Second Language Research 20,1 (2004); pp.

132

A minimalist approach to null subjects and objects in second language acquisition

Hyeson Park University of South Carolina

Studies of the second language acquisition of pronominal arguments have observed that: (1) L1 speakers of null subject languages of the Spanish type drop more subjects in their second language (L2) English than first language (L1) speakers of null subject languages of the Korean type and (2) speakers of Korean-type languages drop more objects than subjects in their L2 English. An analysis of these two asymmetries is conducted within the Minimalist Program framework (MP), which hypothesizes that language acquisition involves the learning of formal features of a target language. I propose, based on Alexiadou and Anagnostopoulou (1998), that the licensing of null subjects is conditioned by the interpretability of agreement features. When a language has [+interpretable] agreement features, raising of the verb to T (X-movement) satisfies the EPP requirement: hence, a null subject is allowed. On the other hand, in a language with [interpretable] agreement features, the subject is obligatory since merger of the subject in the specifier of TP (XP-merge) is required to check the EPP feature. Learning of the obligatory status of English subjects is easier for Korean learners than for Spanish speakers since syntactically both English and Korean have the same feature value [interpretable] (although null subjects are allowed in Korean for pragmatic reasons). Spanish has the opposite syntactic feature value [+interpretable] and resetting of this is more difficult. Licensing of null objects is hypothesized to be related to the strength of theta-features. Languages with strong theta- features, such as English and Spanish, do not allow null objects, whereas languages with weak theta-features like Korean allow null objects. It takes time for Korean speakers to learn the different value of English theta-features, resulting in the extended null object period in L2 English of Korean L1 speakers.

I Introduction There have been extensive studies of the null argument phenomenon in the acquisition of English as a second language (L2), and two observations have drawn L2 researchers attention. The first is that Romance first language (L1) speakers drop many
Address for correspondence: Hyeson Park, Department of English, University of South Carolina, Columbia, SC 29208, USA; email: parkhy@gwm.sc.edu Arnold 2004 10.1191/0267658304sr228oa

A minimalist approach to null subjects and objects in SLA

subjects in the early stages of L2 learning, while East Asian (e.g., Chinese and Japanese) L1 speakers do not drop subjects from the earliest stages of learning English as an L2 (White, 1985; Hilles, 1986; 1991; Phinney, 1987; Lakshmanan, 1991; 1994; Zobl, 1994; Yuan, 1997; Wakabayashi, 2002). The different acquisition patterns between these two language groups were clearly shown in Lakshmanan (1994), which investigated null subjects in the developing English of four children: two Spanish-speaking children, one French and one Japanese-speaking child. Of the four children, the Japanese child did not drop subjects from the very beginning (in fact, there were only three instances in the entire data), while the other three children with Romance L1 backgrounds dropped quite a few subjects. Another interesting finding in previous studies is that East Asian L1 speakers drop more objects than subjects in their L2 English (Yuan, 1997; Zobl, 1994). For example, Chinese adult learners of English in Yuans (1997) study were observed to have more difficulty in judging ungrammatical null object sentences than ungrammatical null subject sentences. In my analysis of null arguments in the interlanguage of six Korean children learning English as an L2 (see Section II), I found that: (1) the Korean children rarely dropped subjects from the early stages, and (2) the Korean children dropped more objects than subjects. These results from another East Asian language, Korean, support the findings of previous studies. The asymmetry between Romance L1 and East Asian L1 learners of English with regard to null subjects, and the asymmetry between subject and object found in East Asian L1 speakers developing English grammars need to be accounted for. In this article, I explore some explanations for these two asymmetries within the framework of the Minimalist Program (MP), which hypothesizes that language learning mainly involves learning feature values of a target language. The specific research questions are: Why is it the case that the Korean L1 children in my data do not drop subjects in their early L2 English, as opposed to Spanish L1 speakers who have been observed by many L2 researchers to drop subjects in their L2 English? Why do the Korean children drop more objects than subjects in their L2 English? The article is organized as follows: in Section II, I present my Korean data, in which the asymmetry between null subject and null object is clearly seen. In Section III, after a brief introduction to the MP, I review and critique two previous studies that examined null arguments in L2 within the Principles and Parameters approach (P&P) and the MP approach. In Section IV, incorporating the

Hyeson Park

strengths of the previous studies, I develop a new theory of null subjects and objects and apply this theory to account for the observed asymmetries. II Data In order to address the research questions, the transcripts of six Korean children were examined. The data were collected by the National Center for Bilingual Research (NCBR) between 1981 and 1983. The following description of the subjects relies heavily on a research report by Kim and Hong (1982), the researchers who directed the collection of the data. The children were aged between 4 and 9 at the beginning of the study (their initials are used to protect their identity):
1) JH: 4;7 YJ: 5;0 KH: 7;1 SM: 7;3 SE: 9;7 JA: 9;6

The data collection began nine months after the children arrived in the USA and continued for three years. The data were gathered at school and at home once a month. At school, the children wore a small tape-recorder on a belt around their waist and were recorded for 60 to 90 minutes while they were engaged in various activities. The children used Korean most of the time at home, and semi- or structured elicitation tasks were used to obtain English utterances. Thus, the data include both spontaneous speech and elicited speech from story-telling and directed conversation. The data consist of both English and Korean, and in some cases KoreanEnglish mixed utterances. For this study I looked at only English utterances. The number of English utterances produced by each child and analysed for our study is as follows:
2) JH: 395 YJ: 469 KH: 823 SM: 838 SE: 843 JA: 813

The Mean Length of Utterance (MLU) for each child at the beginning of the data collection period is shown in Table 1, which is adapted from Kim and Hongs (1982: 42) Table 7.1
Kim and Hong (1982) state that they calculated the MLU and the stages of the subjects based on Brown (1973). MLU, which is calculated based on the number of morphemes per 100 utterances, has been used to measure the level of a childs linguistic development in L1 acquisition. Of the three children studied in Brown (1973), Eve reached stage 3 when she was 22 months old, while Adam and Sarah reached stage 3 when they were about 36 months old. Thus, Kim and Hong estimate that the Korean childrens English at the beginning of the data collection was comparable to or below that of native English-speaking 3-year-old children.
1

A minimalist approach to null subjects and objects in SLA

Mean length of utterance Number of utterances 78 4 74 55 75 319 Number of morphemes 125 4 231 141 216 1022 Mean length of utterance (MLU) 1.60 1.00 3.12 2.56 2.88 3.20 Stage

Table 1

Subject (grade)

JH (pre-K*) YJ (pre-K*) KH (1st) SM (1st) SE (3rd) JA (3rd)

1 1 3 2 3 3

Note: *pre-Kindergarten.

The subjects produced utterances of very limited length. In particular, the data show that the youngest children JH and YJ were at Stage 1 at the beginning of the data collection, which was nine months after their arrival in the USA. The fact that they were at Stage 1 at the beginning is very important in relation to the analysis of the data. According to the data analysis, the children did not drop subjects even in the earliest stages, which is in contrast with the findings of other studies in which child L2 learners are observed to drop subjects in the initial stages. It might be argued that, since the data collection began nine months after the Korean children had arrived in the USA, the children had already passed the null subject stage. However, the MLU shows that the youngest children were in Stage 1 when they were not dropping subjects. It is unlikely that the data collection began too late to reflect the childrens grammar in the initial stages. The general picture of the null argument phenomenon in the developing English of the six Korean children in my study is the following: There is little evidence for the dropping of subjects by any of the early-stage learners. There was asymmetry between null subjects and null objects: the children dropped more objects than subjects; and There was no significant relation between the number of null objects and the length of residence in an English-speaking country; that is, the percentage of null objects did not decrease in proportion as the length of residence increased, as is shown in Tables 34. Table 2 shows the number of null subjects and null objects together with the number of complete utterances produced by each of the children. A sequence was considered to be a complete

