1

Case markers as clause boundary inducers in Japanese

Edson T. Miyamoto NAIST

May 7, 2002

Send correspondence to: Nara Institute of Science and Technology Graduate School of Information Science 8916-5 Takamaya, Ikoma, Nara 630-0101, JAPAN E-mail: [email protected] FAX: +81-743-72-5249

Case markers and clause boundaries

2

Abstract The present paper argues that the processing of verb-final clauses proceeds incrementally based on local information that becomes available with each word. The results of three selfpaced reading experiments are reported in support of the proposal that NPs in Japanese are associated within clauses before a verb is processed. A clause boundary is posited whenever case markers prevent two NPs to be part of the same clause, and slow reading times at the second NP are used as supporting evidence. Moreover, clause boundaries induced by case marking can facilitate processing at later points in the sentence as attested by faster reading times at relative-clause heads. Contrary to previous findings that argued against a subclass of head-driven parsers, the present results are not easily reconcilable with models that delay parsing decisions until a verb is available in the input sentence.

Key words: Incremental processing, case markers, clause boundary ambiguities, Japanese

Case markers and clause boundaries

1

In languages such as English, the verb comes in relatively early in the clause and can immediately influence various aspects of sentence processing. In SOV (subject-object-verb) languages such as Japanese, however, it is often the case that a sentence starts with a sequence of case-marked NPs which must be followed by one or more verbs in order to complete the sentence (see Kuno, 1973; Tsujimura, 1999, for overviews of Japanese syntax within a generative linguistics framework which is adopted here for expository purposes). If parsing takes place without delay, then NPs in such sequences in Japanese should be integrated within a partial interpretation of the sentence even before a verb is processed. Alternatively, it is possible that the NPs are only interpreted jointly when a relevant predicate is processed. The present paper argues that NPs in Japanese are associated incrementally according to their case markers anticipating the verbs to come (for discussions on case markers in Japanese sentence processing, see A. Inoue, 1991; Yamashita, 1994, 1997; also Kim, 1999; Suh, 1994, for related observations for Korean). As evidence for the early association of NPs, it will be suggested that a slow-down is observed whenever a NP cannot be a co-argument of the preceding NP, in other words, whenever a clause boundary has to be posited in order to indicate that two adjacent NPs have to be associated with different verbs. For example, (1) schematically represents a NP sequence in which NP1 and NP2 are associated (either as arguments or as adjuncts) with the main verb V2; whereas NP3, NP4 and NP5 are associated with V1, the verb in the embedded clause. Local information allowing, the boundary between NP2 and NP3 should be created as soon as NP3 is processed, and a slow-down is predicted at this point. (1) [ NP1 NP2

[ NP3 NP4 NP5

. . . V1 ] . . . V2 ]

An example of case markers locally determining clause boundaries comes from the syntax literature and is known as the double o constraint, which states that a clause in Japanese may not contain more than one o-marked NP (Harada, 1973; Kuroda, 1992). In processing terms, when confronted with two successive o-marked NPs, a reader should slow down at the second NP in order to create a clause boundary.

Case markers and clause boundaries

2

The parsing model implicit throughout this paper is consistent with what is generally assumed about the processing of head-initial languages (i.e., languages such as English in which the phrasal head occurs in the initial position of the phrase, as opposed to head-final languages such as Japanese in which the head occurs at the end of the phrase). In particular, the model is incremental (Marslen-Wilson & Tyler, 1980; inter alia) in that there is no delay in parsing, and constituents are associated to each other immediately according to local information available. The main difference lies on the type of local information available in each language (e.g., case markers in Japanese as opposed to word order in English). The model is not head-driven in that parsing decisions are made even when the relevant head (in the present case, the verb) is not available. To illustrate the effect of case markers in Japanese, three self-paced reading experiments are reported. Experiment 1 is a direct application of the double o constraint. Experiments 2 and 3 make use of a more generalized strategy, namely, the local assignment of clause boundaries. Various delays, that would have to be explained in a case by case basis in a head-driven model, are claimed to have a common cause in the processing of clause boundaries as directed by case markers.

Head-driven parsing Head-driven models of parsing are characterized by the following property (Abney, 1989; Pritchett, 1991, 1992; and the partial commitment parser in A. Inoue, 1991). (2) Delay : Constituents are left unattached if a licensing head is not available. For example, in a bottom-up parser for SOV languages, an object NP can only be attached to the tree structure being built when the verb is encountered, because only then is a VP (or a V
) node created. Delay prescribes the presence of a licensing head as a necessary condition for attachment. A natural extension is to assume that a licensing head is also a sufficient condition, so that attachment obligatorily occurs as soon as a relevant head is available (see Bader & Lasser, 1994, for discussion). In other words, as many previously unattached constituents as

Case markers and clause boundaries

3

possible should be integrated in the tree structure when a licensing head is encountered, as the following suggests (see for example the theta attachment principle in Pritchett, 1992). (3) Maximal licensing: As many unattached constituents as possible are attached to the current head being processed. Experimental studies arguing against head-driven models have provided evidence against maximal licensing, but not against delay per se. For example, Bader and Lasser (1994) provided evidence from a verb final clause construction in German, schematically represented in (4). The pronoun sie is ambiguous between nominative and accusative case. If interpreted as nominative, sie must be the subject of V2, the verb in the outer clause; but if interpreted as accusative, the pronoun has to be the object of V1 , the verb in the inner clause. (4) . . . daß sie . . . V1 . . . V2 Bader and Lasser detected a slow-down at V2 if this verb could not take sie as a subject, indicating that readers had not attached sie when they read V1 . Therefore, maximal licensing is unlikely to be determining attachment at V1. A similar result was obtained by Kamide and Mitchell (1999) who examined the Japanese construction schematically represented in (5). The dative NP is optional and can be associated to V1 (the verb in the embedded clause) or to V2 (the verb in the main clause).1 (5) NP-nom NP-dat NP-nom . . . V1 . . . V2 Kamide and Mitchell compared (5) with two unambiguous versions obtained by manipulating the argument structure of the two verbs. In one unambiguous condition, the dative NP could only attach to V1 but not to V2, and a slow-down was observed at the second verb in comparison to the same word in (5). In another condition, the dative NP could only attach to V2 and no slow-down was observed at any point. As in the German construction, the result suggests that readers prefer not to attach the dative NP when they process V1. Although neither result is compatible with maximal licensing, they could be explained by a head-driven model in which some but not necessarily all of the permissible NPs are attached

Case markers and clause boundaries

4

to the verb being currently processed. In this type of model, independent factors would conspire to yield the similar results obtained in the German and Japanese constructions (e.g., a clause-initial sie is preferentially interpreted as nominative in German; whereas, in the Japanese construction, the second nominative NP may prevent NPs to its left to be attached to V1).2 The present paper provides evidence against delay in head-driven models by examining the processing of NP sequences in Japanese.

Processing clauses in Japanese Whenever possible, Japanese speakers interpret a sequence of NPs followed by a verb as a single clause. This initial preference follows from various principles proposed in the sentence processing literature such as minimal attachment (Frazier, 1987), properties of thematic reception (Gibson, 1991), simplicity (Gorrell, 1995), the theta attachment principle (Pritchett, 1992). Experimental evidence comes from a lexical decision task in which native speakers of Japanese were faster to recognize served, when compared to drank, as a valid Japanese word after they were presented with a NP sequence like the one in (6a) (Yamashita, 1994, 1997; adverbs and adjectives are not shown below). (6) a. Seito-ga

sensei-ni

koocha-o . . .

student-nom teacher-dat tea-acc b. . . . dashita served c. . . . nonda drank The verb served allows the sequence in (6a) to be interpreted as part of a single clause (“The student served the tea to the teacher.”); whereas with drank, a more complex structure involving two clauses is necessary (for example, the sentence could be completed with the

Case markers and clause boundaries

5

words man-acc and introduced as in “The student introduced to the teacher the man who drank the tea.”). The result is compatible with two types of explanations. Under an incremental model, the NPs must have been associated without delay. Based on their case markers, there is a bias for the interpretation in which student is an agent, teacher is a goal, and then tea is added as a theme. An upcoming ditransitive verb such as served is recognized more rapidly because it conforms with the expectations created from the way the NPs were associated. A head-driven model could also explain Yamashita’s result based on the number of grammatical violations at each licensing head (e.g., the number of NPs without a thematic role, in other words, NPs whose semantic role in the sentence is still to be determined). Assume that each NP in (6a) is individually built but that they are not associated to each other until a verb is detected. At the verb, readers associate each NP with the verb. In this model, the recognition of served is fast because this verb is able to license all NPs in (6a); with drank, the first and second NPs cannot be attached and incur two violations.

Arguments against head-driven models The head-driven explanation is challenged in a number of aspects. First, there is the intuition that the NPs in (6) are partially interpreted before the verb is read. Although not entirely clear what the event is, the case markers (and possibly the animacy of the NPs) in (6a) imply the transfer of the tea from the student to the teacher (see A. Inoue, 1991, for other intuitive judgements which prompted him to revise his partial commitment parser, a head-driven model; also, A. Inoue & Fodor, 1995, for related discussion). Second, head-driven models imply that speakers of SOV languages should be in severe disadvantage in comparison to speakers of SVO languages. If approximately up to seven unrelated items or chunks can be maintained in working memory simultaneously (Miller, 1963), head-driven models should predict that speakers of SOV languages are not able to process a sentence with seven or more constituents preceding the verb (for example, a ditransitive verb preceded by a subject, two objects, and four or more adverbs). In contrast, speakers of SVO languages are predicted to have no problem with a similar sentence in which an

Case markers and clause boundaries

6

arbitrary number of constituents follows the verb, because each incoming constituent can be immediately associated with the verb in a complex chunk that does not occupy further slots in working memory. Even if such long clauses are rare in everyday use (especially considering the high incidence of argument omission in SOV languages such as Japanese and Korean), head-driven models make similar predictions for shorter sentences. In general, clauses with the same number of constituents are predicted to require more working memory to be processed in SOV languages than in SVO languages. Such a disadvantage should have dire consequences for the acquisition and historical evolution of SOV languages, when in fact typological studies have found that the SOV order is more common than the SVO order across the languages of the world (Dryer, 1989). Third, recent experimental results in Japanese provide converging evidence for the incremental processing of head-final clauses (Kamide et al., 2000; Kamide & Mitchell, 1999; Miyamoto et al., 1999; also Den & M. Inoue, 1997, for evidence that object NPs predict the verb to come).

