Combining Relational and Distributional Knowledge for Word Sense Disambiguation
Richard Johansson, Luis Nieto Piña · DSpace repository (University of Tartu) · 2015
We present a new approach to word sense disambiguation derived from recent ideas in distributional semantics. The input to the algorithm is a large unlabeled cor-pus and a graph describing how senses are related; no sense-annotated corpus is needed. The fundamental idea is to em-bed meaning representations of senses in the same continuous-valued vector space as the representations of words. In this way, the knowledge encoded in the lex-ical resource is combined with the infor-mation derived by the distributional meth-ods. Once this step has been carried out, the sense representations can be plugged back into e.g. the skip-gram model, which allows us to compute scores for the differ-ent possible senses of a word in a given context. We evaluated the new word sense dis-ambiguation system on two Swedish test sets annotated with senses defined by the SALDO lexical resource. In both evalu-ations, our system soundly outperformed random and first-sense baselines. Its ac-curacy was slightly above that of a well-known graph-based system, while being computationally much more efficient. 1