsi (Huang, 1989)

Zhangsan he say no see Lisi'Zhangsani, hej said that (hei) didn't see Lisi.'

b. *Zhangsani, tai shuo Llsi mei kAnjiim [ei]

Zhangsan he say Lisi no see'Zhangsani, hej said that Lisi didn't see (himi).'

(10) [ OPj [ Zhangsanj shiio [ Llsik kanjian [ej] Ie ]]]

Zhangsan say Lisi see ASP'Zhangsanj said that Lisik saw himi/*j/*k.'

(11) a. [s prOi [INFL AGyTense] ...... ]

(identification by AG, Italian)

b. Discourse Topicj [topicj [s tj [INFL] ...... ]

(identification by a discourse topic, Chinese)

c. Subjectj verb [s prOj VP]

(identification by a c-commanding NP, Chinese)

Null Subject 'V5. Null Object: Some Evidencefrom the Acquisition ofChinese and English 229

1.3 Null Subjects in Children's Grammars

From the above, it may be seen that 'Early'English resembles a pro-drop language in threerespects. First, lexical subjects are optional;second, the subject has definite reference evenwhen phonologically null (except in the case ofnull expletives); and third, lexical expletives areabsent (Hyams, 1983; 1989).

How can one account for the development anEnglish-learning child has to undergo in order toarrive ultimately at a steady state grammar so asto speak the right type of English? A recentanalysis by Hyams (in press; Jaeggli & Hyams,1987), following the analysis of null subjects inadult languages by Jaeggli and Safir (1989)discussed above, proposed that the earlygrammar, like adult grammars, is constrained bythe Null Subject Parameter cited above. That is,the early grammar satisfies the requirement ofmorphological uniformity and the requirementthat null arguments be properly identified.

Hyams argues that English-speaking childrenbegin speaking a Chinese-like language, i.e., adiscourse-oriented language. Under the child'sinitial analysis, English is morphologicallyuniform with uniformly simple forms. Hyamstakes children's verb productions, which at thistime are generally not inflected, as evidence forthis position. She further proposes that youngEnglish-speaking children use null topics toidentify the reference of their null subjects. Thechild will then need to learn that English is not a'Discourse Oriented' language in order to properlyexclude null subjects.

In the case ofItalian-speaking children, Hyamsproposes that their early empty subjects areidentified by AG(reement), as is the case in adultItalian. She proposes this early correct nullsubject use since Italian speaking children acquirethe inflectional system fairly early. Thus, for thesechildren resetting of the null subject parameter isnot required.

One potential problem for Hyams' analysis isthat one would expect that a discourse-orientedchild language should have both null subjects andnull objects, since under topic identification thenull subject and null object phenomena aregrammatically equivalent. However, according tothe data she reviewed, Hyams claimed thatEnglish-speaking children do not use null objects.In order to account for this, Hyams thus proposed,following Roeper, Rooth, Mallis, and Akiyama(1984),5 that in the early grammar, the inventoryof null elements includes pro, but not variables.

This hypothesis would predict a null subject/nullobject asymmetry. Since null objects can only bevariables, under this hypothesis null objects wouldnot be allowed in the early grammar until somelater point when variables mature. In order forthis account to hold, Hyams must depart fromHuang's analyses of Chinese, and suggest thatmatrix empty subjects as well as embedded emptysubjects can be pro, although only embeddedempty subjects can be identified by a c-command­ing NP. Hyams says that matrix empty subjectpros are identified by a discourse topic.

According to Hyams' hypothesis, Chinese-speak­ing children, who will ultimately acquire a realdiscourse-oriented language, should first exhibitthe same null subject/null object asymmetry asEnglish-speaking children, and they should notproduce null object structures until the pointwhen they develop variables. Hyams' hypothesiswould also predict one of two null subject-objectasymmetries for English-speaking children. Onthe one hand, if they have not yet reset the NullSubject Parameter by the time that they acquirevariables, then they will produce only null sub­jects early on, but will later include null objects aswell once they have developed variables. On theother hand, if the English-speaking children havereset the null subject parameter before they de­velop variables, they will never use null objects.Thus, knowing when English- and Chinese-learn­ing children use null subjects and objects com­pared to when they develop variables is importantfor evaluating Hyams' proposal.

The evidence regarding the timing of use ofvariables versus resetting the null subject param­eter is not wholly consistent with Hyams' ap­proach. Roeper (1986) gives evidence that childrenhave some uses of variables by age three to fouryears. All of his evidence for the use ofpros ratherthan variables with wh-questions occurs witholder children (ages 8 to 10) and long-distancequestions. However, his proposal that children usepros instead of variables even at this later age canalso be questioned, given new evidence regardingchildren's very early comprehension and produc­tion of wh-questions and strong crossover con­structions (see Thornton, 1990). We therefore usedthe production and comprehension of wh-questionsin the study reported here as evidence for theexistence ofvariables in children's grammars.

The timing of the use of null subjects is easier todetermine. The acquisition data Hyams used tosupport her hypothesis indicate that therestructuring of the Null Subject Parameter takesplace around 26 to 28 months. If Hyams' proposal

230 Wang et al.

that young children do not have variables is true,then we will not expect to see any null objects inthe production of English-speaking children, sincethe restructuring takes place prior to thedevelopment of variables; and of course a cleardecline in their use of null subjects should appearfollowing the resetting of the NA parameteraround 2-112 years. However, if there is evidencethat children do have variables while they still usenull subjects (indicating that the resetting of theNA parameter has not yet taken place), then theywill be expected to use null objects too, accordingto Hyams' account.

In order to more fully evaluate Jaeggli andHyams' proposals, we collected data on theacquisition of English and Chinese. The followingexperimer ' was designed to answer some relevantquestions ...t>out Hyams' hypothesis through first­hand acquisition data. The questions weaddressed include the following:

i. Is a null subject/null object asymmetryexhibited in child Chinese and child English? If so,is it equivalent for the two groups?

ii. If child Chinese or child English does exhibitnull objects, do we have evidence that variablescoexist with null objects? The emergence of wh­questions will be taken as evidence of acquisitionof variables.

iii. Can the presence of lexical expletives betaken by American children as evidence thatEnglish is not [+pro-dropl? The use of overt versusnull expletives will be examined to address thisquestion.

iv. What does the developmental pattern looklike, as far as the null subject and null objectphenomena are concerned, in terms of theparameterized theory of UG?

v. What is the influence of linguisticenvironment during development of earlygrammar between ages 2 - 4-112?

