This is a brief summary of recent research on how adults understand sentences in Japanese. It includes discussions on incremental processing and on filler-gap dependencies. New reading time data are reported on the interaction between lexical frequency and scrambled word orders, indicating that the deleterious effects of low frequency verbs and non-canonical word order are not additive.
This paper investigates mechanisms of long-distance dependency formation in language comprehension, using experimental data on the processing of Japanese interrogatives and exclamatives to explore the nature of locality biases in parsing. Findings on the processing of exclamative wh-phrases are compared to previous results involving the processing of interrogative wh-phrases, revealing both similarities and differences. Experiment 1 uses a sentence fragment completion task with in-situ and fronted exclamative and interrogative wh-phrases. Both types of in-situ wh-phrase show a strong bias for local generation of licensing particles. Conditions with fronted wh-phrases show a contrast between interrogative and exclamative wh-phrases: interrogatives show a bias for interpretation in an embedded clause, replicating previous evidence for a long-distance scrambling bias in Japanese (Aoshima, Phillips, & Weinberg, 2004); in contrast, the long-distance scrambling bias is weaker for fronted exclamative wh-phrases. Experiment 2 uses an on-line self-paced reading task to investigate the processing consequences of expectations for a local licensor for in-situ exclamative wh-phrases. Results indicate processing disruption when readers fail to encounter a licensor for an exclamative wh-phrase at the first possible verb position, although the disruption is weaker than the Typing Mismatch Effect shown for interrogatives in previous studies by Miyamoto and Takahashi (2002). Different possible accounts of the parallels and contrasts between processing of interrogatives and exclamatives are discussed.
The validity of the schema integration model (Fujiki & Chujo, 2005a) was investigated. This model assumes that the process of forming the semantic representation of a sentence consists of repeating the integration of the schema belonging to the syntactic head word with constituents' schemata in the phrase. The model also assumes that a comprehender checks the consistency between the constituents' schemata and the requirements for integration into the slots of the head schema. According to the model, if these requirements were not satisfied, they would be extended to be coherent with the comprehender's world knowledge. In this study, we controlled the compatibility between representations of noun phrases and schema of verbs in order to verify whether the requirements of verb schemata were checked in the process of integration with the noun phrase. The results of Experiment 1 in which the sentences were presented using the moving window paradigm, supported the assumption of the model. In Experiment 2, each sentence was presented in a batch and eye movements were measured. The results indicated that the representation of the sentences were not completed in the first eye fixations on the words of the sentence, and extended the requirements of integration after the second fixation. These results indicate that the schema integration model is valid, and that the comprehender selected an efficient processing strategy to perform the multiple integrations of schemata during the online process of forming semantic representations in a sentence.
According to a widely held view, the object-subject-verb word order in Japanese is derived from the subject-object-verb word order by shifting the object to the sentence-initial position. This movement of the object, called scrambling, is hypothesized to leave “a trace” in the original object position (Saito, 1985). With regard to this view, during online sentence processing, a fronted object must be associated with its trace (filler-driven parsing). If a human actually processes scrambled sentences by filler-driven parsing, it is assumed that an object is reactivated and the processing load increases at the trace position. Although many psycholinguistic studies have been conducted in order to investigate the processing of a trace at the trace position, few studies have focused on processing around the trace positions. In the present study, by using a cross-modal lexical priming (CMLP) task that is capable of measuring the processing load and the activation level of an object at arbitrary positions, we investigated the processing around the trace positions in Japanese clause-internal scrambled sentences. In this study, in order to correct the problem encountered in the preceding study (Nakano et al., 2002) using the CMLP task, we did not measure the direct priming effect; however, we measured the indirect priming effect as a method of investigating the activation level of an object. When the data of all the participants were analyzed together, the increases in the processing load and the reactivation of an object around the trace position were not revealed. However, because of the difficulty of the CMLP task, the previous study (Nakano et al., 2002) presented the reactivation of an object at the trace position for participants who responded to lexical decisions quickly and possessed a high working memory capacity. Therefore, the participants in this study were divided into fast and slow groups based on their lexical decision latencies during the task. The results that reflect the filler-driven parsing were revealed only for the fast group. In the fast group, the processing load at the trace position was found to exceed the load at the position preceding and following the trace position. Further analyses of the results showed that the activation level of an object increased only at the trace position.
Repeated-name and pronoun have been investigated for their referring functions. Reviewing the literature led to the hypothesis that these two types of anaphor might differ in their units of activation. Experiment 1 examined recognition latencies for referents and related words referred by repeated-name or pronoun. Participants recognized referents in both repeated-name and pronoun conditions faster than control condition which had no reference. However, for related words, facilitation was observed in only pronoun condition. This indicated that pronoun and repeated-name activated different units of linguistic representation. Experiment 2 replicated Experiment 1 except that an additional sentence was placed between antecedent and anaphoric sentences. This change affected the patterns of results. Pronoun had no effect on recognition for referents and related words. On the other hand, repeated-name facilitated recognition for their referents, but inhibited for the related words. These patterns were interpreted on the basis of differential functions in which repeated-name helps the integration of text with thematic shift, whereas pronoun contributes to immediate integration in theme-continuous situation.
