A corpus-based approach to the discovery of cross-strait lexical contrasts

Studies of cross-strait lexical contrasts in the use of Mandarin Chinese reveal that a divergence has become increasingly evident. This divergence is apparent in phonological, semantic, and pragmatic analyses and has become an obstacle to knowledge-sharing and information exchange. Given the wide range of divergences, it seems that Chinese character forms offer the most reliable regular mapping between cross-strait usage contrasts. We propose a new approach to discovery of cross-strait contrasts in this paper anchored on the regular correspondences of characters. Our approach is corpus-based and collocation-driven. We use known contrast pairs as seeds in a corpus containing data from both the PRC and Taiwan. Collocation patterns in terms of both lexical lists and grammatical functions of these contrast pairs are studied to semi-automatically discover additional contrast pairs. This approach obtains both NLP applicability and linguistic felicity since it yields both the contrast pairs as well as their usage contexts.