A corpus-based versus experimental examination of word- and character-frequency effects in Chinese reading: theoretical implications for models of reading

Chinese words consist of a variable number of characters that are normally written in continuous lines, without the blank spaces that are used to separate words in most alphabetic writing systems. These conventions raise questions about the relative roles of character versus whole-word processing in word identification, and how words are segmented from strings of characters for the purpose of their identification and saccade targeting. The present article attempts to address these questions by reporting an eye-movement experiment in which 60 participants read a corpus of sentences containing two-character target words that varied in terms of their overall frequency and the frequency of their initial characters. We examine participants' eye movements using both corpus-based statistical models and more standard analyses of our target words. In addition to documenting how key lexical variables influence eye movements and highlighting a few discrepancies between the results obtained using our two statistical approaches, our experiment shows that high-frequency initial characters can actually slow word identification. We discuss the theoretical significance of this finding and others for current models of Chinese reading, and then describe a new computational model of eye-movement control during the reading of Chinese. Finally, we report simulations showing that this model can account for our findings.