Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

1

The processing of object anaphora in Brazilian Portuguese* Marcus Maia Federal University of Rio de Janeiro - Brasil [email protected]

1. Introduction The comprehension of anaphoric relations is an essential feature of human sentence processing. The basic processing issues concerning coreference assignment revolve around the questions of if, when and how

the correct coreferential relationships are established among

elements in a sentence. Typically the focus of the research has centered on the ability of referentially-dependent elements such as overt pronouns and empty categories to facilitate the comprehension of a previously mentioned noun phrase (NP), or antecedent. A number of recent studies in Psycholinguistics have shown the processing relevance of empty categories and pronouns in English. That is to say, these elements have been shown to be psychologically real in the sense that they trigger a reactivation1 of their antecedents. Chang

(1980)

conducted

experiments

that

demonstrated

that

recognition responses for a person's name were significantly faster *

This article is based on research I conducted for my doctoral dissertation (Maia, 1994) at the Psycholinguistics lab in the department of linguistics of the University of Southern California - USC. I am indebted to Maryellen MacDonald, Joseph Aoun, Mark Seidenberg and Maria Luiza Zubizarreta for helpful comments and suggestions. The material in section 3 of the article was presented during the 6th Annual CUNY Sentence Processing Conference at the University of Massachusetts (Amherst) in1993. I am grateful to that audience for many insightful questions. I would also like to thank Zlatka Guentcheva for having encouraged me to submit a first draft of the paper to RLV and the two RLV anonymous reviewers who have made important suggestions concerning the form and the content of the article. Of course, I am the sole responsible for any remaining mistakes.

1 According to Fodor (1989) "reactivation" may be a matter of excitation of the relevant entry in the mental lexicon or of the relevant concept in the semantic representation being constructed for the sentence.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

2

when a final clause contained a pronoun referring to the person than when it did not. Cowart and Cairns (1987) showed that the initial assignment of an antecedent to a pronoun obeys structural constraints, but not constraints on semantic or pragmatic well-formedness. Their experiments suggests that a pronoun triggers all and only those prior referents that are structurally appropriate as the antecedent. They argue that there is a device that computes coreference, and that this device has access to structural information, but not to what is often considered to be "higher level" information, such as semantics and pragmatics.

Several studies have also been conducted in order to assess whether similar effects could be established for antecedents of different types of empty categories. For example, Bever & McElree (1988) have found evidence that gaps access their antecedents during comprehension in the same way as overt pronouns and that gaps produced through movement access their antecedents more strongly than

the null

pronominal PRO. MacDonald (1989) conducted a study in which NPtraces are shown to unequivocally prime their antecedents in passive constructions in English. Gap and non-gap conditions differed by only one word, ruling out the possibility that differences in reaction times could be attributed to differences in the processing loads.

Besides answering the question concerning the perceptual reality of coreference assignment, the research on the processing of gaps has also been particularly informative with respect to the questions of decision principles and constraints on information use in the

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

3

comprehension of sentences. As a matter of fact much of the recent research on sentence understanding has focused on identifying the types of information available to the sentence processor or

parser

during comprehension. Two views about the type of information accessed during comprehension have dominated the literature. According to one position, the parser rapidly builds a syntactic skeleton by making use of phrase structure rules and simple decision principles such as the minimal attachment principle, by means of which "incoming material is attached into the phrase marker being constructed using the fewest nodes consistent with the wellformedness rules of the language" (Frazier, 1979, p.76). As in such a system syntactic category information is the primary input and other sources of information are basically ignored, many mistakes usually arise in the form of the so-called "garden-path" phenomenon. On this view,

combinatory lexical information

such as subcategorization

information can only be used in reanalysis after the parser's original analysis fails (Ferreira & Clifton, 1986). According to Frazier et al. (1983), for example, subcategorization information is not available to the parser in the initial stages of syntactic processing. In keeping with the idea of a parsing system in which syntactic category information constitutes the primary source of information, Clifton & Frazier (1989) also proposed the "active filler hypothesis", according to which a gap would be postulated on the basis of antecedent information well before the verb subcategorization could be accessed.

Clifton & Frazier

propose that as long as it has an unassigned filler in a nonargument position, the parser prefers to posit a gap for the filler rather than take into consideration the lexical properties of the head of the phrase.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

4

If subcategorization information about the verb is inconsistent with the postulation of the gap, the hypothesized gap is deleted.

Another view proposes that the comprehension system rapidly and optimally integrates lexical, syntactic and contextual information in some form of mental representation. In such a system, the immediate access to argument and control structure information would allow the parser to project structure and avoid indeterminacy (Crain & Steedman, 1985). Thus, the question whether lexical information can be accessed in the initial stages of the parsing process is far from settled, since several types of on-line processing evidence have made a strong case that combinatory lexical information is immediately used in parsing (e.g. Stowe, 1989; Tannenhaus, Stowe & Carlson, 1985; Tannenhaus & Carlson, 1989.) In sum, several studies of filler-gap sentences provide evidence that the language processor has rapid access to both argument structure and control information. Such results complement an increasing body of research demonstrating that lexical information may play an immediate role in guiding parsing decisions.

In this article, we present evidence in favor of the processing relevance of gaps which refer to topics (A'-bound gaps) in Brazilian Portuguese (BP). As it will be argued in section 3, our data also seem to confirm that subcategorization information is indeed accessed by the parser at early stages of comprehension.