Underspecified heads One way of characterizing the incremental processing of head-final clauses is to assume the prediction of underspecified nodes (see predictive parsing in Miyamoto et al., 1999, and references therein). For example, in the processing of (6a) above, readers may create a node for the verb even before the verb itself is read. Such a node is underspecified in that only some of its features are known in advance (e.g., it is a verb that takes a subject and two objects; see Scheepers et al., 1999, for a discussion on how the argument structure of a verb is predicted based on its arguments), but this is enough to permit the NPs to be associated together in a coherent chunk. When the verb comes in, only the more specific aspects of its lexical information have to be added to the node previously created, and the features that were underspecified can be instantiated. Pritchett (1991) argues that head-driven parsing is required by a competence level constraint such as the projection principle (Chomsky, 1981). However, the on-line use of predicted heads is not incompatible with the projection principle if we take it to be a well-

Case markers and clause boundaries

7

formedness condition which does not constraint the way in which the representation is built. Moreover, parsing with predicted heads is consistent with the assumption that subcategorization is a symmetrical relation in which verbs require specific complements in the same way that complements require a specific verb (Chomsky, 1965; see also Ades & Steedman, 1982). Processing with underspecified categories can be criticized as being tentative at best given that it is not known for sure what the thematic roles of the NPs are going to be until the verb is processed. However, a similar underspecification is observed in SVO languages. In English, a transitive verb cannot assign a thematic role to its subject until the direct object is determined as suggested by the following examples (Marantz, 1984, p. 25). (7) a. throw a baseball b. throw support behind a candidate c. throw a party In each of the sentences above, the exact role of the subject varies depending on the type of complement that follows the verb. Thus, thematic-role indeterminacy during incremental parsing is not restricted to head-final languages, but is observed in head-initial languages as well. In either case, uncertainty does not prevent incremental processing from taking place. The present paper provides experimental evidence that NP sequences in Japanese are parsed without delay. The difference with languages such as English is that in Japanese the sources of local information available are of a distinct nature. The particular type of information to be considered here, namely case markers, has been discussed elsewhere (A. Inoue, 1991; Yamashita, 1997).

Case markers in Japanese processing A. Inoue (1991) considers a number of Japanese constructions in which case markers are crucial to determine the correct interpretation. Inoue provides a few examples in which intuition supports incremental parsing, but other examples favor head-driven parsing. For example, he suggests that three NPs (as in (6) above) are likely to be associated to each other

Case markers and clause boundaries

8

incrementally before the verb is read; whereas two NPs (for example, a nominative NP and an accusative NP) do not seem to be associated to each other until the verb is processed as would be the case in head-driven models. In order to conciliate these contradictory intuitions, Inoue developed the information paced model. Although incremental in nature, this model may at times appear to delay decisions because its flexible reanalysis routine allows previously made decisions to be undone imperceptibly as relevant information becomes available. Thus, Inoue mainly investigated the processing of a verb or its following word in the sentence in order to determine how reanalysis of previous decisions is accomplished in Japanese. The present paper, in contrast, is primarily concerned with the initial association of the NPs before a verb is processed. Even when intuition could suggest otherwise, it will be argued that disruptions in reading times can be used as evidence for the immediate association of NPs. The evidence used here has a precedent in work done by Yamashita (1997). Yamashita (1997) reported experimental results suggesting that Japanese readers are sensitive to the types of case markers that appear in NP sequences. The present paper extends her result by providing evidence that the slow-down she observed in a particular sequence of NPs (see below for further discussion) also occurs in other constructions and is a special case of a general strategy to assign clause boundaries.

Clause boundaries and the double o constraint Given the preference for a single clause interpretation discussed above, the sentence in (8a) would be initially interpreted as a single clause as in (8b). However, at woman-acc, it is clear that served is part of a relative clause (RC) and reanalysis must take place in order to switch from the initially favored single-clause interpretation to a multi-clausal interpretation.

Case markers and clause boundaries (8) a. Ofisu-de shokuin-ga [RC ti kakaricho-ni ocha-o dashita] joseii -o office-loc employee-nom

manager-dat tea-acc served

9

teineini

woman-acc politely

shoukaishita. introduced “At the office, the employee politely introduced the woman who served the tea to the manager.” b. Ofisu-de shokuin-ga

kakaricho-ni ocha-o dashita.

office-loc employee-nom manager-dat tea-acc served “At the office, the employee served the tea to the manager.” There is extensive literature on the processing of various types of clause boundary ambiguities in English (Ferreira & Clifton, 1986; Sturt et al., 1999; Trueswell et al., 1994; inter alia). Ambiguous fragments in those experiments (e.g., “The evidence examined. . . ”) were compared to unambiguous counterparts obtained through relatively simple modifications (e.g., the insertion of a complementizer or a relative pronoun followed the verb be as in “The evidence that/which was examined. . . ”). However, Japanese does not have relative pronouns and, as a head-final language, complementizers come at the end of the clause; therefore, the beginning of an embedded clause is often left ambiguous. In the following, the italicized case markers were manipulated in order to induce a clause boundary at tea. (9) Ofisu-de shokuin-ga

kakaricho-o [RC ti ocha-o dashita ] joseii -ni

office-loc employee-nom manager-acc

tea-acc served

teineini

woman-dat politely

shoukaishita. introduced “At the office, the employee politely introduced the manager to the woman who served the tea.” In comparison to (8a), the case markers on manager and woman in (9) were swapped so that the former noun has the accusative marker and the latter noun has the dative marker. Although one single clause suffices to accommodate the fragment up to manager in (9),

Case markers and clause boundaries

10

a clause boundary must be posited between the two accusative NPs (manager and tea) as required by the double o constraint, which prohibits two accusative NPs to occur in the same clause (Harada, 1973; Kuroda, 1992). If inserting a clause boundary requires extra processing time, readers should slow down at tea-acc in (9) when compared to the same word in (8a). Furthermore, the RC head woman, which requires an embedded clause interpretation, should be read faster in (9) than in (8a). Head-driven models predict that NPs are left unattached until the verb is processed; hence, a slow-down at tea in (9) would have to be explained in terms of an independent factor such as the potential interference caused by the adjacency of the two accusative NPs in working memory (see Lewis & Nakayama, in press, for related discussion on the difficulty caused by similar constituents). In this type of model, there is no a priori relation between the time to process the NP sequence and the RC head. A self-paced reading experiment with sentences (8a) and (9) was conducted to test the predictions above. The experiment also included a third condition which was a variation of (9) with the NP manager-acc scrambled so that it would not occur next to tea-acc. This latter condition should indicate whether there is processing difficulty even when the two accusative NPs are not adjacent to each other.

Experiment 1 Method Participants Thirty-eight native speakers of Japanese were paid to participate in the experiment. They had gone to the U.S. as adults and were residents of the Boston area at the time of the experiment. The data of three participants were discarded because their overall comprehension performance was below 65% (for the remaining 35 participants, M = 75%, SD = 4.3%). Analyses including all thirty-eight participants revealed the same trends as the analyses with thirty-five participants reported below.

Case markers and clause boundaries

11

Materials Twenty-five sets of sentences with five conditions each were constructed. (Appendix A contains the complete list of items used.) (10) a. Ambiguous relative clause (ARC) Ofisu-de shokuin-ga

kakarichoo-ni ocha-o dashita josei-o

office-loc employee-nom manager-dat tea-acc served

teineini

woman-acc politely

shoukaishita. introduced “At the office, the employee politely introduced the woman who served the tea to the manager.” b. Double accusative sentence (DA) Ofisu-de shokuin-ga

kakarichoo-o ocha-o dashita josei-ni

office-loc employee-nom manager-acc tea-acc served

teineini

woman-dat politely

shoukaishita. introduced “At the office, the employee politely introduced the manager to the woman who served the tea.” c. Scrambled double-accusative sentence (SDA) Ofisu-de kakarichoo-o shokuin-ga

ocha-o dashita josei-ni

office-loc manager-acc employee-nom tea-acc served

teineini

woman-dat politely

shoukaishita. introduced “At the office, the employee politely introduced the manager to the woman who served the tea.” A slow-down related to clause-boundary processing is predicted at tea in the double accusative conditions (the DA and the SDA) in comparison to the same word in the ambiguous RC sentence (ARC). Moreover, woman should be read more slowly in the ARC because this is the first region that indicates the presence of an embedded clause in this sentence. In the

Case markers and clause boundaries

12

SDA, the accusative NP manager was scrambled to the position prior to employee, avoiding a potential confusion in the DA due to the adjacency of the two accusative NPs. (For experimental results and discussions on the processing of scrambling in Japanese, see Mazuka et al., in press; Miyamoto & Takahashi, in press; Yamashita, 1997). The experiment design included two unrelated conditions, which investigated word order differences of the two objects manager and tea (see Miyamoto & Takahashi, in press, Experiment 1, for details). (11) a. Ofisu-de shokuin-ga

kakarichoo-ni ocha-o dashita josei-o

office-loc employee-nom manager-dat tea-acc served hometa-to

teineini

woman-acc politely

Aiharasan-ga hanashiteita.

praised-that Aihara-nom

said

“At the office, Aihara said that the employee politely praised the woman who had served tea to the manager.” b. Ofisu-de shokuin-ga

ocha-o kakarichoo-ni dashita josei-o

office-loc employee-nom tea-acc manager-dat served hometa-to

teineini

woman-acc politely

Aiharasan-ga hanashiteita.

praised-that Aihara-nom

said

“At the office, Aihara said that the employee politely praised the woman who had served tea to the manager.” Five lists were created by distributing the twenty-five stimuli in a Latin Square design. Each participant saw exactly one of the lists intermixed with 53 unrelated foil items in pseudo-random order, so that at least one filler item intervened between any two test items. Procedure The experiment was conducted on a Power Macintosh computer running PsyScope (Cohen et al., 1993) with a button-box. Participants were timed in a phrase-by-phrase self-paced non-cumulative moving-window reading task (Just et al., 1982). The blank spaces in the sentences in (10) and (11) indicate the actual segmentation used in the self-paced reading

Case markers and clause boundaries

13

presentation (see also Table 1 below). Sentences were shown using Japanese characters (as in Appendix A) with the uniform-width font Osaka Toohaba 14. Stimuli segments initially appeared masked with dots, and participants pressed the leftmost button of the buttonbox to reveal each subsequent region of the sentence and cause all other regions to revert to dots. At the end of each sentence, a yes/no question appeared on a new screen, which participants answered by pressing one of the two rightmost buttons. No feedback was given. Corresponding data points were eliminated from the reading time analyses if the participant did not answer the comprehension question correctly. The experimental trials were preceded by one screen of instructions and eight practice trials. All sentences were presented on a single line. The experiment took participants approximately 20 minutes. Data analysis Analyses were conducted on comprehension question response accuracy and reading times. Residual reading times per region (Ferreira & Clifton, 1986) were derived by subtracting from raw reading times each participant’s predicted time to read regions of the same length (measured in number of characters), which in turn was calculated from a linear regression equation across all of a participant’s sentences in the experiment. The residual reading times were trimmed so that data points beyond three standard deviations from the relevant condition × region cell mean were discarded, which corresponded to less than 2.5% of the test data. The means and analyses presented below are based on the trimmed residual reading times. Analyses conducted on raw reading times yielded a similar pattern of results as the residual reading times analyses reported. An item was eliminated from the reading time analyses because all participants answered the comprehension question for the ARC condition incorrectly. In region 3, one participant’s data were not included in the subject analysis because all data points for one condition were lost during the trimming procedure. Overall, reading times per word in Japanese are slower than in English. However, this is probably due to different concepts of word in each language. In Japanese, a word (or rather,

Case markers and clause boundaries

14

a bunsetsu) is comprised of a content word and a functional particle (e.g., case markers, postpositions, complementizers are usually taken to be suffixes). In contrast, the reading times per word reported for English are usually averages over content words and functional words read separately.

