Limited syntactic parallelism in Chinese ambiguity resolution

Abstract

Using the stop-making-sense paradigm (Boland, Tanenhaus, Garnsey, & Carlsen, 1995) and eye-tracking during reading, we examined the processing of the Chinese Verb NP ₁ de NP ₂ construction, which is temporarily ambiguous between a complement clause (CC) analysis and a relative clause (RC) analysis. Resolving the ambiguity as the more complex, less preferred CC was costly under some conditions but not under others. We took this as evidence for a limited parallel processor, such as Tabor and Hutchins’ (2004) SOPARSE, that maintains multiple syntactic analyses across several words of a sentence when the structures are each supported by the available constraints.

Acknowledgements

This research was partly supported by the National Natural Science Foundation of China (#30300112), the Natural Science Foundation of Beijing (#7062035), and a pilot grant jointly sponsored by the University of Michigan and Peking University.

The authors thank San Duanmu Richard Lewis, Samuel Epstein, Janet Fodor, and two anonymous reviewers for helpful comments and discussions on this manuscript. We are also grateful to Grover Yu for his assistance in conducting the norming surveys and to Xiaoming Jiang for his help in collecting the eye-movement data.

Notes

¹These items were selected from the Corpus for Studies of Modern Chinese (Beijing Language and Culture University, 1995), which has 1.24 million words collected from a broad range of genres.

²It is not entirely impossible for the first four words in (3b) and (4b) to have a relative clause continuation, such as [fen3shua1 lao3jiu4 de fang2jian1 de gong1ren2…] [paint old ATT room RC worker] ‘The worker that painted the old rooms …’. However, such a sentence with two nearly adjacent des should be rare. In fact, the first de is usually dropped in an expression like this to avoid redundancy, as the de following an adjective is omissible and in fact omitted about 90% of the time based on our corpus analysis.

³There were ten items per condition for each participant, so at the beginning of the sentence the maximum number of trials on which a participant could respond ‘No’ was ten. If a participant made four ‘No’ judgements at the first opportunity, the participant's score at this word was 40%. At the next word, the number of remaining possible ‘No’ responses would be six. Thus, if two ‘No’ responses were made at this word position, the score was 33%. If all of the trials in a given condition for a particular participant or item received ‘No's before the end of the sentence, the later word positions were assigned a value of 100% rather than 0%.

⁴Lewis’ SNIP operator appears to be sensitive only to syntactic inconsistencies, so the most straightforward assumption is that reanalysis would be attempted at the conjunction in both the Animate and Inanimate Ambiguous conditions. In fact, the repair surpasses the locality constraints of SNIP, regardless of whether the repair is initiated at NP₂ or the conjunction.