The psycholinguistic experiment which is reported in section 3 has also an important bearing on the issue of topic availability or identification

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

5

which has been investigated in the framework of functional grammar. Within this area of linguistic study clauses have traditionally been thought of as made up by two distinct components: the comment or "rheme" which carries the new information and the topic or "theme" which is usually identified with the old information. authors have

Additionally,

also investigated the degree of difficulty that

speakers/hearers may experience in identifying a topic, that is, in filing it appropriately in their internal register, so that the new information transmitted about the topic would in turn be addressed adequately. In Givón (1983) several studies are reported which have attempted to establish a scale in the coding of topic accessibility. By analyzing texts in different languages, authors have endeavored to assess the grammatical devices used by the speaker to code various topics in the discourse. According to Givón, a scale of crosslinguistic coding devices may be used to indicate topic continuity in discourse. These studies have been couched in terms of the speaker-hearer neutral notion of discourse continuity, rather than in terms of the more psychologically oriented notion of accessibility. The assumptions underlying these studies, however, have a psychological import: what is continuing is more predictable; what is predictable is easier to process. However, as these measurements were performed on texts rather than on speakers or hearers, they could not tap into the issue of topic identifiability in a direct way, since they could not reveal the ease or difficulty hearers experience in processing and filing topics in discourse. Therefore, these studies lacked empirical respect, Givón (1983) states :

psychological justification.

In this

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

6

"Hopefully, once stable, strong and cross-linguistically viable correlations are established (...), one may proceed to the obvious next step, that of correlating the grammatical and discourse-distribution data with psycholinguistic experimentation and measurement." (Givón, 1983, p. 13). In this article, we provide psycholinguistic evidence which takes the research on topic accessibility to the "obvious next step" proposed by Givón. One of the findings of the experiment reported in the next section is that the zero-anaphora in object position exhibit a stronger psychological reactivation of topics than the overt lexical pronoun in the same position. These results are entirely in line with the crosslinguistic text measurements performed by Givón et alii (1983), and provide partial psychological justification for Givón's scale of topic accessibility.

This article is organized as follows. In section 2, a general overview on the relevant aspects of the grammar of BP is provided. In section 3, we present experiment 1 which compares the reactivation properties of gaps and overt pronouns in a probe recognition task, and discuss the relevance of the BP facts for the architecture of the human parser. In section 4, we discuss our experimental results in the light of the Overt Pronoun Constraint

proposed by Montalbetti (1984)

and

present experiment 2, which compares the strict/sloppy interpretation of overt and nonovert objects in a speeded grammaticality judgment task. Finally, in section 5, we summarize the general conclusions of the paper.

2. BP : an overview

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

7

Brazilian Portuguese is a pro-drop language with an SVO basic word order. It has a rich verbal inflection system which indicates person, number, tense and mood by means of suffixes. As other Romance languages such as Italian, Spanish, and Catalan, it allows missing subjects. Unlike Italian and like Chinese or Japanese ( cf. Huang, 1984; Hasegawa, 1985) , it allows missing objects, even though there is no object morphological agreement. Although BP also has a paradigm of object pronouns or clitics which is similar to the system generally found in the Romance languages, it is a well known fact that clitics are disappearing in oral BP and even in informal written BP. Tarallo (1984) analyzed 45 hours of recorded data in which no third person clitics were observed whatsoever and the only third person pronouns in object position were the full lexical pronouns. Similarly, Duarte (1989) has also shown in her sociolinguistic study on the use of the accusative clitic, the lexical pronoun and the null object in colloquial BP, that the null object occurs in 62.6% of the utterances in her 40 hour corpus, whereas the clitic appears in only 4.9 % of the sentences, the full nominative pronoun in 15.4 %

and anaphoric NPs in 17.1% of

the cases. Therefore, although BP has a system of object clitics which is still productive in formal registers, BP speakers generally prefer to resort

either to the gap or to the nominative lexical pronoun, as

exemplified in (2) and (3), avoiding clitic constructions such as (4): (1) Você viu o João i ? "Did you see João ? " (2) - Vi [e]i ontem no clube. saw [e]i yesterday in the club "I saw (him) yesterday in the club."

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

8

(3) - Vi ele i ontem no clube. saw he i yesterday in the club "I saw him yesterday in the club." (4) - Vi-oi ontem no clube. saw him i yesterday in the club "I saw him yesterday in the club."

This ability to drop objects, which approaches BP from so called "cool" languages2 such as Chinese, is not shared to the same extent by any other Romance language, including European Portuguese (EP). In EP null objects are not so frequent and alternate only with the clitic. Sentences such as (3) above, as Galves (1989) points out, would be absolutely ungrammatical in EP, what suggests that the underlying structure of (2) is different in the two dialects of Portuguese.

Another important feature of BP which is directly relevant for the purposes of this paper concerns the distribution of topic constructions. Topic structures are so frequent in BP that Pontes (1987) proposed that BP must be classified as either a topic prominent type of language or at least as a language in which both the notions of subject and topic are equally prominent. Additionally, in her important study on topics in BP, Pontes (1987) shows that topic structures in which the comment clause does not contain an element which is anaphorically related to 2

Following Ross (1982), Huang (1984) suggests that languages may be classified as hot, medium or cool, on the basis of the degree of explicitness with which they express certain anaphoric elements. English would be a "hot" language because pronouns cannot in general be omitted from grammatical sentences. Spanish would be a "medium" type of language because it allows the deletion of subject pronouns but not of object clitics. Chinese as well as Brazilian Portuguese would be examples of "cool" languages since pronouns are usually omissible from grammatical sentences.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

9

the element to the left of S are very common in BP. This is what Chafe (1976) calls "Chinese style" topic structure.