Results Comprehension question response accuracy Comprehension performance in the ARC (72%) was reliably better than in the DA (62%; F1(1,34) = 4.95, P < 0.05; F2(1,24) = 4.52, P < 0.05) and marginally better than in the SDA sentences (61%; F1(1,34) = 4.04, P = 0.052; F2(1,24) = 4.72, P < 0.05). The results suggests that the presence of the two accusative NPs in the DA and SDA is unusual and more disruptive than the more common word order in the ARC. The SDA and the DA did not differ (F s < 1). Reading times The analyses with residual reading times yielded the following results for the three conditions in (10). (See Table 1 for a schematic representation of the sentences, and Figure 1 for the residual reading times per region.) ———————— Insert Table 1 About Here ————————–

———————— Insert Figure 1 About Here ————————– In regions 1 and 2, the three conditions did not differ (F s < 1.7). In region 3, the SDA was slower than the ARC in the analysis by subjects (F1 (1,33) = 10.24, P < 0.01; F2(1,23) = 3.29, P = 0.083) and slower than the DA in both types of analyses (F1 (1,33) = 11.2, P < 0.01; F2(1,23) = 20.83, P < 0.01). This may be due to scrambling (see Mazuka et al., in press, for a similar result) or to properties of nominative

Case markers and clause boundaries

15

NPs (Miyamoto & Takahashi, in press; and the general discussion below). In region 4, the ARC was faster than its two controls (vs. the SDA: F1 (1,34) = 7.66, P < 0.01; F2(1,23) = 7.79, P < 0.05; vs. the DA: F1 (1,34) = 9.21, P < 0.01; F2(1,23) = 12.18, P < 0.01). The SDA was only numerically slower than the DA (F s < 1). In region 5, the three conditions did not differ (F s < 1.9). In region 6 (the RC head), the ARC was slower than its two controls (vs. the SDA: F1(1,34) = 6.16, P < 0.05; F2(1,23) = 6.44, P < 0.05; vs. the DA: F1(1,34) = 11.14, P < 0.01; F2 (1,23) = 9.02, P < 0.01). The two double accusative conditions did not differ (F s < 1). In regions 7 and 8, the three conditions did not differ (F s < 1.6). In the two unrelated conditions presented in (11), the only reliable difference was observed at the fourth region as the direct object tea was read faster in (11a) when compared to the indirect object manager in (11b) (F1(1,34) = 6.42; P < 0.05; F2(1,22) = 4.45; P < 0.05; see Miyamoto & Takahashi, in press, for further details).

Discussion The results in regions 4 and 6 support the view that readers are constructing clausal structures based on the case markers on the NPs, thereby predicting the upcoming words necessary to complete the sentence. In region 4 (tea-acc), the DA and the SDA were slower than the ARC suggesting that a clause boundary is created at that point in those two conditions. If readers were not associating NPs to each other, this slow-down would have to be explained in terms of some independent factor. Although adjacency of the two accusative NPs could be a factor in the DA, it is inadequate to explain the slow-down in the SDA, in which the two accusative NPs are not adjacent. In region 6 (the RC head), the ARC was slower than the other two conditions as should be expected if a switch from a single-clause interpretation to a multi-clausal interpretation is taking place. In other words, the RC head is unexpected and the previously favored single-clause interpretation has to be reanalyzed to accommodate a new clause. In the

Case markers and clause boundaries

16

double accusative conditions, reanalysis is unnecessary because the RC head in region 6 is consistent with the clause boundary created in region 4. The next two experiments use the same type of rationale, but with a more generalized strategy to determine clause boundaries.

Local assignment of clause boundaries To create a clause boundary requires determining which NPs are co-arguments and should be associated with the same upcoming verb. In the following example, when the first two NPs are read, readers preferentially interpret this sequence as the beginning of a single-clause sentence which could be completed with a verb such as see (“The woman saw the old man.”). (12) Obasan-ga

toshiyori-o

woman-nom old-man-acc

←− gakuseiga-ga student-nom

When student-nom is detected, it is clear that the sentence involves more than one clause because there is no verb in Japanese that could take those three case-marked NPs as arguments. Readers have to create a clause boundary so that some NPs are going to be associated with the verb in the main clause, and the remainder should be associated with the verb in the embedded clause. The question then is which NPs are going to be assigned to which clause. The following strategy suggests that there is a preference for the accusative NP to remain associated with the first nominative NP and that a clause boundary is most likely created between old-man-acc and student-nom. Local assignment of clause boundaries (LACB): Assign the left boundary of a new clause at the point where it is first clear that this new clause is necessary for the interpretation of the sentence. Because student-nom is the first word that makes it clear that the segment involves more than one clause, this is where the boundary is preferentially set according to the LACB. The LACB can be justified in terms of a conservative reanalysis processes in which readers attempt to change the previous interpretation as little as possible (minimal revisions,

Case markers and clause boundaries

17

Frazier, 1990) or exactly as dictated by violated constraints (constraint driven changes, Miyamoto, 2002). The following sections discuss two ambiguous constructions and ways of creating unambiguous controls using the LACB.

Displacing one or two NPs during reanalysis In the processing of (13a) below, a single clause interpretation as in (13b) is initially favored, but when the noun girl is detected, it is clear that saw is part of a RC. (13) a. Subject reanalysis (SR) Obasan-ga [RC tj toshiyori-o kousaten-de woman-nom

mita] onnanokoj -ni koe-o-kaketa.

old-man-acc intersection-loc saw girl-dat

chatted

“The woman chatted with the girl who saw the old-man at the intersection.” b. Obasan-ga

toshiyori-o kousaten-de

mita.

woman-nom old-man-acc intersection-loc saw “The woman saw the old-man at the intersection.” The sentence in (13a) will be referred to as subject reanalysis (SR for short; following Hirose & A. Inoue, 1998) because only the subject woman has to be displaced out of the original interpretation shown in (13b). Much recent research has been devoted to the investigation of how the initial preference for (13b) can affect the processing of sentences such as (13a) (Gorrell, 1995; Hirose & A. Inoue, 1998; Mazuka & Itoh, 1995; Sturt & Crocker, 1996; inter alia). Of interest here is the observation that the reanalysis at girl in (13a) is easier than a similar reanalysis at taxi in (14) below (Mazuka & Itoh, 1995).

(14) Subject and object reanalysis (SOR) Obasani -ga toshiyori-o [RC proi tj kousaten-de woman-nom old-man-acc

mita] takushiij -ni noseta.

intersection-loc saw

taxi-dat

put

“The woman put the old-man in the taxi that she saw at the intersection.”

Case markers and clause boundaries

18

When processing (14), readers prefer a simple clause interpretation as the one in (13b). It is only at taxi-dat that it is clear that (14) involves a RC. However, contrary to (13a), not only the subject woman but also the object old-man has to the displaced from the original clause built.3 Accordingly, sentence (14) will be referred to as subject and object reanalysis (SOR; following Hirose & A. Inoue, 1998). The following section proposes ways of manipulating case markers in order to obtain baseline reading times for SR and SOR sentences.

Unambiguous control sentences for SRs and SORs The purpose of the following discussion is twofold. First, it introduces sentences in which the early positing of a clause boundary as dictated by case markers avoids a slow-down at the RC head. Second, it will be argued that the reading times in these sentences provide a baseline against which we can test the claim that reanalysis at the RC head is harder in SOR sentences than in SR sentences (Mazuka & Itoh, 1995). Two control sentences for SORs Assuming that clause boundaries are created according to the LACB, consider how the processing of (15) would proceed in comparison to the SOR. (15) Canonical control for the SOR Obasani -ga toshiyori-o [RC gakusei-ga tj kousaten-de woman-nom old-man-acc

mita] takushiij -ni noseta.

student-nom intersection-loc saw taxi-dat

put

“The woman put the old-man in the taxi that the student saw at the intersection.” In the SOR sentence, there should be a slow-down at taxi because this is the first point in which it is clear that the single clause interpretation is incorrect. In (15), however, a clause boundary must be inserted at student according to the LACB; and taxi is a natural continuation as the RC head. Thus, the reading time of taxi in (15) should be faster than the reading time of the same word in the SOR. The two sentences are not a minimal pair in that there is an extra NP in (15), namely student. But this should not be a problem in

Case markers and clause boundaries

19

the present case because, in general, the inclusion of an extra discourse entity makes the memory load heavier and consequently processing should be harder, whereas the prediction to be tested is for the RC head in (15) to be easier to process. One problem with (15) is that, after the clause boundary is assigned between old-man and student, reanalysis may take place at saw in order to lower the accusative NP into the embedded clause and satisfy the argument structure of this verb (see Miyamoto, 2002, for related discussion). In the following sentence, the accusative marked old-man is scrambled to the sentence initial position, thus avoiding reanalysis at saw. (16) Scrambled control for the SOR Toshiyori-o obasani -ga [RC gakusei-ga tj kousaten-de old-man-acc woman-nom

mita] takushiij -ni noseta.

student-nom intersection-loc saw

taxi-dat

put

“The woman put the old-man in the taxi that the student saw at the intersection.” Based on the results of Experiment 1, it is predicted that a slow-down occurs at student in the two controls for the SOR (namely, (15) and (16)) in order to posit a clause boundary. A self-paced reading study with sentences containing the NP sequences below provides supporting evidence (Yamashita, 1997; adverbs and adjectives are not shown). (17) a. Gakusei-ga sensei-ni

koibito-ga

student-nom teacher-dat girlfriend-nom b. Gakusei-ga sensei-ni

tegami-o

student-nom teacher-dat letter-acc c. Sensei-ni

gakusei-ga

tegami-o

teacher-dat student-nom letter-acc Reading times were longer at the third NP when it had nominative case (as in (17a)) rather than accusative case (as in (17b,c)) suggesting that a clause boundary is inserted at the third NP when it has nominative case. However, as Yamashita herself observed, there may be some property intrinsic to nominative NPs that makes them take longer to be processed than accusative NPs.