2. Method2.1 Subjects

2.1.1 Chinese and American children. NineChinese children, 4 female and 5 male, aged from2;0 to 4;6, participated in the experiment. All ofthem were learning some variety of MandarinChinese as their first language. Their parentswere graduate students from either mainlandChina or Taiwan, studying in the United States.Nine English-speaking children, 5 female and 4male, aged from 2;5 to 4;5, were also tested usingthe same procedure. Their parents were membersof the University community. All the subjects had·normal hearing. There were no recorded

developmental delays of any sort. Subjectcharacteristics are given in Appendix 1.

2.1.2 Chinese adult controls. Nine Chinese­speaking female adults participated in theexperiment. They were all born in mainlandChina or Taiwan, speaking some variety ofMandarin Chinese. They were the mothers of theChinese child subjects.2.2 Procedure

2.2.1 Controlled production data collection. Thispart of the experiment was carried out in theexperimenter's home for the Chinese children, andin the observation room at a day care center forthe English-speaking children. There were twostory books used. One was a story book designedby the experimenter (QW) about the daily life of alittle boy named Baldy (who had no hair). A dollhouse with dolls and furniture corresponding tothe settings and characters in the book was usedto familiarize the subject with the main character.Another story used was a pop-up book, "The ThreeLittle Pigs." The testing was carried out after theexperimenter played with the child subject anumber of times and established rapport. Thesubject's task was to tell the experimenter thestory. For the first story, the experimenter and thesubject played with the doll house and dolls. Next,the subject was asked if he or she wanted to reada book about Baldy and then to tell a story abouthim. The answer was invariably positive. Theentire procedure was audio recorded. Allinteraction with the Chinese-speaking childrenwas conducted in Mandarin; that with theEnglish-speaking children was in English.6

2.2.2 Eliciting expletive structures. In this partof the experiment, a number of pictures weredisplayed to the child subject and then he or shewas asked to tell what happened in the pictures.This part of the experiment was designed to elicitexpletive structures for the English-speakingchildren and to compare their productions to thoseproduced by the Chinese-speaking children underthe same situation.

2.2.3 Adult controls. The Chinese adult subjectswere asked to tell the stories and talk about thepictures, while pretending that they were talkingto their own child. The testing was conducted inthe subjects' home without their child or theexperimenter present. The testing materials wereidentical to those prepared for the child subjects.The whole procedure was audio taped.2.3 Data reduction

i. The mean percentage of sentences with nullsubjects for each speaker was calculated based onthe ratios of the sentences with null subjects to

Null Subject t'S. Null Object: Some Evidence from the Acquisition ofChinese and English 231

wash bath

'take a bath'(12) b. xl shOll

wash hands

'wash hands'

the total number of sentences produced whentelling the two stories. These ratios were averagedover the total number of subjects in each languagegroup, over each age level (2-, 3-, and 4-year oIds),and over each MLU level (3.5, 4.5, 5.25)separately. The standard error of the means (s.e.)was also calculated.7

ii. The mean percentage of sentences with nullobjects was calculated using a similar method.The ratio was the total number of sentences withan underlying structure of SVO to the total num­ber of sentences produced with a null object. Forthe Chinese data, in addition to this criterion, anytwo-morpheme compounds which have been iden­tified as a word by the authoritative dictionary­XiimdAi HAnyU' Cidiin (Modern ChineseDictionary) (Institute of Linguistics, ChineseAcademy of Sciences, 1973)-were not included,even if they had the V+O formation. For example,(12a) was identified as a single word, so it was ex·eluded; but (12b) was counted because it was notidentified as a single word. The reason for thisconstraint is that it is generally agreed amongChinese linguists that a verb+complement com­pound is not equal to the structure ofV+O; unlikethe latter, the former is already in its minimalconstruction and is not divisible; therefore, thesetwo types of words are analyzed differently.

(12) a. xl v

zao

lll. The MLU for child subjects in bothlanguages was calculated, using the productionsmade for the stories, according to the method inBrown (1973).

iv. A second measure of the mean percentageof sentences with null subjects for English­speaking children was also calculated in the sameway, excluding the sentences with null subjectsusing a gerund or to-infinitive. The reason for thisexclusion is that given the discourse, these kindsof sentences are also allowed in the adultgrammar of English. This second measure islabelled 'aQiusted' in the figures.

v. The data gathered from testing theexpletive structures was excluded from thecalculation of the mean percentages. This part ofthe data was only evaluated for structuraldifferences among the three testing populations.No quantitative analysis was involved.

vi. The children's comprehension andspontaneous productions of wh-questions duringthe course of the study were evaluated, for thepurpose of determining their use of variables.

3. Results

3.1 An Overall View of the Results (fordetails see Appendices 2 and 3)

3.1.1 Null subjects. From Figure 1, it may beseen that there is a noticeable difference betweenthe mean percentages of sentences with nullsubjects produced by Chinese child subjects andthat by American child subjects at 2 - 4-1/2 years.Examples for such sentences are (l3a,b) for theChinese child subjects, and (14a,b,c) for Americanchild subjects.

fa Chi_ childrMI

100 • Am.nc.n childr.... (UMdjuet.cl)

l lSI Am.nc.n childr.... (lIdju.t.d)

1i 80 !'J Chi_lIdult2iII:;

80c'8

il 40c

I20

oFigure 1. Mean percentage of sentences with nun subjects produced by Chinese and American children and Chineseadults.

232 Wangetal.

[e] shuai. [e] shuai dao Ie.

fall fall down ASP(He) fell. (He) fell down.'

Zhe hUang wawa tiaotiao.

this yellow baby jump'This yellow baby jumped.

(ZY,2;0)

b. [e] wan shishi ne. [e] zang. [e] xl zaozao ne.

play sand NE dirty take bath NE'(He) is playing with sand. (He) is dirty. (He) is taking a bath.'

(AN, 2;3)

(13) a.

(14) a. [e] brush her hair. [e] brush hair.'(She's) brushing her hair. (She's) brushing (her) hair.'

[e] fighting like that, bang!'(They're) fighting like that, bang!'

[e] playing. They all bent. [e] are playing.'(They are) playing. They (are) all bent. (They) are playing.'