Typical examples are (5)

and (6): (5) Aquele fogo, ainda bem que o corpo de bombeiro chegou. “That fire-i, fortunately the fire brigade arrived.” (6) Peixe, Dourado é o melhor para mim. “Fish, Red snapper is the best for me.”

As we will show in section 3, topic-comment constructions in BP are not only frequent, but they also seem to be processed as readily as the subject-predicate type. We also present

experimental data which

seem to suggest that there may be structural and processing constraints underlying the preference for the empty category in object position in BP topic constructions. Our data seem to confirm findings by Teixeira (1985), Pontes (1987) and Callou et al (93)

which indicate that

resumptive pronouns in object position are not productive in BP. We show that

these elements do not exhibit the same

reactivation

properties as topic-bound gaps and a parsing explanation is suggested: in languages which have the option between

overt and empty

elements, a default strategy is operative so that overt elements check for their possible antecedents within the sentence as a first resort, whereas gaps can establish coreference with elements outside S more directly.

3. Experiment 1:

priming effects of gaps and overt pronouns in

object position in BP

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

10

This

experiment compared the processing of overt pronouns and

empty categories in object position in Brazilian Portuguese in structures in which these anaphors are A-bound by a subject and in structures in which they are A-bar-bound by an antecedent in topic position.

3.1. Method In this section we describe the participants, the materials and design as well as the testing procedures used in the experiment.

3.1.1. Subjects Forty-eight

(48)

USC,

UCLA

and

UC

Berkeley

brazilian

undergraduate and graduate students participated in this experiment. All were native speakers of brazilian portuguese with normal or corrected vision and normal hearing.

3.1.2. Materials and Design The stimuli were 12 sets of

60 sentences (see appendix 1 for the

complete lists). Each subject was presented one of these experimental sets embedded in an extra set of 60 filler sentences. Each experimental set was made up of 6 experimental conditions and 6 control conditions with five sentences per condition in a 3x2x2 type of design. There were two levels of matrix (topic/subject), three levels of subordinate clauses( gapless, overt pronoun, empty category), and two levels of probe (antecedent, other).

Table 1 provides

an example of each of the

experimental conditions tested in the experiment. The control conditions were the same as the experimental conditions except that a

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

11

probe other than the antecedent of the overt pronoun or gap was presented. This other probe was the subject of the matrix clause in the case of the topic sentences and the subject of the embedded clause in the case of the subject sentences. Notice that although sentences were very similar, differing only in terms of the tested variables, they were distributed between subjects, so that each subject saw only one version of each sentence type. The distribution of sentence types in 12 sets allowed all sentences of a

type to be compared.

To

provide a

counterbalance for the extra NP required in the topic sentences, all subject sentences had an extra PP in the matrix clause in order to try to rule out the possibility that caused by

differences in reaction times could be

different processing loads rather than by the experimental

factors . Table 1

(7) Subject-bound null object A moradorai disse agora mesmo na entrevista à televisão que os bombeiros já estão ajudando [e] i. "The resident said right now in the interview to the television that the firemen are already helping"

(8) Subject-bound overt pronoun A moradorai disse agora mesmo na entrevista à televisão que os bombeiros já estão ajudando elai. "The resident said right now in the interview to the television that the firemen are already helping her"

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

12

(9) Gapless subject structure A moradora

disse agora mesmo na entrevista à televisão que os

bombeiros já estão chegando. "The resident said right now in the interview to the television that the firemen are already arriving."

(10) Topic-bound null object Os desabrigadosi, a moradora disse agora mesmo na entrevista que os bombeiros já estão ajudando [e] i. "The homeless, the resident said right now in the interview that the firemen are already helping ."

(11) Topic-bound pronoun Os desabrigadosi, a moradora disse agora mesmo na entrevista que os bombeiros já estão ajudando eles i "The homeless, the resident said right now in the interview that the firemen are already helping them."

(12) Gapless topic structure Os desabrigados, a moradora disse agora mesmo na entrevista que os bombeiros já estão chegando. "The homeless, the resident said right now in the interview that the firemen are already arriving ."

3.1.3. Procedure The study used

a

cross-modal priming technique.

Subjects were

given a probe recognition task in which target sentences were orally

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

13

presented and reaction times to visual probes corresponding to the antecedent of the overt pronoun and of the gap were measured in the case of the experimental sentences. For the control sentences,

we

measured reaction times to visual probes corresponding to the subject of the matrix clause in the case of the topic sentences and to the subject of the embedded clause in the case of the subject sentences.

Subjects indicated whether or not the probe word had occurred anywhere in the sentence by pressing a yes or a no key in a button box. This response removed the probe and presented a comprehension question on the screen which should also be answered by pressing a yes or a no key in the button box. Following three practice trials, experimental, control and filler sentences were presented in a different random order to each participant. Subjects were tested individually in sessions of approximately 25 to 30 minutes and generally reported in post-session interviews that the task was at least moderately easy.

3.2. Results The logic of the experiment was as follows. A consistent finding of recent studies on coreference processing is that reference-dependent sentential elements seem to trigger reactivation of the antecedent NP's to which they refer. This reactivation or "priming effect" of overt and implicit anaphoric elements has been found both in studies which checked for on-line probe recognition and in studies which use end-ofsentence probes as it is the case in our experiment. By recording and comparing reaction times to the experimental and control sentences across the 6 conditions we expected to assess whether pronouns and

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

14

empty categories in object position in BP would facilitate the access to their antecedents and whether this facilitation effect would vary with relation to the subject/ topic nature of the antecedent. The extent to which the reactivation of subjects and topics by the empty and by the overt anaphors varied with relation to the gapless conditions is shown in Figure 1. The gapless condition may be taken as a baseline in order to ascertain whether or not the presence of structural constraints will affect comprehension.