Case markers and clause boundaries

20

Yamashita also reported that there were no significant differences at any point when comparing (17b) to (17c). Thus, she concluded that the order in which the NPs are presented does not affect their reading time. However, recent results disconfirm this claim (Mazuka et al., in press; Miyamoto & Takahashi, in press). The processing time for the first three NPs in (15) and (16) can be compared in order to investigate Yamashita’s findings. If the slow-down at girlfriend in (17a) was exclusively caused by its nominative case marker, there should be no differences in the processing time of student in (15) and (16), given that this NP has nominative case in both sentences. However, if a reliable difference is found, it would be evidence that the order of the first two NPs are relevant in the processing of the third NP, thus providing further evidence against Yamashita’s claim that the order of case marked NPs does not influence their processing time. Moreover, the two nominative NPs are adjacent in (16), therefore this is another opportunity to test the hypothesis that the slow-downs investigated are not related to the insertion of clause boundaries, but rather that they are caused by the adjacency of NPs with the same case marker. If this is the case, student in (16) should be read more slowly than the same word in (15). A control sentence for SRs Given (15) and (16) as controls for the SOR, the sentence in (18) provides a baseline for the SR sentence. (18) Control for the SR Obasan-ga [RC toshiyori-ga tj kousaten-de woman-nom

mita] onnanokoj -ni koe-o-kaketa.

old-man-nom intersection-loc saw girl-dat

chatted

“The woman chatted with the girl who the old-man saw at the intersection.” The main difference between (18) and the SR is in the position of the gap (i.e., the empty position in the RC, indicated by tj ). The gap is in object position in (18), and in subject position in the SR; thus, the former sentence has an object gap and the latter has a subject gap. The nominative marker on old-man in (18) requires this NP to be the subject of saw and the RC gap has to be in the object position . Although not obligatory, there may be

Case markers and clause boundaries

21

a tendency to create a clause boundary at old-man-nom in (18), which could cause a slowdown at this point in order to shift from a single-clause to a multi-clausal interpretation (see the general discussion). When saw is processed, it is clear that the sentence contains an embedded clause because this verb cannot take two nominative NPs as arguments. It is assumed that RCs with subject gap (as in the SR) in Japanese are not intrinsically easier to process than RCs with object gap (as in (18)). This is not the case in languages such as English, in which it is well documented that subject-gap RCs are easier to process (King & Just, 1991; and references therein). However, it is not clear how the factors at work in the English constructions would affect their Japanese counterparts given that RCs are considerably different in the two languages. For example, in Japanese RCs, the linear distance to the RC head is shorter when the gap is in object position; RC gaps have been argued to have different properties in the two languages (in generative linguistics terminology, the gap in English is a trace left behind by movement, while the gap in Japanese has been claimed to be a phonologically-null pronominal such as pro, Saito, 1985; Murasugi, 1991). Moreover, in a self-paced reading study with a 2 by 2 design investigating the processing of RCs in Japanese, the position of the gap (in subject position or object position) revealed no significant differences, although RCs modifying the main clause object were reliably slower than RCs modifying the the main clause subject (Kanno & Nakamura, 2000, personal communication). The control sentence (18) in the present experiment is proposed as a tentative comparison which should be refined in future work. A self-paced reading experiment was conducted comparing the SR and SOR sentences to their respective controls.

Case markers and clause boundaries

22

Experiment 2 Method Participants Twenty-three native speakers of Japanese participated in the experiment. They were residents of the Kansai area and graduate students in Information Science at NAIST (Nara Institute of Science and Technology) and had not participated in Experiment 1. Materials Thirty sentences with five conditions each were created. (Appendix B contains the complete list of items used.) (19) a. SR (subject reanalysis) Obasan-ga [RC yoboyobo-no toshiyori-o woman-nom

feeble

guuzen-ni kousaten-de

mita]

old-man-acc by chance intersection-loc saw

onnanoko-ni isoide koe-o-kaketa. girl-dat

hurry chatted

“The woman chatted in a hurry with the girl who saw the feeble old-man at the intersection by chance.” b. Control for the SR Obasan-ga [RC

yoboyobo-no toshiyori-ga

woman-nom

feeble

guuzen-ni kousaten-de

mita]

old-man-nom by chance intersection-loc saw

onnanoko-ni isoide koe-o-kaketa. girl-dat

hurry chatted

“The woman chatted in a hurry with the girl who the feeble old-man saw at the intersection by chance.”

Case markers and clause boundaries

23

c. SOR (subject and object reanalysis) Obasan-ga

yoboyobo-no toshiyori-o [RC guuzen-ni kousaten-de

woman-nom feeble

old-man-acc

mita]

by chance intersection-loc saw

takushii-ni isoide noseta. taxi-dat

hurry put

“The woman put in a hurry the feeble old-man in the taxi that she saw at the intersection by chance.” d. Canonical control for the SOR Obasan-ga

yoboyobo-no toshiyori-o [RC gakusei-ga

woman-nom feeble

old-man-acc

kousaten-de

mita]

student-nom intersection-loc saw

takushii-ni isoide noseta. taxi-dat

hurry put

“The woman put in a hurry the feeble old-man in the taxi that the student saw at the intersection.” e. Scrambled control for the SOR Yoboyobo-no toshiyori-o feeble

obasan-ga [RC gakusei-ga

old-man-acc woman-nom

kousaten-de

mita]

student-nom intersection-loc saw

takushii-ni isoide noseta. taxi-dat

hurry put

“The woman put in a hurry the feeble old-man in the taxi that the student saw at the intersection.” The locative at the intersection is crucial to guarantee that taxi in the SOR sentence is interpreted as the object of saw as intended, and not as a place modifier (see Sturt & Crocker, 1996, for a different type of reanalysis that may take place at the main verb if at the intersection is removed). An adverbial phrase by chance was also included in the first three conditions to compensate for the extra word student-nom which is necessary in the last two conditions. The experiment should provide evidence that a clause boundary postulated according to case markers preempts reanalysis at a later point in the control sentences. Moreover, by

Case markers and clause boundaries

24

comparing the SR and SOR sentences to their respective controls, we can test the proposal that reanalysis in SR sentences is easier than in SOR sentences (Mazuka & Itoh, 1995). Five lists were created by distributing the thirty stimuli in a Latin Square design. Each participant saw exactly one of the lists intermixed with 52 unrelated foil items in pseudorandom order with at least one foil item intervening between two test items. Procedure and data analysis The procedure and data analysis were the same as in Experiment 1. Data points were trimmed as in Experiment 1, removing less than 2.5% of the test data. Analyses with raw reading times revealed similar trends as in the analysis with residual reading times reported below. Items were not included in the item analyses if they lacked data points for the conditions compared.

Results Comprehension question accuracy The comprehension performance in the SR (82% correct response) was reliably better than in its control (68%) in the analysis by subject (F1(1,22) = 6.48, P < 0.05; F2(1,29) = 3.98, P = 0.055). Performance in the SOR (70%) and in its canonical control (71%) did not differ (F s < 1). Performance in the SOR condition was reliably worse than in its scrambled control (78%) in the analysis by subjects (F1(1,22) = 4.97, P < 0.05; F2(1,29) < 1.5). The two controls for the SOR did not differ (F1 (1,22) = 2.65, P = 0.12; F2 (1,29) = 2.54, P = 0.12). Reading times The following are the results for the analyses with the residual reading times per region (see Table 2 and Figures 2 and 3). ———————— Insert Table 2 About Here ————————–

Case markers and clause boundaries

25

———————— Insert Figure 2 About Here ————————–

———————— Insert Figure 3 About Here ————————– In regions 1 and 2, the five conditions did not differ (F s < 1.9). In region 3, the control for the SR was slower than each of the other four conditions (all P s < 0.05); in particular, it was slower than the scrambled control for the SOR (F1(1,22) = 14.34, P < 0.01; F2 (1,28) = 8.05, P < 0.01). In region 4, student-nom in each control for the SOR was ready more slowly than bychance in each of the other three conditions (all P s < 0.01). Moreover, the canonical control for the SOR was slower than its scrambled counterpart (F1(1,22) = 6.62, P < 0.05; F2(1,29) = 5.81, P < 0.05). In region 5, the five conditions did not differ (F s < 2). In region 6 (the embedded verb), the SR and its control did not differ (F s < 1). The SOR was faster than its canonical control in the analysis by items (F1(1,22) = 2.59, P = 0.12; F2 (1,25) = 4.46, P < 0.05). The SOR was only numerically faster than its scrambled control (F1(1,22) = 3.47, P = 0.076; F2(1,25) = 2.17, P = 0.15). The two controls for the SOR did not differ (F s < 1.3). In region 7, the five conditions did not differ (F1 (4,88) = 1.48, P = 0.21; F2 (4,96) = 1.96, P = 0.11). In region 8, the five conditions did not differ (F s < 1). In region 9, the SR and its control did not differ (F s < 1). The SOR was slower than its canonical control in both types of analyses (F1(1,22) = 6.38, P < 0.05; F2(1,25) = 9.9, P < 0.01) and slower than its scrambled control in the subject analysis (F1 (1,22) = 4.4, P < 0.05; F2(1,25) = 4.14, P = 0.053). The two controls for the SOR did not differ (F s < 1.2).

Case markers and clause boundaries

26

Discussion The proposal that NPs are associated according to their case markers is supported by the slow-downs observed in regions 3 and 4 suggesting that the result that Yamashita (1997) obtained for (17) is likely to be due to a clause boundary being inserted. In region 3, the most informative comparison is between the control for the SR repeated below as (20a) and the scrambled control for the SOR in (20b). (20) Region:

1

2

a. Control for the SR:

Woman-nom feeble

b. Scrambled control for the SOR:

Feeble

3 old-man-nom

old-man-acc woman-nom

Although the NPs in region 3 in both conditions have nominative case, (20a) was read more slowly than (20b) at that point. In (20b), the accusative feeble old-man can be adjoined to the left of woman-nom and a new clause does not have to be built (see the general discussion section). In (20a), there are two alternative interpretations that may be competing causing the observed slow-down. Under one interpretation, old-man is the subject of an embedded clause. In the second interpretation, old-man is the object of an upcoming stative verb (e.g., “the woman likes the feeble old-man”). In region 4, the longer reading times in the canonical control for the SOR (repeated below as (21a)) when compared to the scrambled control for the SOR (21b) contradicts Yamashita’s proposal that the order of the NPs in a sequence does not affect their processing time. Moreover, if adjacency of the two nominative NPs were a factor, the prediction would be for student in (21b) to be read more slowly, contrary to what was actually observed. (21) Controls for the SOR Region:

1

2

a. Canonical control: Woman-nom feeble b. Scrambled control: Feeble

3

4

old-man-acc student-nom

old-man-acc woman-nom student-nom

A factor related to competition during the insertion of the clause boundary may explain the result in region 4. In (21b), there is only one possible position for the clause boundary — it must be between the two nominative NPs (assuming that a clause may not contain

Case markers and clause boundaries

27

more than one subject NP, but see Kuno, 1973, for some well-known exceptions). In (21a), in contrast, there are two possible points at which the boundary could be inserted, namely immediately before student or immediately before feeble old-man. The LACB favors the clause boundary at student; but the availability of the competing alternative at feeble oldman may cause the relative slow-down observed. The explanation may also be cast in terms of the clarity of what has to be done to correct the representation (Fodor & A. Inoue, 1994; see also the discussion section of Experiment 3). According to such competition explanations, the slow-downs observed at clause boundaries may not be due to the complexity of the structure that needs to be inserted, but rather they may reflect the competition between alternative interpretations. The exact nature of the slow-down will not be explored in the present paper. Although the pattern of results up to region 4 can be explained within the framework assumed here, the lack of reliable differences in region 7 (the RC head) is not compatible with the present claims. The prediction that the two controls for the SOR would be faster than the SOR is only observed in the last region of the sentences. One possible explanation for this result is that the occurrence of the second nominative NP in the control conditions was so disruptive that participants were still confused when reading region 7. When asked about their general impression about the experiment, several participants complained that the sentences containing two nominative NPs were particularly confusing. Previous off-line results indicate that sequences of multiple nominative NPs are complex to process, and that they become easier when the nominative marker on one of the NPs is replaced by the topic marker wa (Uehara, 1997; also M. Inoue, 1991, for eye-tracking results comparing nominative and topic subjects displaced during reanalysis). Thus, a second version of Experiment 2 was conducted in which the main-clause subject was topicalized.

