Thematic role assignment refers to the on-line processing of assigning semantic roles, such as assigning agents or patients, to arguments (i.e., nouns) related to the verb. Linguistic information provided by arguments (e.g., word order, or case marking), as well as lexical argument representation of the verb is used for assigning thematic roles. The Extended Argument Dependency Model (eADM) suggests that argument cues utilized to assign semantic roles vary across languages. For rigid word-ordering languages (e.g., English) with case marking, readers adopt a position-based assignment, according to which the initial argument is usually analyzed as an agent. By contrast, thematic role assignment in unrestricted word-ordering languages (e.g., German, Italian, Japanese, and Turkish) with case marking exhibits a morphology-based strategy. The eADM model also predicts a reversal of thematic role assignment when the verb's argument representation contradicts with the argument cues, which is based on verb information and induces additional processing costs. Considerable evidence has demonstrated the language-specific weight on argument cues. However, it is unknown whether word order strongly affects thematic role assignment in Chinese (a rigid-ordering language with case marking) reading. In addition, the reanalysis of thematic roles proposed by the eADM model has only been tentatively explored in Spanish. Whether such reanalysis processing exists in other languages, especially in non-alphabetic languages like Chinese, is still lack of evidence. The present study examined the reliance on word order information in the existence of case marking information and the reanalysis of thematic roles when argument representation of the verb was in contradiction with cues of arguments in Chinese. The sentence structure (centered or preposed) and the type of control verb (subject-control or object-control) were manipulated. Sentences in the centered structure provided information of word order and case marking, while sentences in the preposed structure only provided case marking information. Argument representation of object-control verbs incompatible with the information of arguments would lead to a re-assignment of semantic roles. The argument representation of subject-control verbs compatible with the argument cues would cause no reanalysis. Fifty-four pairs of control verbs were selected, each of which was embedded into a centered-structure sentence and a preposed-structure sentence. Eye movements of 24 native Chinese speakers were recorded by the Eyelink. eye tracker. Each participant read 54 experimental sentences, followed by a comprehension question. The results showed that the preposed structure sentences caused longer second-pass reading time and more total incoming regressions in the first noun, longer first-pass reading time in the second noun, and longer regression path duration in the verb region than the centered-structure sentences, which suggested the strategy of position-based assignment for Chinese readers. There were robust main effects of the types of control verb, in that longer first pass reading time, regression path duration, and total incoming regressions were observed in the verb region, and longer second-pass reading time and total incoming (outgoing) regressions were found in the post-verb region in the object control verb condition than in the subject control verb condition. These results indicated that the mismatch of verb argument representation and argument cues contributed to an extra processing load. In addition, interactions between sentence structure and types of control verb were also observed, with longer second-pass reading time and total incoming regression in the second noun and longer second-pass reading time in the verb region in the centered-structure sentences containing object control verbs than those containing subject control verbs. There were longer first-pass reading time and regression path duration in the post-verb region in the preposed-structure sentences in the object-control verb condition than in the subject-control verb condition. In conclusion, these findings indicate that Chinese readers depend on word order information heavily to assign thematic roles even when there is case marking; also, the mismatch between cues of arguments and the argument representation of control verbs in Chinese reading causes reanalysis of thematic roles. Such findings are in line with the claim of the eADM model.