The configuration in Figure 1 indicates that

there are no significant differences between the processing times for the gapless topic and subject constructions. No matter how the human sentence processor copes with topic structures, the final results for the BP data show that comprehenders deal with topic-comment relations as fast as they deal with subject-predicate relations. This finding provides interesting processing evidence in favor of the claim that BP is a language in which topic/comment relations are as prominent as subject/predicate relations.

Graph 1: topics and subjects,antecedents

Reaction Times (ms)

1300

1200 topic subject

1100

1000

900 gapless

pronoun anaphor

[e]

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

15

Figure 1: Topics and Subjects as antecedents Our data show a stronger priming effect for the gap coindexed with the topic (the A'-bound gap) than for the gap coindexed with the subject (A-bound gap) and stronger priming effects for the overt pronoun coindexed with the subject than for the overt pronoun coindexed with the topic. In the topic sentences, reaction times to the antecedent probe were 180 milliseconds faster in the gap condition than the gapless condition, [F(1, 48) = 12.07, p < .05]. In the case of the subject sentences, response times to the antecedent probe were 187 milliseconds faster in the pronoun condition than the gapless condition, [F(1, 48) = 10.06, p < .05]. These are

clearly significant

statistical results, showing an interaction between sentence type and anaphor. In sum, Experiment 1 provides evidence that antecedents in A-position have a preference for overt anaphoric elements, whereas elements in A'-position prefer to be coindexed with empty categories.

3.3. Discussion 3.3.1. Subcategorization Information An important issue

in Sentence Processing has been whether the

language processor has rapid access to argument structure in parsing. Another related issue concerns the positing of gaps based on antecedent information. Clifton and Frazier (1989) claim that gaps are posited as a first resort, that is, the parser makes use of antecedent information in order to decide whether or not to hypothesize a gap in a sentence. Our results suggest that the parser has immediate access to subcategorization information and that a gap does not seem to be posited on the basis of antecedent information. Notice that RT's to the

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

16

gapless cases are not significantly different. That is, subject/predicate relations seem to be handled as readily as topic/comment relations. However, if the parser were to posit a gap as a first resort it would be legitimate to expect that gapless topic sentences should take longer to process than gapless subject sentences since only in the topic case a gap would be

mistakenly posited.

If these BP sentences were parsed

according to the Active filler hypothesis (Clifton & Frazier(1989)), the processor should postulate an empty category right after it identified a filler in nonargument position, in our case the topic. However, by pursuing such a strategy, the processor would be garden-pathed because the positing of a gap would be wrong in this case, since the verb in the embedded clause is intransitive and projects no empty category. Both the positing of the gap and the "surprise" reaction by the processor should presumably take processing time. However this is not the case in our data, since as we have showed above there is no significant difference between the processing times for the topic and the subject gapless sentences.

On the other hand, it seems clear in our data that subcategorization information

guides the parser in the processing of long distance

dependencies. Note that experimental sentences such as (10) in which there is a

transitive verb in the embedded clause, display a clear

priming effect of their antecedents in topic position when compared to the gapless topic sentences discussed above. This happens presumably due to the fact that the processor uses the information about

the

argument structure of the verb in order to postulate the gap and then associate it with the filler in topic position. In the gapless cases there is

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

17

no empty category to be postulated, once the verbs in the embedded clause are intransitive. Since there is no gap to reactivate the antecedent, RT's are longer in the gapless case than in the gap case. Therefore, our results provide additional evidence in favor of

an

"optimal" model of parsing, that is, a comprehension system which has access not only to syntactic category information, but also to other types of information, such as argument structure.

3.3.2.

The

Architecture

of

the

Human

Sentence

Processing

Mechanism Therefore, according to our data, topics do not seem to function as active fillers and

the postulation of a gap is not triggered by the

antecedent in A'-position. If this is the case, what is the architecture of the human sentence processing mechanism (HSPM) that could account for the BP facts presented above? We want to speculate that the HSPM deals with the fragment of BP grammar presented above in the following

manner. First the HSPM constructs a surface structure

representation of the sentences, left to right. In the case of the topic sentences, the topic phrase is represented in a nonargument position. In the subject sentences this position is left empty. As we have argued above,

the

parser

subcategorization

has

access

properties

of

to

information

predicates.

concerning

Therefore

upon

encountering a transitive verb the parser expects an object. If there is not an overt object, a gap is postulated and coindexation with an antecedent must take place.

As suggested by

Nicol (1988), the

assignment of coreference is carried on by a coreference module which constitutes an intermediate stage between purely structural processes

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

18

and interpretive processes. This coreference device must determine which referents are potential antecedents of a referentially-dependent item in order to access those referents. In the light of the results we obtained in Experiment 1, we propose that such a coreference device must have access to information concerning argument/nonargument positions in the following way. If the parser encounters a pronoun after the verb, the coreference device will search first for an antecedent in argument position. If there is not a suitable antecedent in argument position, then a referent must be found in discourse or in the pragmatic context. The crucial fact that is clear from our results is that object overt pronouns in BP "prefer" antecedents in argument position. As a default strategy, the coreference device looks for an antecedent for the pronoun at the sentence level as a first resort. If there is not one antecedent available at this level, then nonargument positions will be checked. Now, what happens if the processor finds a gap after the verb instead of a pronoun? In that case, as it is clear from our data, the coreference device will look for an antecedent in a nonargument position as the best candidate for coindexation with the gap. Thus, in the same way as an overt object pronoun triggers the search for an antecedent within the sentence level, object gaps look for their antecedents outside the scope of the sentence, in a peripheral position (A'-position) or even in the context of utterance.