(AR,2;5)

b. He got in there. [e] fell down.'He got in there. (He) fell down.'(DS,2;10)

c. [e] jumping. [e] fell. They fell down. [e] sleeping.'(They're) jumping. (They) fell. They fell down.(They're) sleeping.'(SP,4;2)

The mean percentage of sentences with nullsubjects produced by Chinese children is 46.54%(s.e. = 3.78); while for the American children, it is33.11% (s.e. = 6.12). The Chinese adults producedsentences with null subjects 36.13% of the time.Given that Chinese is a pro-drop language, all thesentences with null subjects produced by theChinese children are considered grammatical,with the reference of the null subject determinedby the discourse topic. Although English is not apro-drop language, some of the sentences withnull subjects produced by American children, i.e.,sentences with null subjects but using inf"mitivesor gerunds rather than a full verb, can be judgedas pragmatically acceptable in the given context inwhich they were produced. If we exclude thesesentences from our count of sentences with nullsubjects produced by American children, the

mean percentage drops to 14.58% (s.e. = 5.03).Comparing this adjusted mean percentage,14.58%, with the mean percentage of Chinesechildren, 46.54%, and that of Chinese adults,36.13% [one way ANOVA omnibus F(2, 24)=17.80,p=.OOOl], it is clear that Chinese children aredropping their subjects at a much higher ratethan American children, and even a bit higherthan the rate of the Chinese adults. Thedifferences between the American children andthe Chinese children, and between the Americanchildren and the Chinese adults, are bothsignificant by Scheffe's tests [F(1,24)=31.96,p=.OOOl, and F(l, 24)=21.55, p=.0025 respec­tively1; the difference between the Chinesechildren and the Chinese adults is not significant.Even still, it is clear that American children dodrop subjects a relevant amount of the time.

Null Subject tIS. Null Object: Some EtJidencefrom the Acquisition ofChinese and English 233

For both groups of children, the null subject wassometimes clearly related to an antecedent fromthe discourse as shown in examples (15, Chinese)and (16, English). In other cases, the referent ofthe null subject was not previously mentioned in

the discourse, although it was usuallyunderstandable from the context; often, it waspart of the pictures the children were describing.Some examples of this type are given in (17,Chinese) and (18, English).

(15) a. Xiao zhiizhii zhil tangtimg.

little piggy boil soup'Little pig makes soup.'[e] zhu timgtimg.

(He) boil soup

'He makes soup.'(WW,2;5)

b. Da ye langi zai zhen tou kan.

Big wild wolfi ASP here secretly look

'The big wild wolf is here peeping secretly.'[ei] zm kim xiao zhii.(Iti) ASP look little pig

'It is looking at the little pig.'(HE, 3;1)

(16) a. Look at this bad wolf. He got in there. [e] fell down.'Look at this bad wolf. He got in there. (He) fell down.'(DS, 2;10)

b. The big bad wolf coming again and bang the door. [e] want to

blow the house and the house is down.'The big bad wolf (is) coming again and bang the door. (He)

wants to blow the house and the house is down.'(SR, 2;8)

(17) [e] kAn JingJmg. [e] mei chiian xiexie.

(He) look mirror (He) not wear shoe

'He is looking in a mirror. He didn't wear shoes.'[e] mei chiian wawa.(He) not wear sock

'He didn't wear socks.'(ZY, 2;0)

(18) [e] jump up. [e] jump in bed. [e] fall down.

'(He) jumped up. (He) jumped in bed. (He) fell down.'(AR,2;5)

234 Wangetal.

Although both Chinese- and English-speakingchildren thus produced null subjects in asomewhat similar fashion, we believe this does notnecessary show that they use the samemechanism in identifying and licensing the nullsubjects. We will return for further discussion ofthis point.

3.1.2 Null objects. From Figure 2, we may seethat there is a considerable difference between themean percentages of sentences with null objectsproduced by Chinese child subjects, which is22.53% (s.e.=1.76), or by Chinese adults, 10.3%(s.e.=1.58), and that by American child subjects,which is 3.75% (s.e.=1.31), [one way ANOVAomnibus F(2, 24)=37.21, p=.OOOl]. Here, thedifferences between the American children and

100

the Chinese children, the American children andthe Chinese adults, and the Chinese children andthe Chinese adults are all significant by Scheffe'stests [F(1,24)=18.781, p=.0001, F(1, 24)=6.549,p=.0237, and F(1,24)=12.232, p=.0001, respec­tively]. With the Chinese children, only 27.59% ofthe total sentences with null objects areungrammatical. The grammaticality of theChinese object-drop sentences (i.e., whether thenull object was used properly) was judged withrespect to the context in which the sentence inquestion was actually produced. For the Americanchildren, 100% of the sentences with null objectswere ungrammatical. Examples are given in (19)for Chinese child subjects, and (20) for Americanchild subjects.

80 ~ Chine.. children

l • American children

'0 IZJ Chine.. adult.!!.,g

600

:;c-0-c 40CDu..CDQ.

C: 20E

o

Figure 2. Mean percentage of sentences with null objects produced by Chinese and American children and Chineseadults.

(19) a. *Ou, lang lSi cm [e].

oh, wolf come eat (it=pig)lOh, the wolf came to eat (the pig).'(ZY,2;O)

b. *Tamen YElO qiu gSi [e].

they going to build (it=house)'They are going to build (a house).'(WW,2;5)

(ungrammatical)

(ungrammatical)

c. [e] ZSi kankAn [e]. (grammatical)

Null Subject vs. Null Object: Some Et1idena from the Acquisition ofChinese and English 235

(He=wolf) again look look (it=pig)'(He) had another look at (the pig).'(ZY,2;0)

d. [ei] chlwan [ej],

(He=wolf) eat finish (it=pig)'After (he) finished eating (the pig),'laohing dilzl jiou biAn da Ie.

old wolf belly then become big ASP'the old wolfs belly became big.'(LX,3;4)

(grammatical)

(20) a. *Look at [ei] . [ei]

'Look at (him). (He)(DS,2;10)

go a little higher

goes up a little higher.'

(ungrammatical)

b. *The other little pigs worry about [e].'The other little pigs worry about (him).'

(ER,3;8)

(ungrammatical)

3.1.3 Null subject/null object asymmetry.Comparing Figure 1 with Figure 2, it may be seenthat the null subject/null object asymmetry is notunique to the Chinese children. The ratio of themean percentage of sentences with null objects tothose with null subjects is 0.48, 0.23, and 0.24 forChinese children, Chinese adults, and Americanchildren, respectively. If we recalculate the ratiofor the Chinese children, excluding theungrammatical sentences as in example (19a andb), (which may be considered as errors), the ratiodecreases from 0.48 to 0.29. If we do the samething for the English children, considering theirsmall percentage of object-dropping (3.57), whichwas ungrammatical, as errors, the ratio of coursebecomes zero.

The amount of null object use by the Chineseadults is surprisingly low. However, it is impor­tant to note that we believe the ratio for Chineseadults would be higher than the rate we obtainedif the data had been collected in an Adult-to-adultconversational situation, where most object drop­ping takes place, rather than in children's story­telling. Because of this discrepancy, we conducteda follow-up study with Chinese adults.