Experiment 3 The topic marker wa can replace any case marker in Japanese, but there is an overwhelming preference for a wa-marked NP to be interpreted as the subject of the main clause. Hence, substituting a wa marker for the first nominative ga marker in the sentences in Experiment 2

Case markers and clause boundaries

28

should not introduce any further ambiguity, and the resulting sentences should not be harder than the original sentences. Supporting this assumption, on-line pretests did not detect any reliable differences between the SOR sentence in (19c) and its topicalized counterpart to be used below. Moreover, previous eye-tracking results suggest that reanalysis of a topicalized NP is easier than reanalysis of a nominative NP in sentences like the following (M. Inoue, 1991; see also Mazuka & Itoh, 1995, for a related prediction comparing SORs with and without topicalization). (22) Tatsuo-ga

Hanako-ni himitsu-o oshieta yuujin-o nagutta.

Tatsuo-nom Hanako-dat secret-acc told

fried-acc hit

“Tatsuo hit a friend who told a secret to Hanako.” As observed earlier, there is an initial preference to assume that the first three NPs in (22) belong to the same clause (as in “Tatsuo told a secret to Hanako.”). It is only at friend that it is clear that the sentence involves a RC. However, reanalysis at this point becomes easier if Tatsuo is marked with the topic marker wa instead of the nominative ga (the difference was statistically significant in first-pass and total reading time analyses as well as in the number of regressions; M. Inoue, 1991).

Method Participants Twenty-seven native speakers of Japanese participated in the experiment and had not taken part in the previous experiments. They were residents of the Kansai area and graduate students in Information Science at NAIST (Nara Institute of Science and Technology). Materials The conditions were the same as in Experiment 2 (see (19) above), except that in the present experiment, the nominative marker on woman was replaced with the topic marker wa. (See Table 3 for a schematic representation of the sentences and their segmentation for the selfpaced reading presentation. Appendix B contains the complete list of items used.)

Case markers and clause boundaries

29

Five lists were created by distributing the thirty stimuli in a Latin Square design. Each participant saw exactly one of the lists intermixed with 90 unrelated foil items in pseudorandom order. Procedure and data analysis The procedure and the data analysis were the same as in Experiment 1. Data trimming was conducted as in Experiment 1, removing less than 2.5% of the test data. Analyses with raw reading times were also computed and revealed similar trends as in the analysis with residual reading times reported below. Items were not included in the item analyses if they lacked data points for the conditions compared. For a better comparison with Experiment 2, in which only 23 native speakers had participated, analyses were also conducted with the first 23 participants that took part in Experiment 3, and the trends were the same as in the analyses with 27 participants reported below.

Results Comprehension question accuracy Comprehension performance in the SR condition (81%) and its control (73%) did not differ (F1(1,26) = 2.51, P = 0.12; F2(1,29) = 1.99, P = 0.17). Performance in the SOR condition (65%), its canonical control (72%) and its scrambled control (73%) did not differ (F1 (2,52) = 1.98, P = 0.15; F2 (2,58) < 1). Reading times The results of the analyses using residual reading times were as follows (see Table 3 and Figures 4 and 5). ———————— Insert Table 3 About Here ————————–

Case markers and clause boundaries

30

———————— Insert Figure 4 About Here ————————–

———————— Insert Figure 5 About Here ————————– In regions 1 and 2, the five conditions did not differ (all F s < 1.1). In region 3, the five conditions did not differ (F1 (4,104) = 1.8, P = 0.13; F2(4,104) = 2.08, P = 0.088). In region 4, the two controls for the SOR did not differ (F s < 1.9), but each one was reliably slower than each of the other three conditions (P s < 0.05). In regions 5 and 6, the five conditions did not differ (F s < 2.1). In region 7 (the RC head), there was an interaction effect when comparing the SR and its control to the SOR and its canonical control, but it was reliable in the subject analysis only (F1(1,26) = 5.48, P < 0.05; F2 (1,26) = 3.57, P = 0.70). The interaction was reliable in both analyses when the canonical control for the SOR was replaced by the scrambled control (F1(1,26) = 10.73, P < 0.01; F2(1,27) = 6.91, P < 0.05). Pairwise results were as follows. The SR and its control did not differ (F s < 1). The SOR was slower than its canonical control in the analysis by subjects (F1(1,26) = 9.47, P < 0.01; F2(1,27) = 3.75, P = 0.063) and slower than its scrambled control in both analyses (F1(1,26) = 20.55, P < 0.01; F2(1,28) = 11.17, P < 0.01). The scrambled control for the SOR was faster than its canonical counterpart in the item analysis (F1 (1,26) = 3.42, P = 0.076; F2 (1,28) = 4.68, P < 0.05). The five conditions did not differ in regions 8 and 9 (F s < 1.8).

Discussion Prior to the embedded verb, the overall reading-time pattern observed is similar to the one in Experiment 2 but some of the differences were not statistically significant in the present experiment. The reading times in region 7 support the proposal that the early positing of a clause

Case markers and clause boundaries

31

boundary in the two controls for the SOR makes the processing of the RC head easier compared to the SOR. However, because the SOR also requires inserting a gap (indicated by pro in (14)) in the subject position of the RC (which is occupied by student in the controls), it is conceivable that its relative slow-down involves more than just the switching from the monoclausal to the multiclausal interpretation. Further research is needed in order to address this possibility. The interaction observed at the RC head occurs because the SOR is slower than either of its controls whereas the SR does not differ from its control. This result supports the observation that reanalysis is harder in SOR sentences than in SR sentences (Mazuka & Itoh, 1995). However, given the differences between the test sentences and their respective controls (an extra NP in the controls for the SOR, and the position of the gap in the control for the SR), the result should be interpreted with caution. Moreover, in order to test Mazuka and Itoh’s proposal that reanalysis in SOR sentences is harder because two NPs are displaced (in contrast, only one NP is expelled in the SR sentences), it would be desirable to compare SR and SOR sentences with other types of controls and to test other constructions that also vary in the number of NPs displaced during reanalysis. Animacy of the relative clause head It is beyond the scope of this paper to discuss the various possible factors that may be involved in the reanalysis process of SOR sentences (Mazuka & Itoh, 1995; for a discussion on structural factors see Gorrell, 1995; Sturt & Crocker, 1996; for prosodic contours of SOR and SR sentences, see Hirose et al., 1999). The following considers the possibility that information in the RC head is decisive to determine reanalysis difficulty. Based on the proposal that the clarity of the disambiguating signal is the only source of difficulty during reanalysis (Fodor & A. Inoue, 1994), it could be suggested that the slowdown observed at the RC head taxi in the SOR sentence occurs because this noun can be interpreted as the agent or as the theme of saw, and is compatible with an object-gap RC as well as a subject-gap RC. Thus, the slow-down detected may stem from the uncertainty as to which interpretation to favor. In support of this proposal, Hirose and Inoue (1998) provided

Case markers and clause boundaries

32

evidence that the difficulty in processing SOR sentences can be modulated by varying the animacy of the RC head. More specifically, reanalysis was easy when the RC head was an inanimate noun consistent with the object gap interpretation alone. The following provides evidence that the clarity of the disambiguating segment alone is not enough to explain the result in Experiment 3. In five of the thirty items used in Experiment 3, the RC head in the SOR condition could be interpreted as being the subject or the object of the RC. If Fodor and colleagues are correct, reanalysis should be easy in the remaining twenty-five items in which the RC head could only be interpreted as the object of the RC. Analyses were conducted in region 7 (the RC head) excluding the five items with ambiguous RC head. The results of the analyses using residual reading times were as follows. The SOR condition was slower than its canonical control in the analysis by subjects (F1(1,26) = 7.47, P < 0.05; F2(1,23) = 2.30, P < 0.11) and reliably slower than its scrambled control in both analyses (F1(1,26) = 17.78, P < 0.01; F2 (1,23) = 9.15, P < 0.01). The present analyses suggest that even when the RC head unambiguously indicate the type of RC to be built, there is still a relative slow-down in the SOR sentences (see also Tokimoto, 1999, in which unambiguous SORs were found to be harder than unambiguous SRs). In short, the clarity of the signal may be one factor that affects reanalysis difficulty, but it is unlikely to be the only factor involved.

General discussion The present results support the proposal that NPs in verb final clauses are associated incrementally within a partial interpretation even when no verb has been processed. Together with previous findings (Bader & Lasser, 1994; Den & M. Inoue, 1997; Kamide & Mitchell, 1999; Kamide et al., 2000; Miyamoto et al., 1999), the reported data build a strong case against head-driven parsing in the processing of head-final phrases. Converging evidence comes from the processing of non-canonical word orders, in particular, of scrambled NPs, in which difficulty occurs before the verb is read (Mazuka et al., in press; Miyamoto & Takahashi, in press). Although animacy and plausibility play crucial roles, case markers seem to be

Case markers and clause boundaries

33

the single most important piece of information used in this process. This naturally raises questions about the exact nature of each type of case marker and how each one can affect the processing of NPs and clauses in general. The following provides some possible ways of reconciling the syntax literature with the observation in Experiment 2 that the nominative ga marker is particularly salient. The reading times of the controls sentences in Experiment 2 suggest that NP sequences with two nominative ga-marked NPs are confusing to process. The same difficulty was not observed in Experiment 3 in which one of the ga markers was replaced with a wa marker (see Uehara, 1997, for a similar contrast in an off-line task using a within subjects design; also M. Inoue, 1991, for related eye-tracking evidence). Discourse complexity aside (see neutral description, topicalization and exhaustive readings with ga markers and wa markers in Kuno, 1973; also Kim, 1999, for a related discussion on the processing of Korean), one difference between ga NPs and wa NPs is that the latter type of NPs cannot be the subjects of embedded clauses. In processing terms, this suggests that a wa NP must be the subject of the main verb, in general, the last verb in the sentence. In contrast, a ga NP is preferentially interpreted as the subject of the most local verb (i.e., the first upcoming verb). Therefore, in a NP sequence with two ga NPs, there are two potential subjects for the local verb. In a sequence with a wa NP followed by a ga NP, there is no competition for the local verb because only the ga NP can be assigned to it. Difficulty with ga NPs does not seem to be restricted to multiple occurrences. In general, ga NPs seem to be read more slowly than NPs with other case markers. One possible reason for the markedness of ga NPs is that they may be used as clause boundary markers as discussed below.