Hyeson Park
Table 2 Total omission of subjects and objects Subject JH YJ KH SM SE JA Total 6/395 12/469 10/823 2/838 5/843 22/813 57/4181 Percentage 1.5 2.5 1.2 0.2 0.5 2.7 1.4 Object 28/395 26/469 38/823 43/838 33/843 51/813 219/4181 Percentage 7.0 5.5 4.5 5.1 3.9 6.3 5.2

utterance if it contained a finite verb. For example, utterances like we working or to go home were not included in the complete utterances. Routinized expressions such as Thank you and Let me see were also excluded. In counting null subjects, only the nonimperative sentences were considered. In English, object deletion is considered to be a lexically controlled phenomenon (Lehrer, 1970; Allerton, 1975; Rice, 1988; Levin, 1993). Some verbs allow object deletion without changing meaning,whereas some allow a null object only in certain contexts. I counted all cases of object deletion, regardless of whether they were grammatically or pragmatically acceptable, if an object was missing from a verb whose subcategorization information requires an internal argument.2 The whole corpus of data on each child was divided into smaller samples. Each sample covers a three-month period. Sample 1 consists of data collected from March to May 1981, and the last sample covers June to August 1984. Table 3 gives the number of null objects in each sample for each of the six children. Table 4 shows the corresponding percentage of null objects for each child and period. Overall, Korean children rarely dropped subjects from the early stages of L2 learning. In those few cases where they dropped subjects, the omitted subjects were either first person I or retrievable from the context.
Most of the null object utterances the children produced were ungrammatical. However, there were some cases where the four native speakers I consulted did not agree in their judgements. The following are some of the examples where native speakers showed varied judgements. The numbers in parentheses are the number of informants who provided their judgements. a. The girl ran. They were chasing __. (ungrammatical-3, grammatical-1) b. (Whats he doing to the floor?) Hes cleaning __ up. (ungrammatical-1, marginal-2, grammatical-1) c. They watch __ and they tell them. (ungrammatical-2, grammatical-1) d. In order to eat lunch at 2 pm, he prepared __ in advance. (ungrammatical-1,grammatical1, marginally grammatical-2)
2

A minimalist approach to null subjects and objects in SLA

Omission of objects (numbers) JH YJ KH 3/72 8/95 1/23 2/51 5/154 3/132 2/119 0/20 3/155 27/823 43/838 32/843 SM 2/13 0/39 0/4 6/78 14/166 6/261 9/149 4/68 2/60 SE 0/11 2/47 1/19 7/45 7/250 7/185 3/195 5/74 0/17 JA 5/86 1/10 5/58 15/232 3/121 2/60 4/60 8/186 43/813

Table 3 Sample 1 2 3 4 5 6 7 8 9 10 Total

2/13 0/10 4/53 0/25 3/22 5/86 5/88 0/8 6/90 25/395

2/28 3/74 9/120 3/21 5/110 4/37 0/79 26/469

Table 4 Sample 1 2 3 4 5 6 7 8 9 10 Total

Omission of objects (percentages) JH YJ KH 5.6 10.5 4.3 3.9 5.2 3.0 4.2 0.0 0.0 2.6 4.6 SM 15.4 0.0 0.0 7.7 8.4 2.3 6.0 5.9 3.3 5.1 SE 0.0 4.3 0.0 17.8 2.8 4.3 2.1 5.4 0.0 3.9 JA 7.0 10 8.6 7.6 4.1 5.0 6.7 4.8 6.3

15.4 0.0 7.5 0.0 13.6 6.0 7.0 0.0 6.7 7.0

7.1 5.4 7.5 14.3 4.5 8.1 0.0 5.5

3) a. YJ: She give it to her. ___ pay the money b. KH: It was Yoonhees fault. __ keep talking to you.

In the contexts where the children answered an interviewers question, subject deletion was also found:
4) a. b. Interviewer: JA: Interviewer: JA: What do they do to her at the hospital? __ give __ medicine. What does the leader have to do? __ find the thing and then, say a word.

The children dropped more objects than they dropped subjects. Some of the null objects are topic-referring, while others are not.
5) KH: His mother had a bandage and put ___ on his leg. (Q: do you know whos David?) I still remember ____. She did ___ one time. He cried and thats why she give him ____.

Hyeson Park

More examples of null argument sentences produced by the children are provided in Appendix 1. III Previous studies of null arguments in L2: Yuan (1997) and Wakabayashi (2002) 1 Minimalist Program and language acquisition The aim of linguistic theory, according to Chomsky (1986), is to answer the three questions: 1) What constitutes knowledge of language? 2) How is knowledge of language acquired? 3) How is knowledge of language put to use? The endeavour to answer questions (1) and (2) has been at the heart of progress in generative linguistics since the 1960s, and this has led to the development of the Minimalist Program (MP) in recent times. Generative linguists have been interested in studying how language ability is embodied in the human brain and how this ability has come into existence rather than how language is used in real communication in a society. The former is, according to Chomskys terminology, I(nternalized)-language, while the latter is E(xternalized)-language. The I-language, which is a speakers linguistic competence, consists of the lexicon and the principles of combination, which take a finite number of lexical items and form a potentially infinite number of phrases and sentences. The general consensus among generative linguists is that a child is equipped with a mechanism that helps him or her to construct the I-language based on E-language input, called Universal Grammar (UG). In the Principles and Parameters (P&P) approach, UG is assumed to consist of a fixed and invariant set of languageindependent universal linguistic properties, which are inviolable constraints on the well-formedness of constructions, and a finite set of parameters allowing typological variation. In the MP, which is committed to explanation through minimization, the sources of parametric variation have been drastically reduced to Saussurean arbitrariness (association of concepts with phonological matrices), properties of grammatical formatives (inflection, etc.), and readily detectable properties that hold of lexical items generally (the head parameter) (Chomsky, 1995: 169). In the MP, differently from P&P, there are only two levels of representation, PF and LF. These two levels are accessed by performance systems: PF by sensorimotor systems and LF by systems of thought. The reduced parametric variation and the

A minimalist approach to null subjects and objects in SLA

minimized syntactic domain require a different perspective on language acquisition from that of P&P. In the MP, sentence-building begins by forming a numeration, which is a set of lexical items with their indices. The computation selects lexical items from the numeration and builds up an expression. Two basic computational processes involved in sentence building are merge and move. Merge selects a lexical item from the numeration and combines it with a previously formed structure. When two items are combined, the target of merge projects. For example, in building up the sentence Mary kissed Bill, the first step is to select kiss and Bill from the numeration and to merge the two items. Since kiss is the target of merge, it projects:
6) kiss kiss Bill

Next, Mary is selected and merged with kiss and the target of merge, kiss, projects:
7) Mary kiss kiss kiss Bill

Move is a process driven by morphological considerations, that is, the requirement that some feature F must be checked. When target K has a strong feature F, it needs to be checked before Spell-out and it attracts the closest appropriate a, which is the minimal element including the feature F that can enter into a checking relation with K. Thus, in Chomskys terminology, the overt morpheme pied-pipes with its covert feature. When target K has a weak feature, there is no pied-piping of the overt morpheme a and only the feature F moves. The strength of the morphological features, then, is the locus of typological variation in the MP. For example, the difference between wh-movement languages and whin-situ languages can be explained by the different strength of the [+wh] feature. If the head C of the CP of a language has a strong [+wh] feature, the wh-word moves overtly into the specifier of CP to check the strong feature of C [+wh]. In a language with a weak C [+wh] feature, wh-movement is covert, that is, the wh-word moves at LF. Some languages require an obligatory overt subject: within the P&P framework, this was explained by the Extended Projection Principle, that requires that all sentences have a subject. However, the MP explains this requirement by feature movement.A language

Hyeson Park

which requires an obligatory overt subject is assumed to have T(ense) with a strong D feature, which needs to be checked against the D feature of the subject DP. For example, English T(ense) has a strong D feature and requires a subject to check its feature: hence, the subject generated in the specifier of vP moves to the specifier of TP and checks off the D feature of T. On the other hand, Spanish, due to its weak D feature on T(ense), allows a null subject. The Spanish null subject phenomenon will be discussed in more detail later in the article.
8) TP subjectj [+D] T [+D] T vP objecti tj v v v + Vk V tk ti

In the MP framework, the computational process that involves merge and move is assumed to be universal, and typological variation is due to variable morphological properties, whether features or morphemes. Thus, parameter setting, if there is any, seems to involve learning the morphological properties of the target language. The task of a child learning his or her first language may involve selecting and fixing the correct values of features and morphemes out of the restricted options available. I argue using the null argument phenomenon as an example that L2 learning also involves learning of the feature values of a target language. That is, I propose that UG is still available in L2 learning, at least in L2 learning in childhood. This position contrasts with theories such as the Fundamental Difference Hypothesis (Bley-Vroman, 1989) which view L2 learning to be different from L1 acquisition in that L2 learning involves general cognitive abilities rather than a language specific mechanism like UG. In the following I review two previous studies of null arguments within the pre-MP and early MP framework.