In the follow-up study, five Chinese-speakingadults were interviewed by the experimenter in anadult-to-adult conversational setting. These adultswere all women who had recently given birth to

their first child. The interviews took place in thesubjects' homes, and consisted of several parts.First, the subjects were asked to tell their childtwo stories as a warming up. Then, they engagedin conversation with the experimenter. Theconversations all included the same three topics ofdiscussion: the woman's pregnancy and childbirth,her own lifestyle, and the growth and behavior ofher child. The interviews were tape-recorded.Only the conversations were transcribed andscored according to the same procedures discussedpreviously for the initial study. The percentages ofnull subject and null object used by each speakerin this study are illustrated in Figure 3, and moredetailed information is given in Appendix 4.

As this Figure clearly shows, a subject-objectasymmetry remains for the adult subjects, but theoverall percentage of null object use increasesgreatly. Both of these facts are important forcomparison with the children's utterances. In thefollow-up study, the average object drop is 40.1%(s.e.=l.77), while the average subject drop is45.6% (s.e.=2.42). Although the amount of objectdrop is much higher than in the initial study(10.30%), the difference between the subject-dropand the object-drop is significant by a 2-tail pairedt-test (t=4.073, p=0.015). Some examples of theadults' utterances with subject and/or object dropare given in (21) and (22).

238 Wangetal.

The pattern of use of missing objects is quitedifferent (see Figures 5 and 6). Whether dividedby age or by MLU group, the American childrenused missing objects much less frequently thannull subjects. The two-year-olds MLU 3.51) usedmissing objects only 8.3% of the time, whilethe older children used essentially none. Incontrast again, the Chinese children used nullobjects much more frequently than the American

children. They averaged 20.2% to 26.0%null objects, with the figures increasing slightlyover the ageIMLU ranges.s Although the adultsin the initial study produced far fewer null objectsthan the Chinese children, from the follow-upstudy we can see that the overall productionof null objects by the children is approachingthe level of use by adults in conversationalsettings.

o

Year of age: 2

MLU: 3.41

2

3.51

3

4.41

3,4

4.47

4

5.28

adult

Figure 5. Mean percentage of sentences with null subjects produced by Chinese and American children (by MLU,adjusted) and Chinese adults.

443322

100

lIehm-• American

l 80..¥:a0

'5 80c'0

I 40

IE

20

oYear of lIglt:

Figure 6. Mean percentage of smtences with null objects produced by Chinese and American children (by age) andChinese adults.

Null Subject tIS. Null Object: Some Evidencefrom the Acquisition ofChinese and English 239

adult43,4322

100

II Chlneu

~80 • American

--uf0

603c-0c• 40u~•A.CII

~ 20

oVear of age:

MLU: 3.41 3.51 4.41 4.47 5.28

Figure 7. Mean percent.age of smtences with null objects produced by Chinese and American childrm (by MLU) andChinese adults.

The Chinese- and English-speaking children donot differ significantly in their use of null subjectsat the earlier MLU stage tested: MLU level 3.5,but they do at the latter MLU stage: MLU level4.5. These results provide strong evidence for pro­drop in younger English-speaking kids (MLU level3.5). For the use of null objects, however, the twolanguage groups differ significantly across allMLU levels. The differences in the use of nullsubjects and null objects by Chinese and Americanchildren indicate that the factors controlling theuse of the two types of null arguments in the twogroups are distinct. This is counter to the proposalby Jaeggli and Hyams (1987) which suggests thatthe two groups use null subjects for essentially thesame reason.

3.3 Results of Eliciting ExpletiveStructures

In order to determine how the course of thedevelopment of expletive subjects interacts withthe development of null versus overt subjects,children's productions of sentences calling forexpletive subjects were examined. For the

Chinese-speaking children, we examined whetherthey used a null subject as in (5) above, or a non­expletive lexical subject as in (6). For the English­speaking children, we examined whether theyproduced any lexical expletives, and further,whether there was any evidence that lexical andnull expletives coexisted.

In general, there was no evidence of the Chinesechildren producing structures with overt non­expletive subjects, such as those in (6a, b, and c)above, even among the 4-year olds. The onlystructures they used in the weather conditionswere those with null subjects, as in (5a and b).They did not use the structure as in (5c) either.The only exception occurred when they talkedabout a windy condition. In this case they eitherused a structure with a null subject as in (23), orthey used ~ng,' ('wind'), as an overt subject as in(24). The Chinese adults used all the structures asin (5) and (6). They also used ~ng,' the word for'wind,' in the same way as the Chinese children.The observed difference here between the Chinesechildren and the Chinese adults in their use ofnull subjects (as in Sa and b), and non-expletive

240 Wangettll.

lexical subjects, (as in 6a and b), we believe, is dueto a stylistic reason rather than a grammaticalone. In fact, sentences in (5a and b) are morecolloquial than tMse in (6a and b). However, itseems that the absence of the structure like thatin (6c) from the data of the Chinese children is dueto a grammatical reason. While the null subjectsin (5a and b) can be interpreted as referential, theone in (6c) can not. The structure (as in 6c)

requires the ability to raise the subject from theembedded clause to the matrix clause.

The American children had a different pattern.Except for the youngest one, (AR, 2;5), all thechildren showed some kind of evidence for theexistence of expletive 'it' as in example (25). At thesame time, however, they also used some nullexpletives as well, as shown in examples (25) and(26).

(23) [e] yao ba zhege gui diao,(it=wind) want (BA) this blow down,

[e] h8i yao ba zhegeje gui diao.

(it=wind) also want (BA) this too blow down.

'(Wind) wants to blow this down,

(it) also wants to blow this down too.'(ML,4;3)

(24) Xianzm gui feng-Ie. Feng dim tai da-Ie,

now blow wind-ASP. Wind also too big-ASPfangzi dOu chill dao-Ie.

house also blow down-ASP

'The wind began blowing now. The wind was so big

that the house was blown down.'(SK, 4;1)

(25) It is raining.

It's very windy so the clothes are going up.

It's rain. rain. They can't come out.

(SR,2;8)

(SR,2;8)

(DS,2;10)

(26) Snow. Raining

No snow.

Windy now.

Raining.

(DS,2;10)

(SR,2;8)

(EL,3;6)

(AR,2;5)

Null Subject tIS. Null Object: Some Evidencefrom the Acquisition a/Chinese and English 241

Hyams (1986) suggests that one piece ofevidence that English-speaking children use toreset the null subject parameter to [-pro-drop] isthe presence of overt expletives. Hyams arguesthat since it and there are not being used forpragmatic purposes (because they do notcontribute to the meaning of the sentence), theymust therefore be present for strictly grammaticalreasons. Hence, lexical expletives could be used totrigger parameter resetting. Furthermore, asnoted above, Hyams found that children use nullexpletives at the time they use null subjects. Sothe emergence of lexical expletives coincident withrestructuring to [-pro-drop] is predicted.