In the remnant

paragraphs of this section , we attempt to speculate on possible cognitive reasons for this processing difference between subjects and topics.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

19

In the topic-comment type of packaging (Chafe 1976), the speaker seems to recall a concept which is then commented on. In the subjectpredicate type of packaging, the speaker seems to be also recalling an event or concept, however

this informational item is immediately

integrated into a proposition. At the syntactic level, procedures such as GF assignment, ø-role assignment, case assignment, etc. are going to take place in order to license that NP in a well formed sentence. In the topic-comment strategy, on the other hand, the concept is established as a generic frame of reference, a set of potentialities of meaning which may or may not be precised through grammatical operations and processes at the Sentence level. This peripherical

nature of the

position allows the interface between pragmatics and discourse and the grammaticalities of the sentence. This is is the distinction that Chafe proposes between the

topic as the psychological "subject" of the

sentence and the subject as the proper

grammatical "subject" of a

sentence. When we hear a gapless topic sentence such as (12) in BP it does not seem that we have a surprise effect that would delay the comprehension of the topic sentence when we find out that the topic cannot be integrated at the sentence level with the proper GF's, thetaroles, etc. In the production end, it seems that the speaker has an idea, but he does not know or want at this instant to make a predication on this idea that would take it as a specified argument of a proposition. Rather, he came up with this idea and he makes a comment on it without having necessarily to fit it into a specific structural slot. The concept sort of remains activated as "a psychological subject", a source of interpretive possibilities throughout the sentence. If there is a gap at the structural level to integrate the topic, a reactivation effect will

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

20

happen, as we have demonstrated in the experiment. The speaker may refer back to this concept, by means of a gap, an overt pronoun or an NP which in BP may even be the same as the topic alluded to. If the hearer tried to integrate the topic at once, it would be licit to expect that the RT's for gapless topic sentences would be higher than RT's for equivalent "gapless" subject sentences, since as we have already noted, there would be a "surprise effect" for the former only. Topics are concepts which get lexicalized (or not, consider the possibility of null topics), but which have a minimum degree of grammaticalization. Notice the existence in BP of gapless topics, PP topics with chopped preposition, etc,. These facts become even clearer if we consider that topics seem to occur mostly in the flow of spontaneous and colloquial discourse and less in carefully planned and organized speech (Pontes, 1987). Therefore the best candidates for topic anaphora are empty categories, since these elements do not carry any definite inflectional markings, such as pronouns do. Even in terms of the content of their referents, they do not necessarily retrieve any specific and definite entity, since they can always have an arbitrary or pragmatic interpretation. It is thus legitimate to suppose that gaps are more natural anaphoric means to retrieve topics than overt pronouns, since they

do not commit themselves to

inflectional features, such as

pronouns do. Therefore, overt pronouns and gaps seem to differ in terms of their recoverability capacities. While pronouns will recover grammatical features such as gender, number, case etc., which have usually been grammaticalized in the subject position, gaps are simply structural slots, placeholders for antecedents.

In other words, overt

pronouns and gaps adopt distinct types of anaphoric strategies:

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

21

whereas the overt pronoun requires the matching with certain grammatical features, gaps do not. Thus pronouns just go better with arguments,

because

these

have

already

been

adequately

grammaticalized in the sentence and gaps are the best retrievers for topics, given their usual low degree of grammaticalization. Notice that this low level of grammaticalization of topics might

even

be a

characteristic of their nonargument position, in which no case, thetaroles or grammatical functions are assigned. It may also be the case that the preference for zeros as anaphoric devices to make reference to topics is an on-line processing strategy used by the language processor in production in order to avoid identificational errors which might arise as a result of mismatch of feature agreement.

It is interesting to note that this processing default strategy which we are proposing to account for the ability of zero anaphora to display higher levels of reactivation of its topic antecedent than the overt lexical pronoun actually provides a partial explanation for Givón's (1983) scale of phonological size. In his speaker-hearer neutral study of topic continuity in discourse Givón

notes

that zero anaphora is

more used cross-linguistically as a grammatical device to identify topics than other other coding devices with increasing phonological size. Givón proposes that a basic principle of iconicity underlies this scale of phonological size: "the

more disruptive, surprising,

discontinuous or hard to process a topic is, the more coding material must be assigned to it" (p.18). In turn, Givón continues, this principle may translate into a more generic psychological principle: "expend only as much energy on a task as is required for its performance"

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

22

(p.18). If the proposal we developed above is right, it provides a more straightforward psychological content for Givón's principles: gaps are better retrievers of topics

than overt pronouns because they have

different recoverability properties. In any case, regardless of how these facts are

processed, the very observation that there are such

differences in the anaphoric selectional properties of gaps and overt pronouns is a clear indication that even though both elements seem to behave syntactically as pronominals,

they do have different

interpretive properties. In Experiment 2 the interpretive differences between the empty and the overt pronominal objects in BP will be investigated in the light of Montalbetti's Overt Pronoun Constraint (OPC).

3.4. Conclusion In summary, Experiment 1 allows the following conclusions: a) BP facts support the view that the human sentence comprehension mechanism is guided by structural considerations. There seems to be a processing distinction between subjects and topics which can possibly be related to elements which occupy argument position versus elements which occupy a nonargument position. b) Topic bound object gaps in BP are psychologically real in the sense that they allow a faster reactivation of their antecedents. Subject-bound object gaps are either not psychologically real in BP or if real they are interpreted as arbitrary and do not allow a specific and definite interpretation. c) Subcategorization information is rapidly accessed in BP sentence parsing. Unlike the prediction of the Active filler hypothesis (Frazier,

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

23

83), information about the predicate about the antecedent

argument structure

and not

triggers the postulation of a gap in BP topic

sentences. d) A processing explanation is proposed in order to account for Givón's et alii (1983) crosslinguistic finding that zero anaphora is a more common grammatical device for the identification of topics in discourse than stressed/independent pronouns.

