Paper: Towards Dynamic Word Sense Discrimination with Random Indexing

ACL ID W13-3210
Title Towards Dynamic Word Sense Discrimination with Random Indexing
Venue Continuous Vector Space Models and their Compositionality
Session
Year 2013
Authors

Most distributional models of word sim- ilarity represent a word type by a single vector of contextual features, even though, words commonly have more than one sense. The multiple senses can be captured by employing several vectors per word in a multi-prototype distributional model, pro- totypes that can be obtained by first con- structing all the context vectors for the word and then clustering similar vectors to create sense vectors. Storing and clus- tering context vectors can be expensive though. As an alternative, we introduce Multi-Sense Random Indexing, which per- forms on-the-fly (incremental) clustering. To evaluate the method, a number of mea- sures for word similarity are proposed, both contextual and non-contextual, in- cluding new measures based on optimal alignment of word se...