However, as our data show, some children douse both overt and null expletives at the timewhen they are using null subjects. Hence, it seemsthat the presence of overt expletives in the inputis not a type of triggering data for resetting thenull subject parameter. But why do the childrenuse overt expletives when they sanction nullsubjects? Lillo-Martin (1987) has given areasonable solution for this puzzle. She suggeststhat children have misanalyzed the expletives,and instead interpret 'it' as referential, even insentences like, 'It's raining.' Because they havethe wrong analysis of 'it,' they don't have the overtexpletive evidence that English is not [+pro-drop).So at this point, one cannot assume that the timeat which a child starts using overt expletives will

be coincident with the correct setting for the nullsubject parameter.

3.4 Results on the Use of StructuresExhibiting Variables

In our data, both child language populationshave shown some evidence for the existence ofvariables though the production of wh-movement(English), or the comprehension and production ofwh-questions (Chinese). This can be seen in (27)and (28). These questions were produced andcomprehended during the course of the experi­ment described above, at the same time as thesechildren showed evidence of using null arguments.

One might claim, following Roeper et aI. (1984),that the empty categories used in theseconstructions are pros, not variables. However,work by Thornton (1990) and Sarma (1991)suggests that children at least at 3 years do usevariables rather than pros in these constructions,since they correctly produce long distancequestions and obey the strong crossoverconstraint. Therefore, we will assume that theempty categories used in the wh-questions shownabove are variables rather than pros. In any case,it is the difference between Chinese- and English­speaking children with respect to null objects,without a corresponding difference with respect toevidence for variables in the form of wh-questions,that is relevant to our discussion.

(27) a. What's that?(AR,2;5)

b. Who's that? Baldy? Baldy is playing with mud.(SR, 2;8)

c. That's what I think he did.(DR, 3;9)

(28) a. Experimenter: ShUi 1m-Ie?

Who came-ASP'Who came?'

Child subject: Lang, Lang 1m-Ie.

wolf, wolf came-ASP

'The wolf came.'(ZY,2;O)

242 Wang et al.

b. Experimenter: Da hui lang gan shenmo 1m-Ie?big grey wolf do what come-ASP

'Why did the big grey wolf come?'Child subject: [e] Na xiao zhu Ah .

(He) take little pig Ah!'(He) came to take the little pig away, of course.'(AN, 2;3)

c. Na shi shemo? Na sm shill nang de?that is what? that is who did'What is that?' 'Who did that?'CWW,2;5)

4. DISCUSSION: THE PARAMETERIZEDTIIEORY OF UG AND LINGUISTIC

EVIDENCE

A review of Figures 4 through 7 indicates thefollowing:

i. At the earliest age tested, 2 years old oraverage MLU of 3.5, both Chinese and Americanchildren are using null subjects. The Chinesechildren are also using null objects. Although theAmerican children do have a few sentences withnull objects, the mean percentage of theirsentences with null objects is only 3.57, so we willcount these as errors; i.e., outside of the children'sgrammars.

ii. For the Chinese children, as their MLUincreases, the mean percentage of sentences withnull subjects decreases, and the mean percentageof sentences with null objects increases. By theMLU level of 5.28, their subject-dropping rate isvery close to that of Chinese adults, and theirobject-dropping rate is approaching that of theadults in the follow-up study.

iii. For the American children, as their MLUincreases, the mean percentage of sentences withnull subjects (as well as sentences with nullobjects, which we are not counting as part of thechildren's grammar) decreases drastically, thusalso coming in line with the corresponding adultgrammar.

iv. At each MLU level, both mean percentagesare much higher for the Chinese children thantheir American counterparts, although for thefirst MLU group (MLU level 3.5) the differencebetween the Chinese- and English-speaking

children in their use of null subjects is notstatistically significant.

How can the observation that as early as 2 yearsold both Chinese and American children are usingnull arguments be explained? It might beunderstandable that Chinese children do sobecause adult Chinese is a pro-drop language.But then why would the American children alsodo so, given that null arguments are not allowedin adult English? On the other hand, how can theobserved differences between Chinese andAmerican children in the null argumentphenomena be explained along developmentallines?

If we adopt the idea that part of the formulationof UG is a system of parameters, and the initialsetting for a particular parameter is the same forall children constrained by certain principles, thenthe observed phenomena can be explained. Asdiscussed above in detail, the principles of UGmay tell us when a null subject can occur and howit can be identified. The data we obtained supportthe hypothesis that English- and Chinese­speaking children at a very early age have agrammar which allows null subjects.

We are left, however, with three importantquestions for discussion. First, how strong is theasymmetry we found comparing subject and objectdropping in English compared to Chinese, andhow can it be accounted for by parameter theory?Second, how does the child who begins with anincorrect parameter setting make the change tothe adult grammar? Third, how does the linguisticenvironment make an impact on this parameterresetting?

Null Subject vs. Null Object: Some Evidencefrom the Acquisition ofChinese and English 243

4.1 On the Subject/Object Asymmetry

Our data did not confirm Jaeggli and Hyams'hypothesis with respect to null objects. Instead,our data indicate that while the Chinese-speakingchildren used null objects from as early as 2 yearsold (the youngest age tested), the English­speaking children by and large did not use nullobjects. This returns us to the potential problemwith Jaeggli and Hyams' account discussed above.If English-speaking children have a Chinese-typelanguage as their initial parameter setting, thenwe would expect children learning both languagesto progress similarly in terms of the use of nullobjects. However, this was not the case.

We do not think that the null subject/null objectasymmetry we found in Chinese- and English­speaking children's use of null objects can beaccounted for by the non-existence of variables inearly grammar. Both the Chinese- and theEnglish-speaking children provided evidence forthe emergence of variables. According to Hyams'hypothesis, the English-speaking children in thissituation should use null objects at least asproductively as the Chinese-speaking children do,but our data show that they do not. The smallpercentage (3.57) is really within the error range.If the English-speaking children have reset theirnull argument parameter at this point, theyshould have stopped using both null subjects andobjects. Our data show that this is not the case:they continued to use null subjects but essentiallyno null objects even though they had acquiredvariables. At the same time, the Chinese-speakingchildren (who showed the same kind of evidence ofvariables) did use null objects productively.

