ConSenses: Disambiguating content word groups based on knowledge base and definition embedding

We introduce a method for disambiguating a given group of semantically related words with respect to a certain sense inventory, such as WordNet or Cambridge English Dictionary. In our approach, every member word is converted into a set of senses to be disambiguated. The method involves clustering relevant senses and filtering out irrelevant senses, and determining the intended senses for each word in the group based on pairwise sense similarity. A preliminary evaluation of our method on several datasets shows that the method extends and outperforms the previous work that only deals with noun groups [1]. Our method is more generally applicable, allowing nouns, verbs, and adjectives groups, and can be used to aligning two anthologies to combine knowledge resources, as well as to generate training data for word sense disambiguation tasks.