10

A minimalist approach to null subjects and objects in SLA

2 Yuan (1997) Yuan (1997) examined null subjects and null objects in the L2 English of Chinese learners. The theoretical model in which he works is a combination of the P&P approach and the early MP of Chomsky (1995). Yuan administered a sentence acceptability judgement test to 159 adult Chinese learners of L2 English, who were divided into seven groups based on their English proficiency. The acceptability judgement questionnaire included English sentences with null subjects and objects (experimental sentences) and sentences with overt subjects and objects (control sentences). The results of data analysis show that: The Chinese subjects, except for some elementary learners, did not have much difficulty in judging the null subject sentences as ungrammatical. The subjects, including the most advanced learners, had difficulty in rejecting ungrammatical null object sentences. In order to account for the data, Yuan adopts the licensing and identification conditions of null arguments from P&P, but reinterprets them within the MP. Licensing of null subjects, according to Yuan, relies on the strength of inflectional features of the verb and the level where these features are checked. In Spanish, the I(nflectional)-features of the verb, which consist of Agr(eement) and T(ense)- features, are strong and are visible at the level of Phonetic Form (PF) and they must raise to I(nfl) at S-structure to be checked at PF. In English the I-features of the verb are weak and are not visible at PF, hence will not raise to I for feature checking until LF. In Chinese, due to the total absence of agreement morphemes, the verb does not have any I-features; that is, the I-features are underspecified and the verb never raises to I in Chinese. Yuan further hypothesizes that a null subject is licensed in two situations: first, when the verb raises to I at PF, resulting in the government of the subject position by a lexical head at PF, which is the case in Spanish-type languages and, secondly, when the subject position can never be lexically governed, which is the case in Chinese-type languages. A null subject is not licensed if the subject position is governed by a lexical head at LF, as in English. Null arguments have to satisfy identification conditions as well as the licensing conditions. Identification conditions for null subjects vary across language groups. Following the standard approach in P&P, Yuan hypothesizes that in Spanish-type languages, the rich agreement morphology identifies the content of the null subject, while in Chinese-type languages without any agreement

Hyeson Park

11

morphology, null subjects are identified by the topic in a topic chain. Based on these theoretical assumptions, Yuan sets out to explain unlearning of null subjects by Chinese learners of English: these learners abandon null subjects when they learn that in English I-features are specified and weak. There is some possible positive evidence in the data to help them become aware of the specification of I- features: do-support and copular be verbs. That is, when the learners observe that the semantically empty expletive verbs are used as bearers of Agr(eement) and T(ense) features, they come to realize that I-features are specified in L2 English as opposed to the underspecified I-features in their L1. Once they realize the specification of I-features in English, they should learn that the I-features of English are weak, hence cannot license null subjects in the spec of IP. This will happen when the learners notice the bare stems in the English verbal paradigm. Consequently, the Chinese learners of English abandon null subjects in their interlanguage since they now know that the subject position, which is not governed by a lexical head at PF, cannot be null. In his explanation of null objects, Yuan adopts Huangs proposal on the topic chain, and proposes that Chinese is a [+topic-drop] language and null objects are licensed and identified by discourse topic. English is a [topic-drop] language, in which null objects are not allowed. The difficulty the Chinese learners have in detecting ungrammatical null object sentences in English lies in the fact that they were not able to abandon the [+topic-drop] setting of the parameter, possibly due to the lack of positive evidence in the input data to help them become aware of the [topic-drop] setting of English. The asymmetry between null subjects and null objects in the Chinese learners English is, according to Yuan, due to the presence vs. absence of positive evidence in the input data. Though Yuan appears to nicely explain the asymmetry of null subjects and objects obtained in the experimental study, there are some theoretical problems in his analysis. It depends on the hypotheses that: If the subject position cannot be governed by a lexically specified head I until LF due to absence of V to I raising, the subject NP must be phonetically realized. If the subject NP can never be governed by a lexically specified head I at any level, then a phonetically null subject is also possible. These hypotheses, though expressed in different language, are very similar to Jaeggli and Safirs (1989) Morphological Uniformity Principle (MUP), which states:

12

A minimalist approach to null subjects and objects in SLA Null subjects are permitted in all and only those languages that have morphologically uniform inflectional paradigms. An inflectional paradigm P in a language L is morphologically uniform if and only if P has either only underived inflectional forms or only derived inflectional forms.

It has been pointed out that the biggest problem with the MUP is that it is a good description of the null subject phenomenon, but it does not explain why there should be such a relation between agreement morphology and null subjects. The same criticism can be applied to Yuans hypotheses: he assumes these hypotheses as given, but does not explain why the subject position has to be governed before LF or completely ungoverned in order to license the null subject.3 3 Wakabayashi (2002) Wakabayashi (2002) is another study that analysed null arguments in L2 within the MP framework. Differently from Yuan, Wakabayashis main concern is with the contrast between Spanish and Japanese learners of English in their knowledge of the obligatory status of subjects in English. He collected data using an on-line reading time measure and an off-line grammaticality judgement task. The subjects of his study were 44 Japanese learners of English (JLE) (15 advanced and 29 intermediate), 15 Spanish learners of English (SLE) and 24 native speakers of English as a control group. The materials for the reading time and grammaticality judgement task consisted of sentences containing a null referential subject or a null weather subject:
9) a. Type 1: Null referential subject * When Tom saw a woman wearing a hat, he thought ___was Sams elder sister, but she wasnt. b. Type 2: Null weather subject * During the lecture, __ rained very hard but the sun came out after that.

The results of data analysis reveal that the native speakers of English and Japanese learners of English are more sensitive to the ungrammaticality of null subject sentences than the Spanish learners of English. For example, in the grammaticality judgement task, the advanced JLE learners were most accurate in detecting the ungrammatical null subject sentences of both types (type 1: 71%, type 2: 71%), followed by the intermediate JLE group (type 1: 50%, type 2: 51%) and the SLE group (type 1: 42%, type 2: 42%).
Dependence on government is also problematic since the MP abandoned government as a component of the computational system.
3

Hyeson Park

13

Wakabayashi proposes that this difference between JLE and SLE is due to differential L1 transfer effects. In order to account for the difference between the two groups in their knowledge of the English subject, he develops an analysis based on the following theoretical assumptions: In English, T(ense) has a [+merge-in-overtsyntax] feature due to a strong nominal feature, i.e., D feature, and T merges with VP in overt syntax. The strong D feature associated with T attracts the D feature of the subject in the specifier of VP into the specifier of TP. The strong D feature and nominative case are checked and deleted before Spell-out. Thus, in English a subject DP is obligatory. The following is the syntactic object formed at the point of Spell-out in English.
10) TP Ii [1st person, sing] T T ti VP V read the book

