Word sense disambiguation based on memory enhancement mechanism

Word sense disambiguation (WSD) is a very critical yet challenging task in natural language processing (NLP), which aims at identifying the most suitable meaning of ambiguous words in the given contexts according to a predefined sense inventory. Existing WSD methods usually focus on learning the semantic interactions between a special ambiguous word and the glosses of its candidate senses and thus ignore complicated relations between the neighboring ambiguous words and their glosses, leading to insufficient learning of the interactions between words in context. As a result, they are difficult to leverage the knowledge from the other ambiguous words which might provide some explicit clues to identify the meaning of current ambiguous word. To mitigate this challenge, this paper proposes a novel neural model based on memory enhancement mechanism for WSD task, which stores the gloss knowledge of previously identified words into a memory, and further utilizes it to assist the disambiguation of the next target word. Extensive experiments, which are conducted on a unified evaluation framework of the WSD task, demonstrate that our model achieves better disambiguation performance than the state-of-the-art approaches (Code: https://github.com/baoshuo/WSD ).