The purpose of this study was to demonstrate that people's preference of the word order of Japanese sentences correlates with meanings that cannot be reduced to meanings of head verbs. For this purpose, we conducted two experiments to compare two groups of sentences with different “supra-lexical” meanings (of “caused motion” and “caused possession”), with respect to the different configurations of case-marking particles, or case-markers, -ga, -ni, and -wo. In the experiments, participants were presented phrases (NPs and a verb) which made a sentence in a random order. After a short delay, the participants were required to recall and speak out the learned phrases in a natural sentence format. In Experiment 1, 20 caused possession and 20 caused movement sentences were prepared for the experimental materials. All the sentence including nominals marked by -ga, -ni, and -wo. The nouns and verbs in the total of 40 sentences were all different. In Experiment 2, 16 pairs of sentences which had a same verb but had different constructional meanings were prepared for the caused motion and possession sentences. The results of the two experiments showed that the participants recalled the phrases in the order of “N-gaN-woN-niV” for the caused motion sentences more often than for the caused possession sentences in both of two experiments. These results suggest that, while there is an overall tendency for Japanese speakers to prefer “N-gaN-niN-woV” order to “N-gaN-woN-niV” order, the strength of the preference is not constant among different supra-lexical meanings.
The self-paced moving-window reading paradigm was used in two experiments to examine the possibility of retaining ambiguity in comprehending Japanese temporarily ambiguous sentences, consisting of “subject-NP / VP (object-NP and transitive verb) / head-NP of relative-clause”. The biases for the main-clause or the relative-clause interpretations were manipulated by defining the strength of co-occurrence between the subject-NPs or head-NPs and the VPs, using the rating of typicality as an agent of the VP. In Experiment 1, the proper namess in Japanese were used for the weak biases. The rating values of co-occurrence for them were significantly lower than that for the strong biases, although there was neither semantic anomaly nor implausibility. In reading the ambiguous strings having weak biases for main-clause, it was predicted that readers retained both of the possible interpretations because of no trigger about whether the main-clause interpretation or the relative-clause interpretation was assumed, and consequently that the garden path effects at the head-NPs were reduced. The experiment did not, however, show any reductive effect, but the reading time data suggested that the use of the proper names could increased readers' processing load. In Experiment 2, the proper names were replaced to the common nouns which also had weak biases. The results showed that the clear garden path effects were obtained only in the condition with strong main-clause and weak relative-clause biases, while the effects were eliminated in the other conditions with weak main-clause biases or strong relative-clause biases. These results suggested that the interaction between lexical connectivity and processing load decides whether the structural ambiguity should be resolved or retained.
Japanese double-gap relative clause constructions contain two gaps and two modified nouns. The two gap-noun dependencies can be nested or crossed. The results of a questionnaire and a self-paced reading experiment are reported indicating that nested dependencies are easier to understand than crossed ones in this construction. First, the result is the opposite of what has been reported for nested dependencies in German compared to crossed dependencies in Dutch (Bach, Brown, & Marslen-Wilson, 1986). We argue that this contrast is likely to be related to the comprehension of single-gap relative clauses in Japanese, which have also been claimed to present unexpected comprehension patterns compared to their counterparts in various other languages. Second, the results support previous observations that crossed as well as nested dependencies are grammatical in double-gap relative clauses (contra Fodor, 1978). Third, we discuss how the word-by-word processing of double-gap relative clauses can be explained based on the recursive application of the algorithm commonly assumed for the processing of single-gap relative clauses in Japanese.
The present study investigated the canonical positions of four kinds of adverbial expressions (or adjuncts) in Japanese (i.e. modal, time, manner, and resultative adverbs) using a sentence plausibility judgment task measuring reaction times and error rates. Sentences with a modal adverb were processed faster in the Adverb-Subject-Object-Verb (ASOV) order than either the SAOV or SOAV order. For time adverbs, the mean reaction time for SOAV was longer than the mean reaction times for ASOV and SAOV, which did not differ significantly from each other. For manner and resultative adverbs, ASOV took longer to process than SAOV and SOAV, and the latter two orders did not show a reliable difference in reaction times. These results suggest that the canonical word order(s) is(are) ASOV for modal adverbs, ASOV and SAOV for time adverbs, and SAOV and SOAV for manner and resultative adverbs, as predicted by the following syntactic structure of Japanese. [MP (Modal-Adv.) [IP (Time-Adv.) Subject (Time-Adv.) [VP (Manner/Resultative-Adv.) Object (Manner/Resultative-Adv.) Verb] Infl] Modal]
The present study investigates the effects of case marking information on processing an object NP in Japanese. In Japanese, there are two kinds of case markers that are assigned to an object NP, one is a dative case marker “ni”, the other is an accusative case marker “o”, and it is specified for each verb which case marker is assigned to an object NP. We conducted two experiments to verify whether the parser is sensitive to the difference in case markers in object NPs. Results from the experiments show two findings: (i) the parser processes object NPs in different ways depending on the case marker, (ii) based on the difference in the case markers of object NPs, the parser predicts what kind of element is input next. As for the mechanism of the parser, the present results are compatible with the model in which NPs in Japanese are associated with a parsing tree incrementally before verb information becomes available, and give a detailed explanation of how it processes the NPs with local information at the stage of preverbal processing.