In the case of Japanese, there are two possible derivations: a null subject sentence and an overt subject sentence. In Japanese, T does not have a [+merge-in-overt-syntax] feature due to a weak D feature of T. Thus, T merges with VP after Spell-out in order to satisfy the semantic requirement and EPP. The tree in (11) shows the syntactic object formed at the point of Spell-out.
11) VP

sono hon-o yomu the book read

Wakabayashi argues that the subject and T are merged in covert syntax since they do not have phonological and semantic content. When a subject DP has a phonological feature, it merges in overt syntax at [spec, VP].
12) DP watasi-ga VP V sono hon-o yomu

14

A minimalist approach to null subjects and objects in SLA

Thus, in Wakabayashis analysis, the difference between null subject and overt subject sentences depends on the phonological content of the subject DP.4 This analysis does not take into consideration the difference between overt and covert subject sentences at the pragmatic level, that is, the topic relatedness of null subjects. Wakabayashi appears to try to present a simpler and clearer analysis of null subject sentences by excluding pragmatic functions of null subjects. However, it cannot be ignored that speakers choice of overt subject or null subject is strongly affected by the function of the DPs at the discourse level. In Spanish, T has the feature [+merge-in-overt-syntax] due to its strong verbal feature, and T merges with VP in overt syntax. The strong verbal feature of T attracts the verb: hence, verb raising to T instead of subject raising to the specifier of TP satisfies the EPP requirement in Spanish. Thus, according to Wakabayashi, an overt preverbal subject in Spanish is a topic, not a subject. With the above theoretical assumptions as background, Wakabayashi explains the asymmetry between Spanish-type and Japanese-type L1 learners of English: what is involved in the learning of English subjects by Japanese L1 speakers is figuring out that English T has a strong D feature, which requires T to merge in the overt syntax. On the other hand, Spanish learners of English will transfer the strong-nominal feature of T from their L1 to L2, but at the same time they have to learn that what agreement affixes do in their L1 is done by the subject DP in English. That is, they have to learn that the agreement features have different PF forms in L1 and L2: agreement affixes in L1 and subject DP in L2. Wakabayashi suggests that since learning a new feature (JLE) will be easier than unlearning an L1 feature (SLE), Japanese learners of English find it easier to learn obligatory DP subjects in English than do Spanish learners of English.5

Wakabayashi is not explicit about the position into which the covert subject is merged. If he assumes that the covert subject is merged in the specifier of VP at LF, a question that needs to be considered is whether merger of the covert subject in the specifier of VP is possible after the VP has already been spelled out. If he assumes two subject positions, the specifier of VP for the overt subject and the specifier of TP for the covert subject, another question arises. It has been suggested that subjects base-generated in the specifier of TP and the specifier of VP are different in meaning: quantificational and existential, respectively (Diesing, 1992). It is not clear whether there is any difference in meaning between the overt and covert subject in Japanese. 5 A reviewer pointed out that learning that English has a strong D feature on T is just as much unlearning the fact that Japanese does not have such a feature as discovering that English has subjects rather than agreement affixes is unlearning for Spanish speakers. That is, unlearning is involved for both groups of speakers.
4

Hyeson Park

15

Although I agree with Wakabayashi that Spanish verbal inflections have the same function as the subject in English, I disagree with his claim that learning this difference between Spanish and English is more difficult than learning of the strong D feature of English is for Japanese learners. Since the difference between English and Spanish is a PF phenomenon, while the difference between English and Japanese is related to a more abstract feature, the latter should cause more difficulty to L2 learners. Another problem for Wakabayashis analysis is his treatment of Tense in Japanese. Wakabayashi proposes that T in Japanese does not have any phonological content, hence it merges with VP in covert syntax. The null subject is thus base-generated in the specifier of TP and the overt subject in the specifier of VP. His analysis of Japanese, however, raises a technical problem. It is generally assumed that nominative case has a strong formal feature that needs to be checked off before Spell-out, and the checking position is the specifier of TP. If T merges with VP at LF, as Wakabayashi proposes for Japanese, then a sentence with an overt nominative subject will crash since nominative case cannot be checked off before LF. Thus, for a successful derivation, TP must be projected before Spell-out.6 Another problem with the covert TP analysis is that it cannot apply to Korean since there is evidence that Korean does have overt tense and agreement markers, even though they are weak.7 As mentioned above, the discoursepragmatic function of null subjects also needs to be taken into consideration in order to provide a more complete picture of the null argument phenomenon. In Section IV, building upon Wakabayashis proposal, I develop a new analysis of null arguments and apply this new theory to the null argument phenomenon in the L2 data. IV A new analysis 1 Null subjects in MP One of the key findings of studies on null subjects within the P&P approach was that licensing of null subjects has a close relationship
In Chomsky (2001), case checking can be implemented as Agreement without movement. In other words, subject raising does not necessarily involve case checking. Even if we adopt this most recent approach to case checking, it is still a relevant question where the overt subject in Japanese gets its nominative case from if there is an overt subject, and it merges with VP, and T merges covertly. I would like to thank an anonymous reviewer who pointed this out to me. 7 In traditional Japanese grammar, it is proposed that tense is marked by the overt morphemes -u/-ta. Wakabayashi states in a footnote that these may be the reflex of aspect, not tense.
6

16

A minimalist approach to null subjects and objects in SLA

with the agreement system of a language: languages with rich agreement morphology tend to allow null subjects more freely, while languages with weak agreement morphology have a limited distribution of null subjects (Huang, 1984; 1989; Rizzi, 1986; Jaeggli and Safir, 1989; Speas, 1994). Speas (1994) reinterpreted this observation within the early MP framework, which was further developed by Alexiadou and Anagnostopoulou (1998) within a more recent feature checking model of the MP. In order to offer a better explanation of the intuitive idea that rich agreement and the distribution of null arguments are linked, Speas (1994) proposes a new theory based on the Principle of Economy and a proposal by Rohrbacher (1992; see Speas, 1994): Principle of Economy: Project XP only if XP has content. Rohrbachers Generalization: Strong morphemes have individual lexical entries. Weak morphemes do not have individual lexical entries. Speas argues that in languages with strong Agr(eement) morphology, like Spanish and Italian, an agreement morpheme is base-generated in the head position of AgrP. Since the head has content, the AgrP can be projected without its specifier position being filled. In weak Agr languages, such as English and French, the agreement affix is lexically base-generated within the verb. In this case, AgrP does not have a head with content, and [spec, AgrP] must be filled with an NP in order for the AgrP to be projected without violating the Principle of Economy. Therefore, in Englishtype languages, the VP-internal subject moves to [spec, AgrP] or an expletive is inserted. In languages with no agreement morphology, like Korean and Japanese, there is no need for an AgrP projection, and hence no requirement on licensing conditions.8 The structures she proposes for the two types of language are shown in (13):
13) Strong Agr AgrP Agr Agr affix DP VP V Weak Agr AgrP DPi Agr ti Agr VP V + affix

Speas proposes that languages like Korean and Japanese do not have agreement morphology. However, I diverge from her and propose that Korean does have overt tense and agreement markers though they are weak. For a detailed discussion, see Section IV-2.
8

Hyeson Park

17

Based on Speass insight, Alexiadou and Anagnostopoulou (1998) develop a theory more compatible with the feature checking approach. Based on Chomsky (1995), they propose that the null subject phenomenon is closely related to checking of the EPP feature, which involves checking of the nominal D feature of Agr against the D feature of another nominal element such as subject. In English, which is a weak agreement language, the D feature of Agr is checked by merging a subject in the specifier position of Agr. On the other hand, in Spanish, which is a rich agreement language, the agreement morphemes of a verb are nominals with the D feature: hence, this nominal D feature of the verb checks the D feature of Agr when the verb moves to Agr via head-to-head raising. Thus, in this account, the EPP is universally strong across all languages, but the mechanism for checking the EPP feature is parameterized. It is either merger of XP or movement of X. The core idea of Alexiadou and Anagnostopoulous proposal relies on the fact that in Spanish the agreement morpheme of a verb is nominal, that is, it has semantic content. Hence, the feature is [+interpretable], while that of the English verb does not have any semantic content, which means English agreement features are [interpretable]. Results of some psycholinguistic studies are consistent with the hypothesis that the agreement morphemes in Spanish-type languages may have nominal [+interpretable] features. Vigliocco et al. (1996) studied subjectverb agreement errors by speakers of Spanish and English. They wanted to know whether semantic information plays a role in the production of subjectverb agreement in Spanish and English. They compared the relative incidence of subjectverb agreement errors in single token and multiple token preambles:
14) Single token: a. The key to the cabinet . . . b. The key to the cabinets . . .

