Capsule Network Improved Multi-Head Attention for Word Sense Disambiguation

Word sense disambiguation (WSD) is one of the core problems in natural language processing (NLP), which to map an ambiguous word its correct meaning a specific context. There has been lively interest incorporating definition (gloss) into neural networks recent studies, makes great contribution improving performance WSD. However, disambiguating polysemes rare senses still hard. In this paper, while taking gloss consideration, we further improve WSD system from perspective semantic representation. We encode context and glosses target polysemy independently using encoders with same structure. To obtain better presentation each encoder, leverage capsule network capture different important information contained multi-head attention. finally choose representation closest as sense. do experiments on English all-words task. Experimental results show that our method achieves good performance, especially having inspiring effect words senses.