The purpose of the present study was to examine pseudo-homophone effect in a sentence verification task in which sentences included a two-kanji compound nonword in the case that the target was identified by an underline. In a sentence verification task, participants asked to judge whether the presented sentence was acceptable or not. Previous studies showed that participants responded more quickly when sentences included a pseudo-homophone than when they included a non-homophonic nonword. However, participants responded more slowly to pseudo-homophones than to non-homophonic nonwords when a context of the sentence (a sentence without a target) was presented first, and a target followed which required judgment. The current experiment showed that participants responded more quickly when sentences included a pseudo-homophone, even if an underline was added at the target. Thus, the pseudo-homophone facilitation effect was observed even when the participant did not need to search where a target was. The result suggested that the main factor of the pseudo-homophone facilitation effect would be simultaneous processing of the target stimuli and sentence meaning.
In spoken sentence comprehension, prosody is regarded as an important information source for syntax. However, prosody can also be affected by various non-syntactic factors. The present study demonstrates that lexical accent, which itself is not relevant to syntax, is relevant to, and can have a considerable impact on the resolution of syntactic ambiguity in human sentence comprehension. In a (LB) left-vs. (RB) right-branching ambiguity in Japanese as in the example below, the presence or absence of lexical accent (indicated by “’”) on the initial adjective, (e.g., ao’i (blue): accented vs. kiiroi (yellow): unaccented) is known to influence phrasal level prosody. This in turn affected listeners' analyses on branching structure, as will be shown in the production study.
ao’i⁄kiiroi siidi’i-no ke’esu-o sa’sitekudasai blue⁄yellow CD-Gen case-Acc point “Point at the case of the blue⁄yellow CD.” (LB-interpretation) “Point at the blue⁄yellow case of the CD.” (RB-interpretation)
In the production task, naive speakers (n=7) gave verbal instructions to their partners to identify the target object, using a fixed phrase as indicated in the example above, where either RB or LB was intended. The F0 contour on the collected utterances showed little distinctive difference among conditions except for the LB-intended utterances with an accented adjective. The collected utterances were then fed to a perception test (forced choice task identifying the intended meaning) with a separate group of subjects (N=48). Listeners' performance in correctly identifying LB-intended utterances was significantly better for items with an accented adjective (ao’i) than with an unaccented adjective (kiiroi), whereas performance on RB-intended utterances were relatively poor regardless of the presence of an accent on the initial adjective. The results of the perception study can be explained by what was found in the production study: the successive downstep (which is conditioned to occur following an accented item) as a major cue indicating a left-branching structure was not available for items with an unaccented adjective. It is also suggested that when the relevant F0 cue is not available or useful, hearers rely more on durational cues.
A numeral quantifier in Japanese consists of a numeral and a classifier that agrees with the type of entity being counted. In this study we investigated, using event-related potentials, brain activities associated with the integration of a numeral classifier and a noun that denotes an entity⁄entities being counted. Based on previous studies, we considered two hypotheses. If semantic processes are crucially involved in the integration of a noun and a numeral classifier, like selectional restriction between a verb and its object, N400 would be found when subjects read incorrect pairs of a noun and a numeral classifier. On the other hand, if the integration of a noun and a numeral classifier is morpho-syntactic in nature, parallel to gender agreement in European languages, then LAN would be elicited. Results of our experiment showed that mismatch of a noun and a numeral classifier evoked N400. This suggests that a numeral classifier in Japanese semantically selects a noun denoting a certain type of entity.
This study considers the principle that governs syntactic parser's revision processes by investigating how the parser resolves temporary ambiguities which emerge during reanalysis. Previous studies which addressed this issue have argued that the parser prefers preserving as many existing structure and interpretation as possible. This study, however, argues that the hypothesis emphasizing preserving the existing structure needs to be revised and presents the Error Signal-based Revision Principle (ESRP). The ESRP requires the parser to constructs a maximal projection which has a minimal number of nodes dominating the error signal and can be attached into the existing structure in a legitimate way such that the rules of grammar are not violated. We demonstrate that the ESRP can correctly explain the parser's preference in resolving temporary ambiguities during reanalysis. Furthermore, we present empirical evidence from an experiment employing the event-related brain potentials (ERPs) technique which indicate that the ESRP is more plausible than the structure-preserving principle as an ambiguity-resolution principle. Finally, we argue that the ESRP is consistent with the assumption that the parser performs its task in the “least effort” way.