As an alternative to Jaeggli and Hyams' hy­pothesis, we propose that there is more than asingle parameter controlling the use of null argu­ments (following Lillo-Martin, 1986; 1991). Oneparameter, which can be called the DiscourseOriented Parameter (nOP) (following Huang,1984), permits languages with discourse orientedproperties to have both null subjects and null ob­jects. These null arguments can be one of twotypes. Most are variables identified by a DiscourseTopic. In embedded subject position there is alsothe option of pro, identified by a c-commandingNP. These null arguments correspond straight­forwardly to two of the identification options pro­posed by Jaeggli and Safir, given in (llb and c)above. For learnability reasons, assumingparameter setting takes place on the basis ofpositive evidence, we might expect that the initialsetting of the DOP is [-DOl. If so, the performance

of the Chinese-speaking children in our studyindicates that resetting of the DOP to [+DiscourseOrientedl can take place early. Since othercharacteristics of discourse oriented languages,such as topic-comment structures and discourse­bound anaphors, can serve as evidence fordetermining this parameter setting, it isreasonable to assume that the Chinese-speakingchildren have made this setting and produce nullsubjects and null objects in accord with thisgrammatical option.

The second part of our proposal is that nullarguments in adult languages like Italian are dueto a separate parameter, which we will call theNull Argument Parameter. This parameterpermits null arguments when licensed by certainCase-assigning maximal categories, followingRizzi (1986). These null arguments are emptycategories of the type pro, identified by the person,number-, and / or gender-features of the licensingcategory. Although subject-verb agreement isinsufficient to license or identify null subjects inadult English, we take it that English-speakingchildren who use null subjects are doing sobecause of this parameter, rather than the DOP.The subject-object asymmetry is related to thecross-linguistic observation that object agreementis much less common than subject agreement;hence pro null objects are found in many fewerlanguages than pro null subjects. Children willuniversally posit an INFL category with thepotential of being a licenser for empty subjects,but not for empty objects. Hence, universallychildren will begin with a null subject hypothesis.Changing the parameter setting to disallow nullsubjects will thus only take place aftermorphological agreement has been analyzed.

Other proposals have been made arguing thatthe null subject phenomenon in early English isdue to performance factors rather than agrammatical parameter setting (e.g., Bloom, 1990;Gerken, 1990; Mazuka, Lust, Wakayama, andSnyder, 1986). Although these suggestions areworth considering, there is considerable cross­linguistic evidence to take the early null subjectphenomenon as representing a grammatical stage.Performance accounts of the early null subjectphenomenon do not make the same cross­linguistic predictions as grammatical accounts do.More cross-linguistic work can contribute to theresolution of this debate; but the data currentlyavailable support the grammatical account. Forreviews of performance versus grammaticalaccounts, see Hyams and Wexler (1991) and Lillo­Martin (1991).

244 WQngetQI.

4.2 Parameter Resetting

The evidence is quite strong that both Chinese­and English-speaking children have a grammarwhich allows null subjects at an early age, sincethey were both using null subjects even at the ageof 2 (examples 12a, and b and 13a, and b). For theChinese children, since the adult language allowsnull arguments, no change will have to be made intheir parameter setting. However, for the English­speaking children, a parameter will have to be re­set on the basis of evidence for [-pro drop) fromthe linguistic environment. Our data shows thatroughly between the age of 2 and 3 or MLU 3.5 toMLU 4.5, a drastic change has taken place in theEnglish-speaking child's grammatical develop­ment. That is, during this transition the English­speaking children show a dramatic decline in theproduction of null subjects. It seems to be at thispoint that the parameter resetting has takenplace.

How does this resetting occur? It is possible thatthe presence of overt expletives can be used asevidence that English is [-pro-drop), as discussedabove. However, there is now some cross-linguisticdata which indicates that the perfect correlationbetween overt expletives and [-pro-drop) which isneeded for this kind of evidence does not exist (cf.Jaeggli & Hyams, 1987, Hyams, in press). Even ifthis positive evidence is unavailable, however, it ispossible that indirect negative evidence can beused (Lasnik, 1989). For the English children,since the child's initial setting is also [+pro-drop),he would, like the Chinese children, expect to hearsentences with null subjects. When the child failsto hear sentences with null subjects in English,this will then be taken as indirect negativeevidence that such sentences are not allowed inhis language, hence, ungrammatical. Theincorrect positive parameter will then be replacedby the correct negative setting [-pro-drop).

Note that our data do agree with some empiricaldata existing in the literature, which togethermay be taken as evidence for certain a priori,language-independent properties of earlygrammar hard-wired by parameters of UG. Forinstance, with our Chinese child subjects at MLUlevel 3.5, 20% of the transitive verb constructionswere produced with null objects, which is veryclose to the 17% of the similar constructionsobtained in Japanese children (Mazuka et al.,1986). Also, for the American child subjects, themean percentage of sentences with null subjects(15%) is very close to the percentage found inGerken's imitation study (19%, subjects' mean agewas 2;3; Gerken, 1990). Further, the dramatic

decrease in the mean percentage of sentences withnull subjects observed in our American childrenbetween age 2 and 3 is consistent with Hyams'proposal of an inverse relationship between nullsubjects and the use of inflectional morphology.These studies all point to an initial [+pro-drop)setting, with resetting to [-pro-drop) for English­speaking children during the third year.

4.3 Effects of Linguistic EnvironmentWhat role does the linguistic environment play

in this parameter-setting account of languagedevelopment? Clearly, only data from thelinguistic environment can trigger the resetting ofa parameter, such as is needed for English­speaking children. However, the interactionbetween the child's initial setting of this null­subject parameter and the input of the child'slinguistic environment seems to make itself felteven earlier and in more subtle ways thanparameter resetting. Even the 2-year-olds wetested displayed a noticeable difference in the nullsubject/null object phenomena between the twotesting populations. First of all, only the Chinese­speaking children used null objects to any extent.This, as we suggested, can be due to a differentparameter from the one used for null subjects inEnglish-speaking children; one that could possiblybe set on the basis of entirely independent data.

A more extensive consideration of the role of thelinguistic environment is called for if we take intoaccount the proportions of null arguments usedacross the different age ranges in Chinese andEnglish. Although the English-speaking childrenused null subjects frequently, they still used themless frequently than the Chinese children. In thecase of null objects, we have suggested that thedifference between English- and Chinese-speakingchildren is a difference related to their grammars:the Chinese-speaking children's grammars allownull objects, while the English-speaking children'sgrammars do not. However, we do not make theclaim that the difference in the use ofnull subjectsis a grammatical difference. This seems to be aprime example of an area where the force of thelinguistic environment is felt. Furthermore, asthey develop, the use of null arguments by theChinese-speaking children approaches that of theadult subjects. For example, the Chinese adultsproduced sentences in which the null argument isinterpreted by virtue of a discourse topic estab­lished several sentences earlier, as in example(22) above. The youngest children did not exhibitthis kind of long distance topic chaining. The fac­tors that control the pragmatically acceptable use

Null SUbject vs. Null Object: Some Evidencefrom the Acquisition o[Chinese and English 245

of null arguments (as opposed to their generalgrammaticality) will need to be learned byChinese-speaking children, independent from thesetting of grammatical parameters. This will bedirectly related to the linguistic environment.9

5. CONCLUSION

In general, this study has shown some supportfor the hypothesis that English-speaking childrenbegin speaking a [+pro-drop] language. Thespecific hypothesis of Jaeggli and Hyams (1987),that early English is a Chinese-type language,received mixed support. Support in favor ofJaeggli and Hyams' proposal may be seen throughthe following points:

i. As early as 2 years old, which was theearliest age tested, the English-speaking childrenproduced sentences with null subjects at 34.57%.

ii. The English-speaking children did displayan asymmetry in the use of null subjects, com­pared to their very low incidence of null objects.