Ga NPs as clause boundary markers In the following, it is argued that ga NPs mark the left periphery of a clause, thus serving as approximate markers for the beginning of a new clause. The NPs in (23a) are in the canonical word order with the subject before the object; whereas in (23b), the accusative NP was scrambled to the beginning of the sentence.

Case markers and clause boundaries

34

(23) a. Canonical word order: Obasan-ga

toshiyori-o mita.

woman-nom old-man-acc saw “The woman saw the old man.” b. Scrambled word order: Toshiyori-o obasan-ga

mita.

old-man-acc woman-nom saw “The woman saw the old man.” The usual assumption in transformational approaches to Japanese syntax is that the nominative subject cannot be scrambled and remains fixed in its structural position in the initial segment of the clause (the equivalent of Spec-IP in Government and Binding terminology; Saito, 1985). Moreover, a nominative subject can only occur as the argument of a tensed predicate, therefore this type of NP necessarily requires an inflection related node (Takezawa, 1987). Given those properties and the lack of better indicators in Japanese, it is likely that nominative NPs are used to signal the beginning of a tensed clause. Note that nominative NPs do not always mark the exact beginning of a clause (for example, a constituent can be adjoined to the left of a nominative NP as in (23b)); nevertheless, they provide a fixed point around which other NPs in the sentence can be interpreted. In the discussion of (6), it was suggested that a sequence containing a ga NP, a dative NP and an accusative NP may predict an underspecified ditransitive verb. Similarly, it is conceivable that the processing of a ga NP may predict an underspecified IP node. In this case, the preference to associate a ga NP to the most local verb may originate from this relation between this type of NP and the IP nodes it requires.

Conclusion Three self-paced reading experiments were reported arguing that case markers determine how NPs are to be associated together within a partial interpretation of the sentence. Even when NPs with the same case marker were compared across conditions, a slow-down was

Case markers and clause boundaries

35

observed in the NP sequence that required a clause boundary to be posited, and still longer reading times were detected when more than one alternative position was available for the clause boundary to be inserted. The present results argue against models that require a verb to be present in order for a NP sequence to be interpreted, and suggest instead that a complex representation can be built despite the absence of a licensing head.

Case markers and clause boundaries

36

References Abney, S. P. (1989). A computational model of human parsing. Journal of Psycholinguistic Research, 18, 129-144. Ades, A. E., & Steedman, M. (1982). On the order of words. Linguistics and Philosophy, 4, 517-558. Bader, M., & Lasser, I., (1994). German verb-final clauses and sentence processing: Evidence for immediate attachment. In C. Clifton Jr., L. Frazier, & K. Rayner (Eds.), Perspectives on Sentence Processing (pp. 225-242). Mahwah, NJ: Lawrence Erlbaum. Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press. Chomsky, N. (1981). Lectures on government and binding. Dordrecht, the Netherlands: Foris. Cohen, J.D., MacWhinney, B., Flatt, M., & Provost, J. (1993). PsyScope: An interactive graphic system for designing and controlling experiments in the psychology laboratory using Macintosh computers. Behavior Research Methods, Instruments and Computers, 25, 257-271. Den, Y. & Inoue, M. (1997). Disambiguation with verb-predictability: Evidence from Japanese garden-path phenomena. Proceedings of the 19th Annual Conference of the Cognitive Science Society (pp. 179-184). Dryer, M. (1989). Large linguistic areas and language sampling. Studies in Language, 13, 257-292. Ferreira, F., & Clifton, C., Jr. (1986). The independence of syntactic processing. Journal of Memory and Language, 25, 348-368. Fodor, J. D., & Inoue, A. (1994). The diagnosis and cure of garden paths. Journal of Psycholinguistic Research, 23, 407-434.

Case markers and clause boundaries

37

Frazier, L. (1987). Sentence processing: A tutorial review. In M. Coltheart (Ed.), Attention and Performance XII (pp. 559-586). Hillsdale, NJ: Lawrence Erlbaum. Frazier, L. (1990). Identifying structure under X0. In G. Booij, & J. van Marle (Eds.), Yearbook of Morphology, 3 (pp. 87-105). Amsterdam, the Netherlands: Foris Publications. Gibson, E. (1991). A computational theory of human linguistic processing: Memory limitations and processing breakdown. Unpublished doctoral dissertation, Carnegie Mellon University, Pittsburgh, PA. Gorrell, P. (1995). Syntax and parsing. Cambridge, UK: Cambridge University Press. Harada, S. I. (1973). Counter equi NP deletion. Annual Bulletin of the Research Institute of Logopedics and Phoniatrics (University of Tokyo), 7, 113-147. Hirose, Y., & Inoue, A. (1998). Ambiguity of reanalysis in parsing complex sentences in Japanese. In D. Hillert (Ed.), Sentence Processing: A Crosslinguistic Perspective (pp. 71-93). Syntax and Semantics, 31. San Diego, CA: Academic Press. Hirose, Y., Inoue, A., Fodor, J. D., & Kakehi, K. (1999). Evidence for an effect of implicit prosodic phrasing on ambiguity resolution in reanalysis. Talk presented at the 12th Annual CUNY Conference on Human Sentence Processing. New York, NY. Inoue, A. (1991). A comparative study of parsing in English and Japanese. Unpublished doctoral dissertation, University of Connecticut, Storrs, CT. Inoue, A., & Fodor, J. D. (1995). Information-paced parsing of Japanese. In R. Mazuka, & N. Nagai (Eds.), Japanese Sentence Processing (pp. 9-63). Mahwah, NJ: Lawrence Erlbaum. Inoue, M. (1991). Bun-no tougo shori-ni okeru joshi-wa-no kinou. Proceedings of the 33nd Nippon Kyouiku Shinri Gakkai (pp. 55-56; in Japanese). Just, M. A., Carpenter, P. A., & Woolley, J. D. (1982). Paradigms and processes in reading comprehension. Journal of Experimental Psychology: General, 3, 228-238.

Case markers and clause boundaries

38

Kamide, Y., Altmann, G. T. M., & Haywood, S. L. (2000). Predictive eye-movements in incremental processing of head-final structures. Poster presented at the 13th Annual CUNY Conference on Human Sentence Processing, San Diego, CA. Kamide, Y., & Mitchell, D. C. (1999). Incremental pre-head attachment in Japanese parsing. Language and Cognitive Processes, 14, 631-662. Kanno, K., & Nakamura, M. (2000). Processing of relative clauses by Japanese native speakers and L2 learners. Manuscript. University of Hawaii. Kim, Y. (1999). The effects of case marking information on Korean sentence processing. Language and Cognitive Processes, 14, 687-714. King, J., & Just, M. A. (1991). Individual differences in syntactic processing: The role of working memory. Journal of Memory and Language, 30, 580-602. Kuno, S. (1973). The structure of the Japanese language. Cambridge, MA: MIT Press. Kuroda. S.-Y. (1992). Japanese Syntax and Semantics: Collected Papers. Dordrecht, the Netherlands: Kluwer Academic Publishers. Lewis, R., & Nakayama, M. (in press). Syntactic and positional similarity effects in the processing of Japanese embeddings. In M. Nakayama (Ed.), Sentence Processing in East Asian Languages. Stanford, CA: CSLI. Marantz, A. (1984). On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. Marslen-Wilson, W., & Tyler, L. K. (1980). The temporal structure of spoken language understanding. Cognition, 8, 1-71. Mazuka, R., & Itoh, K. (1995). Can Japanese speakers be led down the garden-path? In R. Mazuka, & N. Nagai (Eds.), Japanese Sentence Processing (pp. 295-329). Mahwah, NJ: Lawrence Erlbaum.

Case markers and clause boundaries

39

Mazuka, R., Itoh, K., & Kondo, T. (in press). Cost of scrambling in Japanese sentence processing. In Nakayama, M. (Ed.), Sentence Processing in East-Asian Languages. Stanford, CA: CSLI. Miller, G. (1963). The magical number seven, plus or minus two: Some limits on our capacity for processing information. In R. Luce, R. Bush, & E. Galanter (Eds.), Readings in Mathematical Psychology, 1 (pp. 135-151). New York, NY: John Wiley & Sons. Miyamoto, E. T. (2002). Reanalysis of clause boundaries in Japanese as a constraint-driven process. Unpublished manuscript, Nara Institute of Science and Technology, Ikoma, Nara, Japan. Miyamoto, E. T., Gibson, E., Pearlmutter, N. J., Aikawa, T., & Miyagawa, S. (1999). A U-shaped relative clause attachment preference in Japanese. Language and Cognitive Processes, 14, 663-686. Miyamoto, E. T., & Takahashi, S. (in press). Sources of difficulty in processing scrambling in Japanese. In M. Nakayama (Ed.), Sentence Processing in East-Asian Languages. Stanford, CA: CSLI. Murasugi, K. (1991). Noun phrases in Japanese and English: A study in syntax, learnability and acquisition. Unpublished doctoral dissertation, University of Connecticut, Storrs, CT. Pritchett, B. L. (1991). Head position and parsing ambiguity. Journal of Psycholinguistic Research, 20, 251-270. Pritchett, B. (1992). Grammatical competence and parsing performance. Chicago, IL: The University of Chicago Press. Saito, M. (1985). Some asymmetries in Japanese and their theoretical implications. Unpublished doctoral dissertation, MIT, Cambridge, MA.