4. Experiment 2: the Interpretation of Object Anaphora in BP The results of Experiment 1 strongly suggest the existence of different interpretive properties between the overt pronoun and the empty category in object position in Brazilian Portuguese (BP): overt pronouns reactivate subjects but not topics and gaps reactivate topics but not subjects. These results seem to provide processing confirmation to the Overt Pronoun Constraint (Montalbetti (1984)), which predicts that overt pronouns seem to be restricted to the coreferential reading whereas empty categories interpretation.

can display

In Experiment 2,

a bound pronoun

the difference

between the

coreferential vs. bound readings in BP is further investigated by comparing the possibility of the strict

and sloppy readings for overt

and nonovert pronouns in object position.

In certain constructions, pronouns which have non quantificational NP's as antecedents can display

either a coreferential or a bound

reading. Thus (13) can be interpreted as either (13a) or (13b):

(13) John thinks that he caught a fish and so does Peter.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

24

(13a) John thinks that John caught a fish and Peter thinks that John caught a fish. (13b) John thinks that John caught a fish and Peter thinks that Peter caught a fish.

(13a) indicates that the pronoun "he" in (13) is interpreted in a strictly coreferential sense whereas (13b) constitutes what Ross (1967) called the sloppy identity of the pronoun. As pointed out by Montalbetti (1984), the strict reading in (13a) is obtained on purely coreferential grounds: the reference of "he" is established by assigning to it the value "John" both in the first and in the second conjunct. However, in order to obtain the sloppy reading in (13b)

we must assume that the first

conjunct of (13) contains an open sentence of the form: x thinks that x caught a fish. This open sentence is satisfied by "John" in the first conjunct and by "Peter" in the second, indicating that some form of a variable binding process is taking place in order to obtain such reading. Let us consider now a BP sentence such as (14):

(14) A professora encontrou o irmão no mercado e a advogada também encontrou ele/[e]. "The teacher met (her) brother in the market and the lawyer also met him/[e]"

If the OPC holds for BP we should expect the construction with the overt pronoun ele "he" to allow only the strict reading in (14a) whereas

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

25

the construction with the null object should also be able to allow the sloppy interpretation in (14b).

(14a) A professora encontrou o irmão dela e a advogada também encontrou o irmão da professora. "The teacher met her brother and the lawyer also met the teacher's brother."

(14b) A professora encontrou o irmão dela e a advogada também encontrou o irmão da advogada. "The teacher met her brother and the lawyer also met the lawyer's brother."

Two notes are in order. First notice that the experimental sentences were deliberately designed with an implicit possessive argument of an inalienable possession rather than with an overt possessive, following the most natural usage

in colloquial BP. That is,

the

implicit

possessive construction with the definite article (e.g. o irmão brother" )

was used

instead of the construction

"the

with the overt

possessive pronoun (e.g. seu irmão "her brother") or with the analytic genitival form of the type "de + N" [of +N] (e.g. o irmão dela "the brother of hers"), since the implicit possessive construction is the preferred strategy in colloquial BP.

Secondly,

notice that the

continuation sentences make a distinction between one entity (the strict case)

vs. two entities (the sloppy cases).

experimental sentences ,

Thus,

in

the

the continuation sentence with the strict

interpretation was construed as, for example, O irmão de Helena e José

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

26

estava no mercado "Helena and Jose's brother was in the market" , that is presupposing the existence of only one entity (one brother), regardless of the fact that this entity could be related both to Helena and José. The continuation sentence with the sloppy interpretation, on the other hand, was construed as O irmão de Helena estava no mercado e o de José também

"Helena's brother was in the market and Jose's too",

establishing the existence of two different entities (two brothers). Since the main objective of the experiment was to check if indeed we should expect the construction with the overt pronoun ele "he" to allow only the strict reading and whether the construction with the null object should also be able to allow the sloppy interpretation, we considered that the difference between

one entity vs. two entities would

be

appropriate to establish the distinction. Indeed as we discuss below, the results of experiment 2 indicated that the construction with the null object is ambiguous between the two readings (one entity vs. two entities) whereas the construction with the overt pronoun allows only the strict reading (one entity).

4.1. Method In this section we describe the participants, the materials and design as well as the testing procedures used in the experiment.

4.1.1. Participants 20 brazilian USC undergraduate or graduate students volunteered to serve as participants. They all spoke BP as their first language and had normal hearing and normal or corrected eye sight.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

27

4.1.2. Materials and Design The stimuli were 4 sets of 24

two-sentence Experimental passages.

Each subject was presented one of these experimental sets embedded in an extra set of 46 two-sentence filler passages and 4 two-sentence practice passages.

Each

experimental set was made up of 4

conditions with 6 passages per condition in a 2x2 type of design. The first sentence in each passage ended with either an overt pronoun or with a gap. The second sentence was a continuation sentence with either a strict interpretation or a sloppy interpretation

of the first

sentence. Notice that although sentences were very similar, differing only in terms of the tested variables, they were distributed between subjects, so that each subject saw only one version of each sentence type. The distribution of sentence types in 4 sets allowed all sentences of a type to be compared.

