A maximum entropy approach to HowNet-based Chinese word sense disambiguation
Wong Ping Wai, Yang Yongsheng · 2002
This paper presents a maximum entropy method for the disambiguation of word senses as defined in HowNet. With the release of this bilingual (Chinese and English) knowledge base in 1999, a corpus of 30,000 words was sense tagged and released in January 2002. Concepts meanings in HowNet are constructed by a closed set of sememes, the smallest meaning units, which can be treated as semantic tags. The maximum entropy model treats semantic tags like parts-of-speech tags and achieves an overall accuracy of 89.39%, outperforming a baseline system, which picks the most frequent sense.