Case markers and clause boundaries

40

Scheepers, C., Hemforth, B., & Konieczny, L. (1999). Incremental processing of German verb-final constructions: Predicting the verb’s minimum (!) valency. Proceedings of the Second International Conference on Cognitive Science (ICCS/JCSS99). Sturt, P., & Crocker, M. (1996). Monotonic syntactic processing: A cross-linguistic study of attachments and reanalysis. Language and Cognitive Processes, 11, 449-494. Sturt, P., Pickering, M. J., & Crocker, M. (1999). Structural change and reanalysis difficulty in language comprehension. Journal of Memory and Language, 40, 136-150. Suh, S. (1994). The syntax of Korean and its implications for parsing theory. Unpublished doctoral dissertation, University of Maryland, College Park, MD. Takezawa, K. (1987). A configurational approach to case-marking in Japanese. Unpublished doctoral dissertation, University of Washington, Seattle, WA. Tokimoto, S. (1999). Serial sentence processing in Japanese. Proceedings of the Second International Conference on Cognitive Science (ICCS/JCSS99). Trueswell, J. C., Tanenhaus, M. K, & Garnsey, S. M. (1994). Semantic influences on parsing: Use of thematic role information in syntactic ambiguity resolution. Journal of Memory and Language, 33, 285-318. Tsujimura, N. (Ed.). (1999). The Handbook of Japanese Linguistics. Oxford, UK: Blackwell. Uehara, K. (1997). Judgements of processing load in Japanese: The effect of NP-ga sequences. Journal of Psycholinguistics Research, 26, 255-263. Yamashita, H. (1994). Processing of Japanese and Korean. Unpublished doctoral dissertation, Ohio State University, Columbus, OH. Yamashita, H. (1997). The effects of word-order and case marking information on the processing of Japanese. Journal of Psycholinguistic Research, 26, 163-188.

Case markers and clause boundaries

41

Author Note I would like to thank the participants in the meetings of the sentence processing laboratory at MIT (1995-1998) and the audiences at the 10th and 11th CUNY Conference on Human Sentence Processing for comments on presentations of this work at various stages. I would also like to thank Takako Aikawa, Janet D. Fodor, Ted Gibson, Yuki Hirose, Neal Pearlmutter, Shoichi Takahashi and Hiroko Yamashita for insightful discussions, and Yuko Maekawa and Lisa Shimizu for invaluable help developing the items used in the experiments. All remaining errors are my own. I am also indebted to professor Yuji Matsumoto for allowing me to use his laboratory facilities at NAIST during the Summer of 1996. Special thanks to Takehito Utsuro who was instrumental in recruiting participants at NAIST, where Experiments 2, 3 and several pre-studies leading to the present proposal were conducted. Funding for the project was provided by the MIT-Japan Program and by the JST/MIT joint international research project “Mind/Articulation”.

Case markers and clause boundaries

42

Footnotes 1

The following abbreviations will be used to indicate Japanese particles (e.g., case mark-

ers, postpositions): nom for nominative, acc for accusative, dat for dative, loc for locative, top for topic. 2

The two factors may be unifiable as a single constraint according to which nominative

NPs preferentially coincide with the beginning of a clause. 3

In (14), the NP woman is said to have been displaced or expelled, even though it is

eventually interpreted as the subject of saw through a co-indexed empty pronoun, pro.

Case markers and clause boundaries

43

Appendix A. The following were the items used in Experiment 1. Slashes indicate the segmentation used in the self-paced reading presentation. The initial segment of the items was partially based on items in Yamashita (1994), Appendix A. For the first item, all five conditions are presented, including the two conditions not discussed in the present paper. For the other items, only the ARC sentence is provided, based on which the DA and SDA can be derived. Item 9 was the item eliminated from the reading time analyses because all participants answered the comprehension question for the ARC incorrectly.

1a. ARC オフィスで / 職員が / 係長に / お茶を / 出した / 女性を / 丁寧に / 紹介した。

1b. DA オフィスで / 職員が / 係長を / お茶を / 出した / 女性に / 丁寧に / 紹介した。

1c. SDA オフィスで / 係長を / 職員が / お茶を / 出した / 女性に / 丁寧に / 紹介した。

1d. Unrelated condition オフィスで / 職員が / 係長に / お茶を / 出した / 女性を / 丁寧に / 誉めたと / 相原さんが / 話していた。

1e. Unrelated condition オフィスで / 職員が / お茶を / 係長に / 出した / 女性を / 丁寧に / 誉めたと / 相原さんが / 話していた。

2a. バーで / 訪問客が / やくざに / 花びんを / 見せた / ホステスを / ひそかに / 殺させた。 3a. 教室で / 先生が / 大学生に / 公式を / 説明した / 研究者を / 簡単に / 紹介した。 4a. 喫茶店で / 不良が / 友人に / 後輩を / 紹介した / オーナーを / 思いきり / なぐらせた。 5a. 京都駅で / 少年が / 係員に / 荷物を / 渡した / 駅員を / 徹底的に / 捜させた。 6a. 食堂で / 運転手が / おばさんに / 定食を / 注文した / 若者達を / すぐ / 告訴させた。 7a. 田舎で / 作家が / アシスタントに / カメラを / 譲った / 写真家を / すぐに / 首にさせた。 8a. 駐車場で / 主人が / マネージャーに / 車を / まかせた / 有名人を / 堂々と / 説得させた。 9a. 自宅で / お兄さんが / 友人に / ビデオを / 貸した / 仲間を / ためらいがちに / 紹介した。 10a. 下町で / おばあさんが / 米屋に / 借金を / 払った / 酒屋を / 簡単に / 追い出させた。 11a. カラオケで / 歌手が / オーナーに / 歌を / 捧げた / お客を / 優しく / 引き会わせた。 12a. 大阪で / 女優が / 監督に / 手袋を / 投げた / 観客を / 一生懸命 / 思いださせた。

Case markers and clause boundaries 13a. 駅で / 婦人が / 長男に / かばんを / 預けた / 友達を / 静かに / 紹介した。 14a. 会議で / 社長が / 部長に / 新製品を / 披露した / 副部長を / 怒って / やめさせた。 15a. 農家で / 老人が / 一人娘に / 土地を / 残した / 親戚を / すぐに / 呼び出させた。 16a. デパートで / 店員が / 女の子に / 洋服を / 見せていた / 高校生を / 長い間 / 見はらせた。 17a. 運動会で / 役員が / お母さん達に / お菓子を / 焼いた / 子供達を / 大声で / 呼ばせた。 18a. 広島で / 政治家が / 校長に / ピアノを / 届けた / 市長を / 無理に / 推薦させた。 19a. バーで / 音楽家が / 踊子に / 花束を / あげた / 知人を / さりげなく / 紹介した。 20a. 長崎で / 少女が / お姉さんに / おみやげを / 頼んだ / 級友を / 楽し気に / 会わせた。 21a. レストランで / 男性が / 恋人に / 指輪を / おくった / 売り子を / すばやく / 誘拐させた。 22a. 公園で / 男の子が / 両親に / キャンディを / ねだった / 仲良しを / 元気よく / 会わせた。 23a. 研究室で / 教授が / 先輩に / 壷を / 作った / 後輩を / 周到に / おとし入れさせた。 24a. 横浜で / 女優が / 母親に / 車を / 買った / 男性を / 親切に / 世話させた。 25a. 大学で / 教授が / 学生に / 古文書を / 貸した / 図書館司書を / 急いで / 呼び出させた。

44

Case markers and clause boundaries

45

Appendix B. The following items were used in Experiment 3. Slashes indicate the segmentation used in the self-paced reading presentation. For the first item, all five conditions are provided. For the remaining items, only the SR and the canonical control for the SOR are shown, and from them the other three conditions can be derived. In the analysis with the unambiguous RC heads, items 1, 18, 23, 25 and 29 were not included because the RC head in the SOR condition of these items can be interpreted as the subject or the object of the RC. The same items were used in Experiment 2, except that the topic wa of the matrix subject was replaced with the nominative ga in all conditions.

1a. SR おばあさんは / よぼよぼの / 年寄りを / 偶然に / 交差点で / 見た / 女の子に / 急いで / 声をかけた。

1b. Control for the SR おばあさんは / よぼよぼの / 年寄りが / 偶然に / 交差点で / 見た / 女の子に / 急いで / 声をかけた。

1c. SOR おばあさんは / よぼよぼの / 年寄りを / 偶然に / 交差点で / 見た / タクシーに / 急いで / 乗せた。

1d. Canonical control for the SOR おばあさんは / よぼよぼの / 年寄りを / 学生が / 交差点で / 見た / タクシーに / 急いで / 乗せた。

1e. Scrambled control for the SOR よぼよぼの / 年寄りを / おばあさんは / 学生が / 交差点で / 見た / タクシーに / 急いで / 乗せた。

2a. 店員は / 怪しい / 男を / ちらっと / 通りで / 見た / 女性に / その場で / お礼を言った。 2d. 店員は / 怪しい / 男を / 運転手が / 通りで / 見た / 交番に / その場で / 突き出した。 3a. 店員さんは / 酔っ払いの / おじさんを / じっと / パチンコ屋で / 狙っていた / ホームレスに / 怒って / 話しをした。

3d. 店員さんは / 酔っ払いの / おじさんを / 乞食が / パチンコ屋で / 狙っていた / パチンコ台に / 怒って / 押し付けた。

4a. 公務員は / 太った / 政治家を / 静かに / 長いこと / 見つめていた / 係員に / 確かに / 悪口を言った。 4d. 公務員は / 太った / 政治家を / 大臣が / 長いこと / 見つめていた / 椅子に / 確かに / すわらせた。 5a. 心臓病学者は / 年寄りの / 病人を / 突然に / 廊下で / 見た / 看護婦に / 丁寧に / 注意した。 5d. 心臓病学者は / 年寄りの / 病人を / 助手が / 廊下で / 見た / 車椅子に / 丁寧に / 乗せた。

Case markers and clause boundaries

46

6a. 警察官は / やくざの / 幹部を / 内緒で / 空港で / 探していた / 女の人に / うまく / 尋問した。 6d. 警察官は / やくざの / 幹部を / 探偵が / 空港で / 探していた / 飛行機に / うまく / 閉じ込めた。 7a. 男の子は / 黒い / 犬を / どきどきと / いつも / 怖がっていた / おばあさんに / とうとう / 話しかけた。 7d. 男の子は / 黒い / 犬を / おばさんが / いつも / 怖がっていた / お化け屋敷に / とうとう / 入らせた。 8a. 看護婦は / 背の低い / 患者さんを / 一生懸命 / 医務室で / 探していた / 学生に / 急いで / 呼びかけた。 8d. 看護婦は / 背の低い / 患者さんを / 事務員が / 医務室で / 探していた / 名簿に / 急いで / 入れた。 9a. 女子大生は / 内気な / 会社員を / こっそり / 奈良で / 訪ねた / 踊子に / 突然 / はち合わせた。 9d. 女子大生は / 内気な / 会社員を / いとこが / 奈良で / 訪ねた / お寺に / 突然 / 行かせた。 10a. お兄さんは / 出来の悪い / 弟を / 必死で / 図書館で / 探していた / 家庭教師に / すぐに / 謝った。 10d. お兄さんは / 出来の悪い / 弟を / 後輩が / 図書館で / 探していた / 勉強部屋に / すぐに / 閉じ込めた。 11a. 老人は / 酔っ払いの / 暴走族を / 偶然に / 高速道路で / 見た / 学生に / 頭から / 衝突された。 11d. 老人は / 酔っ払いの / 暴走族を / ひ孫が / 高速道路で / 見た / 電柱に / 頭から / 衝突させてしまった。 12a. 社長は / 貧乏な / お客さんを / わざと / 長期間 / 無視していた / 社員に / さりげなく / 注意した。 12d. 社長は / 貧乏な / お客さんを / 社員が / 長期間 / 無視していた / 取引に / さりげなく / 誘いました。 13a. 係長は / 若い / お客を / こっそり / 旅行会社で / 気に入っていた / ガイドに / 喜んで / 案内させた。 13d. 係長は / 若い / お客を / ガイドが / 旅行会社で / 気に入っていた / ツアーに / 喜んで / 参加させた。 14a. エージェントは / 新人 / 歌手を / 一目で / テレビ局で / 気に入った / カメラマンに / 何回も / 話しを した。