4.1.3. Procedures In order to assess BP speakers' interpretative preference intuitions concerning these constructions, Experiment 2 uses a measure which is a unimodal

auditory adaptation of the method employed by

Kurtzman & MacDonald (1993) in their study of quantifier scope ambiguities. Subjects hear

sentences such as (15) or (16) followed by

the auditory presentation of a continuation sentence such as (a) or (b). Participants then are asked to judge whether the continuation sentence is indeed a reasonable continuation of the first sentence. For example, for (15) and (16) the continuation sentence would be either ( a) or ( b):

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

28

(15) O Pedro castigou a filha e a Maria também castigou ela. "Pedro punished (his) daughter and Maria also punished her."

(16) O Pedro castigou a filha e a Maria também castigou [e]. "Pedro punished (his) daughter and Maria also punished [e].

(a) A filha de Pedro e Maria se portou mal. (strict reading) "Pedro and Maria's daughter misbehaved."

(b) A filha de Pedro se portou mal e a de Maria também. (sloppy reading) "Pedro's daughter misbehaved and so did Maria's daughter."

If the OPC holds in BP

only the (a) sentence should be judged as a

reasonable continuation for the sentence in (15), since, according to Montalbetti's generalization, the sloppy reading should not be available for the overt pronoun. For sentence (16), both the sloppy reading in (b) and the strict reading in (a) should be possible.

Right after the presentation of the second sentence a question mark would show in the CRT screen. At this point

subjects indicated

whether or not the continuation sentence was a reasonable discourse continuation of the first sentence by pressing a yes or a no key in a button box. Response times were measured. Following four practice trials, experimental and filler sentences were presented in a different random order to each participant. Subjects were tested individually in

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

29

sessions of approximately 15 to 20 minutes and generally reported in post-session interviews that the task was at least moderately easy.

4.2. Results The extent to which continuation sentences were judged compatible with the anaphor sentences is presented in Table II and displayed in Figure 2. TABLE II. Compatibility judgments (%)

Condition Null Strict

% acceptance (nst)

88 %

Null Sloppy (nsl)

89 %

Pronoun Strict (pst)

94 %

Pronoun Sloppy (psl)

36 %

Judgment %

100

% compatible judgment %incompatible judgment

% judgment

80 60 40 20 0 nst

nsl

pst

Condition

FIGURE 2. % Judgment

psl

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

30

For the null object sentences,

participants judged

both the

continuation sentences with a strict interpretation (NST) and the continuation sentences with a sloppy interpretation (NSL)

as

compatible. For the overt pronoun sentences, an effect between the strict vs. sloppy readings obtained. Subjects judged the continuation sentences with a strict interpretation (PST) as compatible with the overt pronoun first sentences, but they

judged

the continuation

sentences with a sloppy interpretation (PSL) as incompatible in 64% of the cases. This pattern produced a robust anaphor X interpretation interaction, [F (1,19) = 51.1, p < .05] . Judgment Latencies

Response Time (msec)

2000

YES NO

1000

0 nst

nsl

pst

psl

Condition

FIGURE 3. Reaction Times for yes and no answers to continuation sentences Figure 3 compares the response times for the Participants took

judgment

task.

in average one second to say yes to the NST, NSL

and PST continuation sentences, but in those 36 % of the observations which judged the PSL continuation sentences as compatible,

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

31

participants needed in average 800 ms more to reach their decision. For the null object, there is a nonsignificant response times difference of only 50 ms. between those who accepted the strict interpretation (NST)

and those

who accepted the possibility of the sloppy

interpretation (NSL). However, for the overt pronoun, there is an RT difference of more than 900 ms. between the acceptance of the strict and of the sloppy interpretation. Since there were 5 subjects out of the 20 who took the experiment who either did not give any yes answer to the possibility of the sloppy interpretation of the overt pronoun or who took too long to reach their decision (above the 2.0 Standard Deviation cutoff point), there were empty cells in the data set which made it impossible to calculate the ANOVA here. Nevertheless, the RT difference of over 800 ms. seems to indicate that those few subjects who did accept the possibility of the sloppy interpretation displayed a considerable degree of hesitation to reach their decision.

4.3. Discussion In Experiment 2, the difference between the coreferential vs. bound readings of overt and empty anaphors in BP was further investigated by comparing the possibility of the strict

and sloppy readings for

overt and nonovert pronouns in object position. Subjects tested in a speeded judgment task

showed a preference for

the strict

interpretation of the overt pronoun, whereas the null object construction is also able to allow the sloppy interpretation. Experiment 2 clearly supports the hypothesis that the

Thus, sloppy

interpretation is not available to the overt pronoun, as predicted by Montalbetti's generalization.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

32

5. General Discussion In section 3, we presented experiment 1, which established the central facts in the study: the empty category in object position reactivates its subject antecedent but not its topic antecedent whereas the opposite configuration holds for the overt lexical pronoun. We argued that our experimental results clearly teased apart the A-bound from the A-bar bound null objects and revealed that the overt lexical pronoun does not have the ability to reactivate its topic antecedent. Additionally, we argued that our experiment

demonstrated that topic-comment

constructions in BP are processed as readily as the subject-predicate type. We also showed that the results

of Experiment 1

provided

partial psychological justification for Givón's scale of topic accessibility and suggested a processing explanation to the observed difference in reactivation properties between the empty and the overt elements: a default, first resort parsing strategy

in the referent search process

which makes empty categories the ideal retrievers of topics. We finally claimed our experiment to be

informative with relation to several

theoretical psycholinguistic issues: the psychological reality of gaps, the role of grammatical structure in the priming process, the effect of the global topic of discourse in reference resolution, the rapid access by the human parser

to information on

the argument frames of

predicates.

