2018/04/27 by Dmitry Ustalov, Денис Тесленко, Ustalov, Dmitry +9
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Speech and dialogue systems #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1804.10686
openalex publication_date 2018/04/27 · openalex created_date 2022/10/02 · openalex updated_date 2026/07/28
In this paper, we present Watasense, an unsupervised system for word sense\ndisambiguation. Given a sentence, the system chooses the most relevant sense of\neach input word with respect to the semantic similarity between the given\nsentence and the synset constituting the sense of the target word. Watasense\nhas two modes of operation. The sparse mode uses the traditional vector space\nmodel to estimate the most similar word sense corresponding to its context. The\ndense mode, instead, uses synset embeddings to cope with the sparsity problem.\nWe describe the architecture of the present system and also conduct its\nevaluation on three different lexical semantic resources for Russian. We found\nthat the dense mode substantially outperforms the sparse one on all datasets\naccording to the adjusted Rand index.\n