Character-Level Dependencies in Chinese: Usefulness and Learning

We investigate the possibility of exploiting character-based dependency for Chinese information processing. As Chinese text is made up of character sequences rather than word sequences, word in Chinese is not so natural a concept as in English, nor is word easy to be defined without argument for such a language. Therefore we propose a character-level dependency scheme to represent primary linguistic relationships within a Chinese sentence. The usefulness of character dependencies are verified through two specialized dependency parsing tasks. The first is to handle trivial character dependencies that are equally transformed from traditional word boundaries. The second furthermore considers the case that annotated internal character dependencies inside a word are involved. Both of these results from character-level dependency parsing are positive. This study provides an alternative way to formularize basic characterand word-level representation for Chinese.