However, this data also throws Jaeggli andHyams' (1987) theory into a dilemma. They useRoeper's (1986) proposal for the later developmentof variables in order to account for the proposednull subject/null object asymmetry. Our resultshows that apart from the low level of null objecterrors, the English-speaking children never usedany true null objects, consistent with Jaeggli andHyams' analysis. However, we found this even af­ter the children had developed variables (as indi­cated by production ofWh-questions). According toHyams, the English-speaking children shouldhave displayed null objects when they developedvariables, or else they should have gone throughthe business of null argument parameter restruc­turing by this time, and displayed no null sub­jects. But our data shows that they did use nullsubjects at this age. Furthermore, the English­speaking children were different from theChinese-speaking children, in that the latter usedboth null subjects and null objects during the timewe tested them. These observations provide coun­terevidence to the Jaeggli and Hyams proposal.

This study also shows that although it isimportant to have theory guide research in thefield of language acquisition, it is likely that thedata will show where the predictions of the theoryare in error, or where the theory needsrefinement. Even if the parameter theorygenerally holds, it still could be true that theprocess of resetting might be slower for someparameters than others; in other words, in someaspects of the use of null subjects, therestructuring can be gradual and take a longer

time than was previously thought. The result ofthis study also suggests that the linguisticenvironment or linguistic input shapes the child'sgrammar from a very early stage, e.g., as seen inthe early cross-language differences in use of bothnull subjects and null objects.

REFERENCESBloom, P. (1990). Subjectless sentences in Child language.

linguistiC I"'Iuiry, 21, 419-504.Brown, R (1973). A first limguRge: The erlrly stages. Cambridge, MA:

Harvard University Press.Gerken, L. (1990). A metrical account of children's subjectless

sentences. In J. Carter, R Dechaine, B. Philip, &: T. Sherer (Eds.),Proceedings of NELS 20, Volume 1, (pp. 121-134). Amherst, MA:GLSA.

Huang, C. T. J. (1982). Logical relations in Chinese and the theory ofgrrmrtrIIlr. Doctoral Dissertation, MIT.

Huang, C. T. J. (1984). On the distribution and reference of emptypronouns. linguistic I"'Iuiry,15, 531-574.

Huang, C. T. J. (1989). Pro-drop in Chinese: A GeneralizedControl Theory. In Jaeggli &r: Safir (Eds.), The null subjectparameter, (pp. 185-214). Dordrecht: Kluwer AcademicPublishers.

Hyams, N. (1983). The pro-drop parameter in Child Grammars.WCCFL, 2, (pp. 12~139).

Hyams, N. (1986). LanguR~ acquisition and the throry ofpa1'llmeters.Dordrecht: Kluwer Academic Publishers.

Hyams, N. (1989). The null subject parameter in language acquisi­tion. In O. Jaeggli &r: K. Safir (Eds.), The null subject p41'11meter,(pp. 215-238). Dordrecht: Kluwer Academic Publishers.

Hyams, N. (in press). A reanalysis of null subjects in childlanguage. In J. Weissenborn, H. Goodluck, &: T. Roeper (Eds.),Theoretical issULS in limguRge acquisition. Hillsdale, NJ: LawrenceErlbaum Associates.

Hyams, N., &: Wexler, K. (1991). On the grammatical basis of nullsubjeds in child IllngUllge. Manuscript, UCLA and MIT.

Institute of Unguistics, Academy of Sciences. (1973). ModernChinese Dictionary. Beijing, PRC: The Commercial PublishingHouse.

Jaeggli, 0., &r: Hyams, O. (1988). Morphological uniformity andthe setting of the null subject parameter. In J. Blevins &: J.Carter, (Eds.), Proc«dings ofNELS, 18, Volume 1. Amherst, MA:Graduate linguistic Student Association.

Jaeggli, 0., &: Safir, K. (1989). The null subject parameter andparametric theory. In O. Jaeggli &: K. Safir (Eds.), The nullsubject parrnneter, (pp. 1-44). Dordrecht: Kluwer AcademicPublishers.

Lasnik, H. (1989). On certain substitutes for negative data. In R.Matthews &: W. Demopoulos (Eds.), Leul7lllbility and linguisticthrory (pp. 89-105). Dordrecht: Kluwer Academic Publishers.

Lillo-Martin, D. (1986). Parameter setting: Ellidence from use,acquisition, and breakdown in American Sign Language. Doctoraldissertation, University of California at San Diego.

Lillo-Martin, D. (1987). Parameter setting in the acquisition ofAmerican Sign Language. Presented at the Boston UniversityConference on Language Development

Lillo-Martin, D. (1991). Unitlersal Grrnnmar and American SignLanguage: Setting the Null Argument ParAmeters. Dordrecht:Kluwer Academic Publishers.

Mazuka, R., Lust, B., Wakayama, T., &r: Snyder, W. (1986).Distinguishing effects of parameters in early syntax acquisition:A cross-linguistic study of Japanese and English. PRCLD, 25,73-82.

246 Wangetal.

McCloskey, J., &: Hale, K. (1984). The syntax of person-numberinflection in modem Irish. Natural LmgWlge and LinguisticTheory, 1,487-533.

McDaniel, D., &: McKee, C. (in press). Which children did theyshow obey strong crossover? In H. Goodluck &: M. Rochemont(Eds.), The psycholinguistics ofisland constraints.

Rizzi, L. (1982). Issues in Italilln syntax. Dordrecht: Faris.Rizzi, L. (1986). Null objects in Italian and the theory of pro.

Linguistic Inquiry, 17, 501-557.Roeper, T. W. (1986). How children acquire bound variables. In B.

Lust (Ed.), Studies in the acquisition ofAnaphora Vol. I, (pp. 191­2(0). Dordrecht: Kluwer Academic Publishers.

Roeper, T., Rooth, M., Mallis, L., &: Akiyama, S. (1984). Theproblem of empty categories and bound variables in languageacquisition Manuscript, University of Massachusetts, Amherst.

Sarma, J. (1991). The acquisition ofwh-questions in English, doctoraldissertation, UniVersity of Connecticut.