15) Multiple token: a. The label on the bottle . . . b. The label on the bottles . . .

In (14b), the key can be associated with multiple cabinets, that is, the key refers to a single token; on the other hand, in (15b) the label represents multiple tokens, that is, one label per bottle. The head noun label is grammatically singular, but notionally plural. The one label per bottle reading is also called a distributive reading. The subjects were presented with the preambles and asked to complete the sentences as quickly as possible. The findings of the study show that:

18

A minimalist approach to null subjects and objects in SLA

1) The Spanish speakers made more agreement errors with the multiple token preambles, that is, they showed the distributivity effect.9 2) The English speakers did not show the distributivity effect, that is, there was no significant difference in agreement errors between the single token and the multiple token preambles. These results replicate the findings reported by Vigliocco et al. (1995) in Italian, and Bock and Miller (1991) in English. Based on Givons (1976) claim that agreement markers are reduced forms of pronominal elements, Vigliocco et al. (1996: 289) propose:
In terms of the processing model we outlined, features such as number would be retrieved from the conceptual representation in the construction of the verb phrase only if they represent pronominal features. Spanish as well as other Romance languages still preserve these features in the verb phrase, while they have been lost in the evolution of modern English. According to this very speculative hypothesis, the likelihood of semantic agreement would be related to the presence of pronominal features in the agreement target.

The results of the experiments that show that Spanish speakers may access the conceptual representation in the construction of the verb phrase are consistent with Alexiadou and Anagnostopoulous (1998) proposal; in Spanish, the agreement morphemes have semantic content due to their pronominal features, and they can check the EPP feature of Agr. On the other hand, the result that English speakers were insensitive to conceptual factors in constructing the verb phrases seems to be consistent with Alexiadou and Anagnostopoulous claim that in English-type languages, agreement features associated with the verb are not pronominal, and merging of the subject in the specifier of Agr is required in order to check the EPP feature. 2 Null subjects in Korean Alexiadou and Anagnostopoulous (1998) approach seems to be able to explain the difference between English-type and Spanishtype languages. However, they do not include Korean-type null subject languages in their analysis, and we cannot tell how their approach can be applied to Korean, which does not have rich agreement morphology but allows null subjects. I propose that the
The Spanish speakers produced more errors of the type The label on the bottles are, thus showing the distributivity effect.
9

Hyeson Park

19

distribution of null subjects in Korean is conditioned by both syntactic and discourse rules of grammar. First, at the syntactic level, Korean has a strong EPP feature and weak agreement features. Since a Korean verb does not have rich agreement morphology, the EPP is satisfied via XP-merge in the specifier of AgrP the same way as in English, but differently from Spanish, where EPP is satisfied by X-movement. My proposal that Korean has agreement features is based on the Korean honorific agreement system and subjectmood agreement markers. One typological characteristic of Korean is that it has a welldeveloped honorific system. A speaker conveys her respect or deference toward a referent using grammatical agreement morphemes: hence, a subject which has the [+hon(orable)] feature triggers the verbal agreement morpheme -si-:
16) sensayngnim-i osinta Teacher-NOM come- HON- PRES- DECL10 The teacher comes. 17) haksayng-i onta student-NOM come PRES DECL A student comes.

In (16) the subject, which denotes a respected person, triggers the honorific suffix -si-, while in (17), the subject does not trigger honorific agreement.The correct use of -si- also involves knowledge of grammatical person. When the subject is a first person, the use of -si- results in an ungrammatical sentence:
18) * na- nun osinta I- TOP come HON PRES DECL I am coming.

Thus, it appears that Korean has agreement features even though they are very limited. Cho (1994) argues that although the honorific morpheme -sihelps to identify null subjects, the information in the honorific marker is not rich enough to retrieve the content of null subjects. Cho proposes that mood categories in Korean agree with the subject in person and number, and thus license and identify null subjects. Some examples of mood categories are:

10

The abbreviations used for Korean data are: ACC: accusative case DECL: declarative HON: honorific PRRES : present tense TOP: topic marker

NOM : PST :

nominative case past tense

20

A minimalist approach to null subjects and objects in SLA

19) nol nol-ca no-sey nol-a-la nol-tela nontay nol-tekwun nol-key(na) play (stem) play-propositive, 1 pl play-propositive, 1 pl play-imperative, 2 sg/pl play-reportative, 2 sg/pl, 3 sg/pl play-reportative, 3sg/pl play-reportative, 2sg/pl, 3 sg/pl play-propositive, 2sg/pl

(Cho, 1994: 456)

The following examples show cases where disagreement between the subject and the mood marker results in ungrammaticality:
20) a. b. * * Yenghi-ka nol-ca. Yenghi-NOM play-propositive (1 pl) Let us (1 pl) Yenghi (3 sg) play. Ney-ka nontay. you-NOM play-reportative (3 sg/pl) You (2 sg) were reported to play (3 sg/pl).

(Cho, 1994: 456)

Cho further argues that null arguments in Korean are licensed in the same way as in Italian-type languages via Infl which has person and number information. Although I agree with Chos proposal that Korean does have agreement features, I diverge from her and propose that Korean agreement features are not as strong as those of Italian-type languages. The fact that the most common mood markers, declarative and question markers do not have person and agreement features indicates that Chos position is too strong. Thus, it appears that contrary to the widely accepted view that Korean does not have any agreement features, Korean does have some agreement features, although very weak ones. That is, Korean belongs to the weak agreement language group together with English, and the EPP feature in Korean is checked the same way as in English by merging the subject in the specifier position of AgrP. If both Korean and English check the EPP the same way, then the question is why Korean allows null subjects while English does not. I would propose that this is due to licensing conditions on NP at the pragmatic level. Previous studies of the distribution of null subjects in Korean-type languages have focused on discourse aspects. For instance, Huang (1984) proposes that empty categories in discourse-oriented languages such as Chinese, Japanese and Korean are identified by a null sentence topic, which in turn is linked to a discourse topic. The topic-identified null arguments are possible in Chinese, he claims, due to a rule of Topic NP Deletion, which operates across a discourse to delete the topics of sentences and form a topic chain.

Hyeson Park
21)

21

[Zhongguo, difang hen da.] [ e renkou hen duo.] [ e tudi hen feiwo.] China place very big population very many land very fertile [e qihou ye hen hao.] [ e women dou hen xihuan.] climate too very good we all very like (As for) China, (its) land area is very large. (Its) population is very big. (Its) land is very fertile. (Its) climate is also very good. We all like (it).