Finally, in section 4, we further investigated the results of experiment 1 in the light of Montalbetti's generalization: the fact that in several Romance languages overt pronouns cannot be locally A-bar bound, but empty categories can (Montalbetti, 1984). We articulated an evaluation

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

33

of the applicability of Montalbetti's generalization in BP and presented experiment 2 which provided further confirmation to the interpretive distinction between overt and nonovert elements in BP : unlike the empty category, the overt lexical pronoun in object position in BP cannot be assigned a sloppy identity.

Therefore, our research provided strong processing evidence to the claim that empty categories and overt pronouns display important syntactic and semantic differences. As it was demonstrated in experiment 1, a strictly syntactic difference

in behavior bettween

empty and overt elements was captured in a priming paradigm: nulls were shown to reactivate topics and overts were shown to reactivate subjects. In experiment 2 , nulls and overts were also shown to differ on semantic grounds: the sloppy interpretation is not available to overt pronouns.

References BEVER, T. G. & MCELREE, B. (1988). Empty categories access their antecedents during comprehension. Linguistic Inquiry, 19, 35-43. CALLOU, D. et al. (1993) Topicalização e Deslocamento à esquerda: Sintaxe e Prosódia. In A. T. Castilho (org.), Gramática do Portugues Falado, São Paulo, Editora da Unicamp/FAPESP. CHAFE, Wallace (1976). Giveness, contrastiveness, definiteness, subjects, topics and point of view. In Li, C.N. Subject and Topic, New York: Academic Press.

CHANG, F.R. (1980) Active memory processes in visual sentence

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

34

comprehension: Clause effects and pronominal reference. Memory and Cognition, 8, 58-64. CLIFTON, C. ; FRAZIER, Lynn (1989). Comprehending sentences with long distance dependencies. In Greg N Carlson and Michael K. Tanenhaus (eds.), Linguistic Structure in Language Processing, Kluwer Academic Press. COWART, W. & CAIRNS, H. (1987). Evidence for an anaphoric mechanism within syntactic processing: some reference relations defy semantic and pragmatic constraints. Memory and Cognition, 15, 318331. CRAIN, S. & STEEDMAN, M. (1985). On not being led up the garden path: The use of context by the psychological parser. In D. Dowty, L. Kartunnen, & A. Zwicky (Eds.), Natural Language Parsing.. Cambridge: Cambridge University Press. DUARTE, Maria Eugênia (1989). Clítico acusativo, pronome lexical e categoria vazia no Português do Brasil. In Tarallo, Fernando,Fotografias Sociolingüísticas. Pontes, Campinas, 19-34. FERREIRA, F. & CLIFTON, C. (1986). The independence of syntactic processing. Journal of Memory and Language, 25, 348-368. FODOR, J. D. (1989). Empty categories in sentence processing. Language and Cognitive Processes, 4 (3/4) SI 155-209. FRAZIER, L. (1979). On comprehending sentences: Syntactic parsing strategies. Bloomington: Indiana University Linguistics Club. FRAZIER, L., CLIFTON, C., & RANDALL, J. (1983). Filling gaps: Decision principles and structure in sentence comprehension. Cognition, 13, 187-222. GALVES, Charlotte (1989). O objeto nulo no Português Brasileiro: Percurso de uma pesquisa., ms. Unicampi. GIVÓN, T. (1983) Topic Continuity in Discourse: A Quantitative CrossLanguage Study. Amsterdam/Philadelphia: John Benjamins Publishing Company.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

35

HASEGAWA, N. (1985). On the so-called "zero pronouns" in Japanese. The Linguistic Review, 4, 289-341. HUANG, C.-T. J. (1984a). On the distribution and reference of empty pronouns. Linguistic Inquiry 15, 531-574. KURTZMAN, H. & MACDONALD, M.C. (1993) Resolution of Quantifier Scope Ambiguities, Cognition, 48, 243-279. MACDONALD, M.C. (1989). Priming effects from gaps to antecedents. Language and Cognitive Processes, 4, 35-56. MAIA, M. (1994). The Comprehension of Object Anaphora in Brazilian Portuguese. Doctoral Dissertation, USC, Los Angeles. MONTALBETTI, Mario (1984). After Binding : On the Interpretation of Pronouns. Doctoral Dissertation, MIT, Cambridge, Massachusetts. NICOL, Janet (1988). Coreference Processing during Comprehension. Doctoral Dissertation, M.I.T.

Sentence

PONTES, Eunice (1987). O tópico no Português do Brasil. Campinas, SP: Pontes. ROSS, J.R. (1967). Constraints on Variables in Syntax. MIT dissertation. TANENHAUS, M., STOWE, L., & CARLSON, G. (1985). The interaction of lexical expectation and pragmatics in parsing filler-gap constructions. In Proceedings of the Seventh Annual Cognitive Science Society Meetings, Lawrence Earlbaum Associates, 361-65. TANENHAUS, M. & CARLSON, G. (1989) Lexical structure and language comprehension. In W.D. Marslen-Wilson (Ed.), Lexical representation and process. Cambridge, MA: MIT Press. TARALLO, F.L. (1983) Relativization Strategies in Portuguese. PhD dissertation. University of Pennsylvania.

Brazilian

TEIXEIRA, R.F. (1985) Zero Anaphora in Brazilian Portuguese Subjects and Objects. UC Berkeley PhD dissertation.

Artigo publicado em Recherches Linguistiques de Vincennes nº 26, p.151-172, França, 1997.

36