Why are Chinese characters so damn hard?

Abstract

This thesis looks into the problem of learning Chinese characters for foreign language learners and focuses on learning approaches that stress recognising characters without writing them by hand, which are becoming popular due to the widespread use of computer-based input methods. Mistaking characters for each other has been identified as an important problem that learners need to overcome. The empirical investigations in the thesis include a self-observation diary study and connectionist simulations of learning Chinese characters. The diary study collected over 1,500 pairs of characters that had been mistaken for one another in the process of learning. The analysis of these cases revealed an interplay of various factors that led to character confusion: graphical, semantic and phonetic similarity, as well as association caused by frequent co-occurrence of given characters in some words. A more detailed analysis distinguished character components that have a semantic or phonetic value in modern Chinese. It showed how the presence of similar components may contribute to character confusion, and found more complex cases of relationships between the value of the components of the target character and the actual pronunciation and meaning of the character it was confused with. The connectionist simulation of character acquisition presented in this thesis is based on the DISLEX model, which consists of two self-organising maps and aims to provide a neurobiologically plausible account of word learning. An evaluation of the first version of the model showed that the pairs of confused characters collected in the diary study were represented significantly closer to each other than the average. Nevertheless, the model had major flaws, which were addressed in the second version. It included a more sophisticated representation of the semantic, phonetic and graphemic features of the characters. The second model showed a significant improvement over the first one. The model accounted for character confusion by representing the approximate pronunciation of the characters, the approximate pronunciation indicated by their phonetic components, frequently recurring graphical components and the semantic classification of the characters (as indicated by the hypernyms). These results give an indication of what a psychologically plausible representation of Chinese characters may look like. Experiments with more learners are required to assess the scope of applicability of these findings and the predictive value of the model.