(Huang, 1984: 549)

In order to incorporate the Topic NP Deletion rule into an interpretive framework, Huang introduces an LF module after LF, which he proposes is active in discourse-oriented languages:
We may assume that there is a rule of coindexation, in the discourse grammar of a discourse-oriented language (in the LF module of grammar following LF), which coindexes an empty topic node with an appropriate preceding topic. On the other hand, a sentence-oriented language has a less substantive discourse grammar in that it lacks this topic-chain interpretation rule, among others. (Huang, 1984: 550)

Contrary to Huang, I would like to argue that the LF module, that is, the pragmatic module is active both in discourse-oriented languages and in sentence-oriented languages. The topic chain is formed in both types of languages, and the difference among languages lies in whether they choose overt pronominal or zero anaphora to refer to the topic. According to Givon (1983a), the choice of anaphoric devices follows a scale that reflects the discourse pattern of topic continuity:
22) Most continous/accessible topic Zero anaphora Unstressed/bound pronouns (agreement) Stressed/independent pronouns Full NPs Most discontinuous/inaccessible (Givon, 1983a: 18)

Korean-type languages choose zero anaphora in this scale to refer to the topic, while English-type languages usually select independent pronouns to encode a topic-referring NP. Spanish uses bound pronouns (grammatical agreement) for the same purpose (Bentivoglio, 1983). However, it is not ungrammatical to use zero anaphora to refer to the topic in English or overt pronouns in Korean. For example, Givons (1983b) analysis of spoken English text shows that both overt pronouns and null pronouns are used to refer to the topic.11 Thus, we propose that all languages have topic
Givon (1983b) analysed the transcript of a narrative of a New Mexico man in order to examine topic marking devices in spoken English. According to his analysis, pronouns constituted about 57% of the topic marking devices used by the subject (423 out of 742) and null pronouns about 16% (117 out of 742).
11

22

A minimalist approach to null subjects and objects in SLA

chains formed at the pragmatic level. The only difference between languages is whether they prefer zero anaphora, overt pronouns or bound pronouns to code topic-referring NPs. 3 Null subjects in L2 English Now we return to the original question raised at the beginning of this study. Most studies of the acquisition of subjects in L2 English have shown that Spanish-type L1 speakers typically drop a noticeable number of subjects at the early stage, while Korean-type L1 speakers do not (White, 1985; Hilles, 1986; 1991; Phinney, 1987; Lakshmanan, 1991; 1994; Zobl, 1994; Yuan, 1997; Wakabayashi, 2002). I propose to explore an answer to this asymmetry at both the syntactic and pragmatic levels. Table 5 summarizes the licensing conditions of null subjects in English, Spanish and Korean at the syntactic and pragmatic levels of grammar. First, we consider the pragmatic level. As discussed in the preceding section, topic-referring NPs in English are usually expressed by overt pronouns. Thus, Korean learners of English have to learn that topic-referring NPs, which are encoded by zero anaphora in their L1, are encoded by overt pronouns in L2 English. On the other hand, Spanish learners of English have to realize that they can no longer encode topic-referring NPs with bound pronouns. The question is which of these two cases involves a more difficult process: learning of overt pronouns by bound pronoun L1 speakers vs. learning of overt pronouns by zero anaphora L1 speakers. Since it seems impossible to distinguish between these two processes in degree of difficulty, we look for an answer to the asymmetry between Korean and Spanish learners at the syntactic level. At first sight, this asymmetry seems difficult to explain if we consider only the strength of the EPP feature, since all languages are hypothesized to have the same strong EPP. However, the way the EPP requirement is satisfied is different between English and Spanish, but the same between English and Korean: XP-merge in Korean and English and X-movement in Spanish. This similarity between Korean and English on the one hand and difference
Table 5 Licensing conditions on null subjects Pragmatic level (topic-referring NP) English Spanish Korean pronoun bound pronoun zero anaphora Syntactic level

[interpretable]; XP-merge [+interpretable]; X-movement [interpretable]; XP-merge

Hyeson Park

23

between English and Spanish on the other result in the asymmetry between Korean and Spanish learners of English in their acquisition of English subjects. Korean learners do not have to learn the [interpretable] feature value of the agreement morphology in L2 English, while Spanish learners have to learn that English has [interpretable] agreement features, which requires XP-merge, in contrast to the X-movement in their L1. 4 Null objects Huang (1989) is one of the few studies that has examined the distribution of null objects. Huang distinguishes between two different types of null objects: variables and pros. He predicts that null objects are possible in languages with rich object agreement morphemes like Pashto, and these null objects are pros coindexed with object Agr. On the other hand, null objects in languages with no agreement morphemes are variables bound by the discourse topic. Thus, according to Huang, null objects in Korean are variables licensed at the pragmatic level. However, contrary to Huangs analysis, it has been observed that languages like Korean and Thai, which do not have object Agr, allow pro in the object position, and this null object is bound by the main clause subject (cf. Cole, 1987; Speas, 1996). For example in Korean, a null object in the embedded clause can be bound by the main clause subject:
23) a. b. Chelswu1-ka [Yenghi-ka Chelswu-NOM Yenghi-NOM Chelswu1-ka [Yenghi-ka Chelswu-NOM Yenghi-NOM Chelswu1 thinks that Yenghi ku1-lul po-ass-ta]-ko he-ACC see-pst-DECL]-COMP e1 po-ass-ta]-ko (him) see-PST-DECL]-comp saw him1. sayngkak-hanta. think sayngkak-hanta. think

In this case, the null object is not a variable that refers to discourse topic, but a pronominal that is bound by the main clause subject. This kind of omission is possible when the content of the object is more or less accessible due to the presence of its antecedent in the main clause. I would propose that null objects in Korean are licensed at two different levels: syntax and pragmatics. In English and Spanish, null objects are not allowed at the pragmatic level since topic-referring NPs in these languages are encoded by overt and bound pronominals, respectively. Speas (1994; 1996) tries to find the licensing conditions for null objects within the syntactic module. Adopting the proposal of Rohrbacher (1992) who claims that in languages with strong Agr agreement morphemes have their own lexical entry, while in languages with weak Agr agreement morphemes do not have independent lexical entries Speas tries to connect the occurrence

24

A minimalist approach to null subjects and objects in SLA

of null objects with the projection of the VP-shell. In Larson (1988) and Hale and Keysers (1991) original theory of the VP-shell, the verb is generated in the V1 position and moves to V2, forming the surface structure in (24).
24) V2 put j V VP V V1 tj PP on the shelves

NP the books

Speas (1996) diverges from this view, and suggests that in some languages a verb is base-generated in the V2 position and controls the V1 position. In this case the specifier of the lower VP has to be filled in order for the lower V to be projected, following the principle of economy of representation. In languages in which a verb is generated in the V1 position and moves to the V2 position, the specifier of the lower VP can be empty since V1 is filled. At first sight this approach seems plausible. In Korean, it is possible to say that a verb moves from V1 to V2. Adopting Chomsky (1995), I assume the following structure for Korean:
25) AgrP Agr vP v VP obj V v Agr

In Korean, v is the position where the light verb ha-ta and the causative morphemes i, hi, li, and ki are generated. At the syntactic level, the lower verb V moves to v and forms a Vv compound. For example, the verbal noun kongpwu in the V position moves to v and merges with v hata and becomes a compound verb kongpwu-hata (to study). In the morphological causative construction, an intransitive verb like wul-ta (to cry) is generated in the V position, and a causative morpheme -li- is in the v position. The verb moves to the v position where it combines with the causative morpheme li- and becomes a causative verb wul-li-ta (to make someone cry). According to Speas theory, in Korean the object position can be

Hyeson Park

25

null; the head of the lower VP is always filled by a verb, and thus the VP can be projected either with or without an object in the specifier position, without violating the principle of economy of representation. Thus, Korean is a null object language. In English there is no explicit evidence that the lower verb moves to the light verb v in syntax. The conflation of V + v appears to be a lexical process. For example, the verb chwuk-i-ta (to kill) in Korean is a conflation of an intransitive verb chwuk-ta (to die) and a causative morpheme -i-, which occurs in syntax. In English kill must be generated as one word in v, and the lower V is controlled by v. Since the head of the lower VP is empty, [spec, VP] cannot be null or there would be a violation of the principle of economy of representation. This approach, however, cannot account for the lexical variation with regard to the possibility of null objects in English. Some verbs in English allow a null object, while some obligatorily require an object, as can be seen in the following contrast:
26) a. b. c. d. John already ate. * John devoured. They ran away and we followed (them). They ran away and we chased *(them).

