A Semi-Supervised method for Persian homograph Disambiguation

摘要

One of the major challenges in the most natural languages processing (NLP) tasks such as machine translation, text to speech and text mining is Word Sense Disambiguation (WSD). Supervised methods are the most common solutions for WSD. However, they need large tagged corpuses which are not available in some languages such as Persian. The Semi-Supervised methods can solve this problem by using small tagged corpus and large untagged corpus. This paper presents a coarse-grained work in WSD that uses tri-training as the semi-supervised method and decision list as supervised classifier for training. The proposed method was evaluated on a corpus. The results show that the proposed method is more precise than the conventional Decision list when the tagged corpus is small.