14d. エージェントは / 新人 / 歌手を / 友人が / テレビ局で / 気に入った / テレビ番組に / 何回も / 出演さ せた。

15a. 写真家は / 内気な / モデルを / 内緒で / 一目で / 気に入った / 会社員に / 親切に / けちをつけた。 15d. 写真家は / 内気な / モデルを / 社員が / 一目で / 気に入った / 喫茶店に / 親切に / 誘った。 16a. 美容師は / お金持ちの / お客を / ぼんやり / 鏡で / 見ていた / アシスタントに / 素早く / カットをや らせた。

16d. 美容師は / お金持ちの / お客を / 副店長が / 鏡で / 見ていた / シャンプー台に / 素早く / 連れて 行った。

17a. 教授は / 内気な / 留学生を / 気長に / コンピューターで / 探していた / 大学院生に / 親切に / 電子 メールを送った。

17d. 教授は / 内気な / 留学生を / 秘書が / コンピューターで / 探していた / アパートに / 親切に / 住まわ せた。

18a. 大学生たちは / せわしい / 小学生を / 気長に / 駅前で / 待っていた / 友達に / 慌てて / 道を聞いた。 18d. 大学生たちは / せわしい / 小学生を / 先生が / 駅前で / 待っていた / バスに / 慌てて / 乗らせた。 19a. 新聞記者は / 素人の / 役者を / 偶然に / テレビで / 見た / 監督に / 何回も / 質問をした。 19d. 新聞記者は / 素人の / 役者を / 友人が / テレビで / 見た / 劇場に / 何回も / 連れて行った。 20a. ガイドさんは / にぎやかな / 観光客を / 上手に / パノラマで / 写した / 青年達に / 約束通りに / お金 をあげた。

Case markers and clause boundaries

47

20d. ガイドさんは / にぎやかな / 観光客を / 青年が / パノラマで / 写した / 富士山に / 約束通りに / 登ら せた。

21a. ボディガードは / アイドル / 歌手を / たまたま / 楽屋で / 見つけた / ファンの子に / すばやく / キス をした。

21d. ボディガードは / アイドル / 歌手を / 知合いが / 楽屋で / 見つけた / 衣装だんすに / すばやく / 隠 した。

22a. 課長は / しつこい / 社員を / できるだけ / いつも / 避けていた / 同僚に / 一緒に / 外周りをさせた。 22d. 課長は / しつこい / 社員を / 新入社員が / いつも / 避けていた / 出張に / 一緒に / 行かせた。 23a. おじいさんは / 可愛い / 小学生を / すんなりと / 自転車で / 追い越した / 先生に / 遠慮なく / お願い を頼んだ。

23d. おじいさんは / 可愛い / 小学生を / 青年たちが / 自転車で / 追い越した / バスに / 遠慮なく / 乗せて あげた。

24a. 老人は / 疲れていた / おばさんを / つい / 普通電車の中で / 押し倒した / おとこの人に / 少し / 怒っ た。

24d. 老人は / 疲れていた / おばさんを / 孫が / 普通電車の中で / 押し倒した / 段ボール箱に / 少し / 座ら せた。

25a. お父さんは / 眠たそうな / 子供を / 一生懸命に / 急行列車の中で / 支えていた / 老人に / 丁寧に / 話 しかけた。

25d. お父さんは / 眠たそうな / 子供を / お母さんが / 急行列車の中で / 支えていた / 荷物に / 丁寧に / 寄 りかからせた。

26a. 浪人は / 痩せっぽちの / 高校生を / さんざん / 運動場で / 馬鹿にしていた / コーチに / しつこく / 文 句をいった。

26d. 浪人は / 痩せっぽちの / 高校生を / コーチが / 運動場で / 馬鹿にしていた / ゲームに / しつこく / 誘った。

27a. 精神科医は / 狂暴な / 患者を / ひどく / リハビリテーションで / 怖がっていた / こどもに / とうとう / 声をかけた。 27d. 精神科医は / 狂暴な / 患者を / 同僚が / リハビリテーションで / 怖がっていた / 特別室に / とうとう / 入れた。 28a. 家主は / 貧しい / 学生達を / いじわるく / 長い間 / 放っておいた / 不動産屋に / 急に / 声をかけた。 28d. 家主は / 貧しい / 学生達を / 不動産屋が / 長い間 / 放っておいた / アパートに / 急に / 引っ越しさ せた。

29a. 運転手は / 無邪気な / 子供を / やさしく / 公園で / なでていた / 子守に / そっと / 道を聞いた。 29d. 運転手は / 無邪気な / 子供を / 高校生が / 公園で / なでていた / りすに / そっと / 近づけさせた。 30a. 事務員は / 鋭い / インターンを / ようやく / 先月 / 探してきた / 秘書に / 即刻 / 連絡をした。 30d. 事務員は / 鋭い / インターンを / 副部長が / 先月 / 探してきた / 仕事に / 即刻 / つかせた。

Case markers and clause boundaries

Table 1: Regions for the self-paced reading presentation in Experiment 1 Sentence Type ARC DA SDA

1 2 3 4 5 6 7 8 Adv N1 -nom N2 -dat N3 -acc V N-acc Adv Vmain Adv N1 -nom N2 -acc N3 -acc V N-dat Adv Vmain Adv N2 -acc N1 -nom N3 -acc V N-dat Adv Vmain

48

Case markers and clause boundaries

49

Table 2: Regions for the self-paced reading presentation in Experiment 2 Sentence Type SR Control for SR SOR Canonical control for SOR Scrambled control for SOR

1 N1 -nom N1 -nom N1 -nom N1 -nom Adj

2 Adj Adj Adj Adj N2 -acc

3 N2 -acc N2 -nom N2 -acc N2 -acc N1 -nom

4 Adv1 Adv1 Adv1 N3 -nom N3 -nom

5 Adv Adv Adv Adv Adv

6 V V V V V

7 N-dat N-dat N-dat N-dat N-dat

8 Adv Adv Adv Adv Adv

9 Vmain Vmain Vmain Vmain Vmain

Case markers and clause boundaries

50

Table 3: Regions for the self-paced reading presentation in Experiment 3 Sentence Type SR Control for SR SOR Canonical control for SOR Scrambled control for SOR

1 N1 -top N1 -top N1 -top N1 -top Adj

2 Adj Adj Adj Adj N2 -acc

3 N2 -acc N2 -nom N2 -acc N2 -acc N1 -top

4 Adv1 Adv1 Adv1 N3 -nom N3 -nom

5 Adv Adv Adv Adv Adv

6 V V V V V

7 N-dat N-dat N-dat N-dat N-dat

8 Adv Adv Adv Adv Adv

9 Vmain Vmain Vmain Vmain Vmain

Case markers and clause boundaries

51

Figure Captions

Figure 1: Experiment 1: residual reading times and standard errors for each region.

Figure 2: Experiment 2: residual reading times and standard errors for each region of the SR and its control.

Figure 3: Experiment 2: residual reading times and standard errors for each region of the SOR and its two controls.

Figure 4: Experiment 3: residual reading times and standard errors for each region of the SR and its control.

Figure 5: Experiment 3: residual reading times and standard errors for each region of the SOR and its two controls.

Case markers and clause boundaries Figure 1:

1500

1250

ARC SDA DA

Residual Reading Times (msec)

1000

750

500

250

0

-250

-500

-750 0

1

2

3

4

5

6

7

8

Regions

Figure 1: Experiment 1: residual reading times and standard errors for each region.

52

Case markers and clause boundaries

53

Figure 2:

1500

Residual Reading Times (msec)

1250

SR Control

1000

750

500

250

0

-250

-500 0

1

2

3

4

5

6

7

8

9

Regions

Figure 2: Experiment 2: residual reading times and standard errors for each region of the SR and its control.

Case markers and clause boundaries

54

Figure 3:

1500

SOR Canonical Control Scrambled Control

Residual Reading Times (msec)

1250

1000

750

500

250

0

-250

-500 0

1

2

3

4

5

6

7

8

9

Regions

Figure 3: Experiment 2: residual reading times and standard errors for each region of the SOR and its two controls.

Case markers and clause boundaries

55

Figure 4:

1500

Residual Reading Times (msec)

1250

SR Control

1000

750

500

250

0

-250

-500 0

1

2

3

4

5

6

7

8

9

Regions

Figure 4: Experiment 3: residual reading times and standard errors for each region of the SR and its control.

Case markers and clause boundaries

56

Figure 5:

1500

Residual Reading Times (msec)

1250

SOR Canonical Control Scrambled control

1000

750

500

250

0

-250

-500 0

1

2

3

4

5

6

7

8

9

Regions

Figure 5: Experiment 3: residual reading times and standard errors for each region of the SOR and its two controls.

Case markers and clause boundaries

57

Contents 1 Head-driven parsing

2

Processing clauses in Japanese

4

Arguments against head-driven models . . . . . . . . . . . . . . . . . . . . . . . .

5

Underspecified heads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

Case markers in Japanese processing . . . . . . . . . . . . . . . . . . . . . . . . .

7

Clause boundaries and the double o constraint Experiment 1

8 10

Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

Materials

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11

Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12

Data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13

Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

Comprehension question response accuracy . . . . . . . . . . . . . . . . . . .

14

Reading times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15

Local assignment of clause boundaries

16

Displacing one or two NPs during reanalysis . . . . . . . . . . . . . . . . . . . . .

17

Unambiguous control sentences for SRs and SORs . . . . . . . . . . . . . . . . . .

18

Two control sentences for SORs . . . . . . . . . . . . . . . . . . . . . . . . .

18

A control sentence for SRs . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20

Experiment 2

22

Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

Case markers and clause boundaries Materials

58

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

Procedure and data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

Comprehension question accuracy . . . . . . . . . . . . . . . . . . . . . . . .

24

Reading times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

Experiment 3

27

Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

28

Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

28

Materials

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

28

Procedure and data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . .

29

Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29

Comprehension question accuracy . . . . . . . . . . . . . . . . . . . . . . . .

29

Reading times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

30

Animacy of the relative clause head . . . . . . . . . . . . . . . . . . . . . . .

31

General discussion Ga NPs as clause boundary markers . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion

32 33 34

Appendix A.

43

Appendix B.

45