While eat and follow allow a null object, devour and chase, which are similar in meaning, do not. If we adopt Speas analysis, we have to assume that eat is generated in the lower verb position and moves to v, while devour is generated in v and controls the lower verb. This seems very arbitrary and stipulative. It may be better to explain licensing of null objects in English as a lexically controlled process rather than a syntactic process, as has been argued by many researchers (see Lehrer, 1970; Allerton, 1975; Fillmore, 1986; Rice, 1988). For instance, Cote (1996) claims that the occurrence of null objects in English is a lexically constrained phenomenon; that is, null objects are implicit arguments listed in the lexicon as a subcategorization specification of a verb. Null objects in most Romance languages also seem to be a lexically constrained phenomenon, which are allowed when an object has an arbitrary or an indefinite interpretation. For example, a verb like comprar (to buy) subcategorizes for an NP, as in (27). But comprar can be used with a null object when the object is interpreted as indefinite, as in (28):
27) a. Compr un/el libro. I bought a book. b. Lo compr it I bought I bought it. c. * Compr. I bought.

26

A minimalist approach to null subjects and objects in SLA

28) a. Compraste caf? Did you buy coffee? b. Si, compr Yes, I bought (some)

(Campos, 1987)

Since the null object in these languages is basically a lexically controlled phenomenon, a null object is possible only when a verb subcategorizes for it. In the P&P approach, the selection of the object by the verb, i.e., subcategorization, is viewed as a thematic relationship involving theta-role assignment by a verb to its object. Within the MP approach, theta-roles are considered to be formal features and theta-role assignment is reinterpreted as checking of theta-features (Lasnik, 1999). Boskovic and Takahashi (1998) apply the feature checking approach to theta-role assignment in their research on Japanese scrambling.12 Scrambling, which involves free word order, is very common in Japanese (29) and Korean (30).
29) a. John-ga [Mary-ga sono hon-o katta-to] omotteiru. John-NOM Mary-NOM that book-ACC bought-COMP thinks John thinks that Mary bought that book. b. Sono hon-o John-ga [Mary-ga t katta-to] omotteiru. a. John-i [Mary-ka ku chayk-ul sassta-ko] sayngkakhanta. John-NOM Mary-NOM that book-ACC bought-COMP thinks John thinks that Mary bought that book. b. ku chayk-ul John-i [Mary-ka t sassta-ko] sayngkakhanta.

30)

A traditional analysis of scrambling is that (29a) and (30a) represent the canonical word order, and in (29b) and (30b) the object of the embedded clause the book moves and adjoins to the main clause (IP-adjunction). Under this analysis, scrambling is an optional movement without any semantic effects. Optional movement, however, is not compatible with the spirit of the MP, in which all movements must have some motivation. That is, movement is not economical, hence is chosen as the Last Resort to check formal features which need to be checked off in order for a derivation to converge. In order to solve the problem of optionality with scrambling, Boskovic and Takahashi (1998) propose that the scrambled object is base-generated in its dislocated position, that is, in an IP-adjoined position, and at LF it moves down to the complement position where it can get its theta-feature checked by the verb. Within the P&P model, which has both D-structure and S-structure, theta-roles are assigned at D-structure. However, within MP which does not have D- and S-structure any more thetaroles can be assigned (= checked) at LF.
For arguments against Boskovic and Takahashi (1998), see Bailyn, 2001. I would like to thank an anonymous reviewer who referred me to this article.
12

Hyeson Park
Table 6 Licensing conditions on null objects Pragmatic level (topic-referring NP) English Korean pronoun zero anaphora Syntactic level

27

strong theta-feature weak theta-feature

In order to explain the difference between scrambling and nonscrambling languages like English, Boskovic and Takahashi propose a parameter based on the strength of the theta-features. In English, the theta-feature is strong and must be checked before Spell-out. Thus, the object must be adjacent to the verb so that the theta-feature of the verb and that of the object can be checked against each other. In Japanese-type languages, the theta-feature is weak, and checking of theta-features can be postponed until after Spell-out, and hence verb and object do not need to be adjacent. Extending Boskovic and Takahashis hypothesis to the null object phenomenon, I propose that the parameterized theta-features are responsible for null objects in Korean and overt objects in English. Since Korean verbs have weak theta-features, which do not need checking before LF, null objects are possible at the syntactic level. Null objects in Korean are, then, the result of two interacting factors: weak theta-features at the syntactic level and encoding of topic-referring NPs by zero anaphora at the pragmatic level. Obligatory objects in English, on the other hand, are due to the strong theta-feature at the syntactic level and use of overt pronouns to encode topic-referring NPs at the pragmatic level.13 For those English verbs that allow null objects, I would assume that their theta-features are also strong since the null objects are subcategorized by the verb. Table 6 summarizes the licensing conditions of null objects in English and Korean at the syntactic and pragmatic levels.
There is a case where topic-referring object drop is allowed in English, namely, the recipe genre. Haegeman (1987) suggests that the diverse registers and dialects of a language are a form of language internal parametric variation. Thus, according to her, the register variation of the recipe is a form of variation mirroring the contrast between topic prominent discourse oriented language (TP) and subject prominent sentence oriented language (SP). English is considered to be a SP language, and null objects are not allowed. However, Haegeman argues that English speakers have both TP and SP settings; they generally turn on the SP setting, but when they access recipes, they turn off the SP setting and turn on the TP setting. Culy (1996), on the other hand, argues that there is only one grammar, and the null objects in English recipes belong to the language use module, that is, pragmatics. Culys approach is based on the assumption that pragmatics is not part of core grammar. However, if we adopt the view that the pragmatic level is a module of linguistic competence, we can say that topicreferring objects can be deleted in English, but the domain in which this rule applies is restricted.
13

28

A minimalist approach to null subjects and objects in SLA

Now let us consider our original question: why do Korean learners of English drop more objects than subjects in their L2 English? The explanation for this asymmetry has to be sought at the syntactic level since the distribution of subjects and that of objects are dictated by the same condition at the pragmatic level: in English, both subjects and objects are encoded by overt pronominals when they refer to the discourse topic, while in Korean zero anaphora is used for the same purpose. Thus, at the pragmatic level, what Korean learners of English have to learn is the same for both subject and object: zero anaphora is not allowed for topicreferring NPs whether subject or object. At the syntactic level, however, learning of subject and of object involve different conditions: while both English and Korean have the same [interpretable] agreement features, Korean and English have different theta-features: strong in English and weak in Korean. Korean learners may find it difficult to learn the different thetafeatures and the lexicon specific subcategorization information in English, leading to the extended null object period. V Summary and conclusions Studies of null arguments in L2 English have found that: 1) Spanish-type L1 learners drop subjects in the early stages of learning English as an L2, while Korean-type L1 learners do not drop many subjects even in the earliest stages. 2) Korean-type L1 speakers drop more objects than they drop subjects in their developing English grammars. In this article, I have explored a possible analysis of these two asymmetries within the Minimalist Program. In the MP approach, language variation is hypothesized to be limited to the properties of formal features, and hence L1 acquisition involves learning of these formal features. I have proposed that not only L1 but also L2 acquisition involve learning of formal features of a target language. The formal features involved in allowing null arguments are agreement features for subjects and theta-features for objects. I have argued that languages with [+interpretable] agreement features like Spanish allow null subjects while languages with [interpretable] agreement features such as English and Korean do not. In the case of objects, I proposed that languages with strong theta-features do not allow null objects since the formal thetafeatures of verbs need to be checked against those of objects before Spell-out. Languages like Korean and Japanese have weak thetafeatures, thus feature checking can be postponed until after Spell-

