Pseudohomophone effects in processing Chinese compound words

Abstract

The issue of how phonological information becomes available in reading Chinese and the role that it plays in lexical access was investigated for Chinese compound words, using pseudohomophone effects in lexical and phonological decision as a diagnostic tool. Pseudohomophones were created by replacing one or both constituents of two-character compound words with orthographically dissimilar homophonic characters. Experiment 1 found that mixed pseudohomophones sharing one constituent with their base words were more difficult to reject than control nonwords in lexical decision. Pure pseudohomophones sharing no constituents with their base words did not show this effect. Experiment 2 used mixed pseudohomophones and found an interaction between base word frequency and the frequency of constituent characters in determining pseudohomophone effects. Experiment 3 used a phonological decision task and found exceptionally poor performance for pure pseudohomophones. These results are interpreted in an interactive framework where the direct mapping from orthography to semantics is dominant and phonology plays a subsidiary role.

Keywords:

• compound word
• phonological processing
• pseudohomophone effect
• reading Chinese

Acknowledgements

The research reported here was supported in part by grants from Natural Science Foundation of China (30070260, 30470569, 60435010) to Xiaolin Zhou and in part by a grant from UK Economic and Social Research Council (ESRC) to William Marslen-Wilson and Xiaolin Zhou. We thank Xueming Lu, Jie Zhuang, and Yan Liu for helping us test participants in Beijing and the reviewers for their constructive comments on an earlier version of the paper.

Notes

¹Throughout the paper, the pronunciations of Chinese characters are given in pinyin, the Chinese alphabetic system. Numbers in brackets represent the lexical tones of syllables.

²It may be possible to realise this framework in a distributed, connectionist model, where orthographic, phonological, and semantic representations are viewed as activation patterns distributed over large numbers of simple processing unit (Rueckl, Mikolinski, Raven, Miner, & Mars, 1997; Seidenberg, 1987, 1989; Seidenberg & McClelland, 1989; Zhou & Marslen-Wilson, 2000b). There would be no need for explicit representation of morphological structure at either semantic or form levels. Lexical representations for compound words are ‘whole-word’ activation patterns at these levels. The morphological structure of compound words emerges from the dynamic interaction between semantic and form representations. At the orthographic and phonological levels, morphological structure may be cued by changes in transitional probability at morpheme boundaries. At the semantic level, semantic representations of both whole words and constituent morphemes are activated in parallel from their form representations, resulting in competition between the two sets of semantic features not shared between constituent morphemes and whole words. Patterns corresponding to constituent morphemes are well established through the acquisition and use of the monomorphemic words. In visual word recognition, although the mapping between orthography, phonology, and semantics establishes stable lexical activation at these levels for the whole words, the statistical regularities encoded within these levels and, more importantly, the interaction between different levels also establish in parallel sub-patterns corresponding to constituent morphemes, which are necessarily of higher frequency than compound words themselves.