Thornton, R. (1990). Adventures in long-distance mornng, doctoraldissertation, University of Connecticut.

Weissenborn, J. (in press). Null subjects in early grammars:Implications for parameter setting theory. In J. Weissenborn, H.Goodluck, &. T. Roeper (Eds.), Theoretical issues in languageacquisition. Hillsdale, NJ: Lawrence Eribaum Associates.

FOOTNOTES"LAnguage Acquisition, 2(3), 221-254 (1992).t Also University of Connecticut

ttAlso Wesleyan UniversitytttAlso Wellesley College

1The following abbreviations are used in the glosses:

[el: null argumentASP: AspectDE (footnote 7); NE (p.14); MA (p.20): Chinese particles whichhave no stress, and no meaning of their own when used in astatementBA (PE): a passivizing morpheme in Chinese

2Chinese examples not otherwise credited are provided by QW.3The null subjects in (Sa, b &. c) can be interpreted or understood

as "sky."41n his (1989) paper, Huang amends this option in a way which

also allows the matrix subject to be pro, by saying that an

empty pronominal (pro) must be identified by the closestnominal element if there is one. We will continue to adopt the(1984) analysis, by which only embedded subjects can be pro.

5Roeper, Roath, Mallis, and Akiyama make this suggestion for acompletely different reason. They discuss an experiment inwhich children appear to violate strong crossover for a longperiod of time. They account for this finding with thehypothesis that children begin with pro but not variables asempty categories. However, there is new evidence whichsuggests that children do not actually violate strong crossover(see McDaniel &: McKee, in press, Thornton, 1990), and thatthey do have variables.

6The experimenter, QW, is a native speaker of Mandarin fromthe People's Republic of China. She is also fluent in English.

'None of the Chinese children in MLU group 3.5 (2-year-olds)and 4.5 (3-year-olds) produced any sentences with embeddedclauses. Only one of the 4-year-olds (YO) produced fewsentences with embedded clauses. However, all three of hissentences with embedded clauses were produced with an overtsubject, e.g.,

18 xiing, !Jio IIIng chiii iN dao ~mutbu (angzi de.He thought, old wolfblow not down this wood house DE'He thought that the old IDOlf could not blow down the woodhouse:

8Statistical comparison between the use of null objects by theAmerican children and the Chinese children was wmecessarygiven the big differences between the ranges of thepercentages.

9An interesting comparison can be made with the acquisition ofGerman. Weissenborn (in press) claims that adult German islike Chinese in allowing null arguments identified by discoursetopics, but he says that the occurrence of null arguments inGerman is more restricted than in Chinese, according topragmatic factors. As he points out, Gennan-speaking childrenwill then need to learn those pragmatic factors which allow fornull arguments in German on the basis of more linguisticexperience than that which allows the Discourse Oriented

•Parameter to be set. He indicates that the development of thecorrect use of null arguments in German takes some time.

IOCC=Chinese Children; AC=American Children;AAC=Adjusted American Children; CA=Chinese Adults.

Null Subject vs. Nun Object: Some Etlidencefrom the Acquisition ofChinese and English 247

APPENDIX 1: CHILD SUBJECTS

Subject Language age Sex MLU Subj.drop Obj.drop Adj.Subj.drop

zy Chinese 2;0 F 2.41 48.103 15.952

AN Chinese 2;3 M 3.60 62.144 21.335

WW Chinese 2;5 F 4.23 56.937 23.077

HE Chinese 3;1 F 4.44 58.669 24.159

LX Chinese 3;4 M 4.27 44.532 12.827

ZZ Chinese 3;5 F 4.52 33.750 27.143

SK Chinese 4;1 M 5.04 45.439 22.479

ML Chinese 4;3 M 4.83 40.756 29.365

YD Chinese 4;4 M 5.98 28.572 26.250

AR English 2;5 M 2.69 58.636 8.333 51.177

SR English 2;8 F 4.10 27.922 9.091 17.388

DS English 2;10 F 3.74 17.156 7.500 9.091

EL English 3;6 F 4.58 11.395 3.125 3.949

ER English 3;8 M 4.80 25.981 5.179 5.390

DR English 3;9 F 4.65 14.063 0.000 4.087

SP English 4;2 F 4.49 59.524 0.000 18.831

SM English 4;4 M 3.84 45.834 0.000 4.167

PT English 4;5 M 4.51 37.436 0.000 17.179

248 Wang et al.

APPENDIX 2: RESULTS FROM ADULT SUBJECTS

Subject Subj.-drop Obj.-drop

BM 33.670 6.719

BX 39.136 22.028

ET 43.363 10.417

LM 32.834 10.976

LP 25.322 8.495

QG 26.423 7.143

QQ 40.94 11.334

we 40.298 8.929

YL 43.177 6.667

Null Subject V5. Null Object: Some Evidence from the Acquisition of Chinese and English

APPENDIX 3: RESULTS FROM CHILD SUBJECTS

Mean percentages Qf sentences with null subjects and wjth null Qbjects

Subj.ll Subj.-drQp (s.e.) Obj.-drQp (s.e.)

CC 46.543 3.776 22.533 1.761

AC 33.105 6.120 3.572 1.313

AAC 14.584 5.025

CA 36.129 2.296 8.387 2.123

Testing results arranged accQrding tQ chrQnQlQgical age

Subj. Age MLU Subj-drQp (s.eJ Adj.SD (s.e.) Obj.-drQP (s.e.)

CC 2 3.41 55.728 4.098 20.192 2.165

CC 3 4.41 45.650 7.215 21.376 4.361

CC 4 5.28 38.252 5.026 26.031 1.991

AC 2 3.51 34.571 12.427 25.885 12.871 8.308 0.459

AC 3 4.65 17.146 4.484 4.475 0.459 2.948 1.653

AC 4 4.28 47.597 6.437 13.392 4.637 0 0

Testing results arranged accQrding tQ MLU

Subj. Age MLU Subj-drQp (s.e.) Adj.SD (s.e.) Obj.-drQP (s.e.)

CC 2 3.41 55.728 4.098 20.192 2.165

CC 3 4.41 45.650 7.521 21.376 4.361

CC 4 5.28 38.252 5.026 26.031 1.991

AC 2 3.51 34.571 12.427 25.885 12.871 8.308 0.459

AC 3,4 4.48 32.372 7.660 8.933 2.884 1.474 0.991

249

25lJ Wangetal.

APPENDIX 4: THE FOLLOW-UP STUDY

Subject Total # of # of sentences % %sentences with transitive verbs Subj.-drop Obj.-drop

HD 295 176 41.36 38.07

HH 264 132 47.73 43.94

LQ 288 97 38.54 35.05

SL 316 122 49.68 39.34

TJ 344 167 50.87 44.31