Hyeson Park

29

out. Due to the weak theta-features, null objects and scrambling are allowed in these languages. Since Korean and English have the same [interpretable] agreement features, it is not difficult for Korean learners of English to learn English subjects. However, Korean and English differ from each other in the strength of thetafeatures, and it takes time for Korean learners, whose L1 has weak theta-features, to learn the strong theta-features of English. The analysis that attributes the observed asymmetries in the distribution of null arguments in L2 English to the properties of formal features provides support for the core idea in the MP. VI References
Alexiadou, A. and Anagnostopoulou, E. 1998: Parameterizing Agr: word order, V-movement and EPP-checking. Natural Language and Linguistic Theory 16, 491539. Allerton, D.J. 1975: Deletion and proform reduction. Journal of Linguistics 11, 21337. Bailyn, J.F. 2001: On scrambling: a reply to Boskovic and Takahashi. Linguistic Inquiry 32, 63558. Bentivoglio, P. 1983: Topic continuity and discontinuity in discourse: a study of spoken Latin-American Spanish. In Givon, T., editor, Topic continuity in discourse: a quantitative cross-language study, Amsterdam: John Benjamins, 255311. Bley-Vroman, R. 1989: What is the logical problem of foreign language learning? In Gass, S. and Schachter, J., editors, Linguistic perspectives on second language acquisition, Cambridge: Cambridge University Press, 4168. Bock, J.K. and Miller, C.A. 1991: Broken agreement. Cognitive Psychology 23, 4593. Boskovic Z. and Takahashi, D. 1998: Scrambling and last resort. Linguistic N , Inquiry 29, 34766. Brown, R. 1973: A first language: the early stages. Cambridge, MA: Harvard University Press. Campos, H. 1987: Indefinite object drop. Linguistic Inquiry 18, 35459. Cho, S.-W. 1994: The grammar of null arguments in early child Korean. In Kim-Renaud, Y.-K., editor, Theoretical issues in Korean linguistics, Stanford, CA: CSLI Publications, 44362. Chomsky, N. 1986: Knowledge of language: its nature, origin and use. New York: Praeger. 1995: The Minimalist Program. Cambridge, MA: MIT Press. 2001: Derivation by phases. In Kenstowicz, M., editor, Ken Hale: a life in language, Cambridge, MA: MIT Press, 152. Cole, P. 1987: Null objects in universal grammar. Linguistic Inquiry 18, 597612. Cote, S.-A. 1996: Grammatical and discourse properties of null arguments

30

A minimalist approach to null subjects and objects in SLA

in English. Unpublished PhD Dissertation, University of Pennsylvania, Philadelphia: PA. Culy, C. 1996: Null objects in English recipes. Language Variation and Change 8, 91124. Diesing, M. 1992: Indefinites. Cambridge: MA, MIT Press. Fillmore, C.J. 1986: Pragmatically controlled zero anaphora. Proceedings of Berkeley Linguistics Society 12, 95107. Givon, T. 1983a: Topic continuity in discourse: an introduction. In Givon, T., editor, Topic continuity in discourse: a quantitative cross-language study, Amsterdam: John Benjamins, 142. 1983b: Topic continuity in spoken English. In Givon, T., editor, Topic continuity in discourse: a quantitative cross-language study, Amsterdam: John Benjamins, 34364. Haegeman, L. 1987: Register variation in English: some theoretical observations. Journal of English Linguistics 20, 23048. Hale, K. and Keyser, S. 1991: On the syntax of argument structure. MIT Lexicon Project Working Papers 34. Cambridge, MA: MIT Press. Hilles, S. 1986: Interlanguage and the pro-drop parameter. Second Language Research 2, 3352. 1991: Access to universal grammar in second language acquisition. In Eubank, L., editor, Point counterpoint: universal grammar in the Second Language, Amsterdam: John Benjamins, 30538. Huang, J. 1984: On the distribution and reference of empty pronouns. Linguistic Inquiry 15, 53174. 1989: Pro-drop in Chinese. In Jaeggli, O. and Safir, K., editors, The null subject parameter, Dordrecht: Kluwer Academic, 185214. Jaeggli, O. and Safir, K. 1989: The null subject parameter. Dordrecht: Kluwer Academic. Kim, K. and Hong, N.-S. 1982: Baseline report for the KoreanEnglish longitudinal study. National Center for Bilingual Research, Los Alamitos, CA. ED 234 634. Lakshmanan, U. 1991: Morphological uniformity and null subjects in child second language acquisition. In Eubank, L., editor, Point counterpoint: universal grammar in the second language, Amsterdam: John Benjamins, 389411. 1994: Universal grammar in child second language acquisition. Amsterdam: John Benjamins. Larson, R. 1988: On the double object construction. Linguistic Inquiry 19, 33592. Lasnik, H. 1999: Minimalist analysis. Chapter 7. Oxford: Blackwell. Lehrer, A. 1970: Verbs and deletable objects. Lingua 25, 22753. Levin, B. 1993: English verb classes and alternations: a preliminary investigation. Chicago, IL: University of Chicago Press. Phinney, M. 1987: The Pro-drop parameter in second language acquisition. In Roeper, T. and Williams, E., editors, Parameter setting, Dordrecht: Reidel, 22138. Rice, S. 1988: Unlikely lexical entries. Proceedings of Berkeley Linguistic Society 14, 20212.

Hyeson Park

31

Rizzi, L. 1986: Null objects in Italian and the theory of pro. Linguistic Inquiry 17, 50157. Speas, M. 1994: Null arguments in a theory of economy of projection. In Benedicto, E. and Runner, J., editors, University of Massachusetts Occasional Papers in Linguistics 17. Amherst, MA. 1996: Null objects in functional projections. In Rooryck, J. and Zaring, L., editors, Phrase structure and the lexicon, Dordrecht: Kluwer Academic, 187211. Vigliocco, G., Butterworth, B. and Semenza, C. 1995: Constructing subjectverb agreement in speech: the role of semantic and morphological factors. Journal of Memory and Language 34, 186215. Vigliocco, G., Butterworth, B. and Garrett, M. 1996: Subjectverb agreement in Spanish and English: differences in the role of conceptual constraints. Cognition 61, 26198. Wakabayashi, S. 2002: The acquisition of nonnull subjects in English: a minimalist account. Second Language Research 18, 2871. White, L. 1985: The pro-drop parameter in adult second language acquisition. Language Learning 35, 4762. Yuan, B. 1997: Asymmetry of null subjects and null objects in Chinese speakers L2 English. Studies in Second Language Acquisition 19, 46797. Zobl, H. 1994: Prior linguistic knowledge and the conservation of the learning procedure: grammaticality judgements of unilingual and multilingual learners. In Gass, S. and Selinker, L., editors, Language transfer in language learning, Amsterdam: John Benjamins, 17696.

Appendix 1
JH:

Examples of null argument sentences for each subject

__ get this chalk __ do up here __ play with He make __. I think __. Hes kicked __. Hes giving __ to bird. Bird see he, um stolding __. __ go to ambulance __ pay the money I am putting __ right here The boy want to climb up __and get the cat He got food and stole __. I pick that up and give her __. The birds got food, and theyre sharing __. She took off this and she is hanging __ up there. __ keep talking to you. Its my cousin, __ have a mustache. __study Bibles and sings, then free time, then __ come in.

Null subjects: Null objects:

YJ:

Null subjects: Null objects:

KH: Null subjects:

32

A minimalist approach to null subjects and objects in SLA

Yes, I said ___. He give ___ to you? She did ___ one time. She threw __ into the basket. His mother had a bandage and put ___ on his leg. Thats why she give him ____. __ call her mother. He put the money in and then he push __. And then Coca Cola come and then he drank __. And he put __ in the trash can. Dont take __. Can you hear __ now? They dont want to have __ together. We bought __ at Korea, too. He didnt hold __ well. We had . . . just me . . . invite friends and __ came, __ eat the cake. Because __ isnt comfortable for me If __ hide in all the places in . . . at outside Dont use that. __ gonna break Could you say __ in Korean? Oh, look at __. Putting the ice cream and he gave__ to the man. She paid the money and get __ and she gave her change. My teacher bought __ for me. She just made __ up. How come you are not making __? She wasnt going to sharpen __ like that. They watch __ and then they tell them . . . __ give __ medicine. __ find the thing and then, say a word. __ drop in the floor. Dont give __ to girl. I dont like __. Lets go over there and wait for __. Because he tried to steal __. He cant blow __. Can you see through __. He was try to found __. I just do __ by myself. She sneak a banana and eating __.

Null objects:

SM: Null subjects: Null objects:

SE:

Null subjects:

Null objects:

JA:

Null subject: Null objects:

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.