Unsupervised word sense disambiguation in dynamic semantic spaces

In this paper, we are mainly concerned with the ability to quickly and automa cally dis nguish word senses in dynamic seman c spaces in which new terms and new senses appear frequently. Such spaces are built “on the fly” from constantly evolving data sets such as Wikipedia, repositories of patent grants and applica ons, or large sets of legal documents for Technology Assisted Review and e-discovery. This immediacy rules out supervision as well as the use of a priori training sets. We show that the various senses of a term can be automa cally made apparent with a simple clustering algorithm, each sense being a vector in the seman c space. While we only consider here seman c spaces build by using random vectors, this algorithm should work with any kind of embedding, provided meaningful similari es between terms can be computed and do fulfill at least the two basic condi ons that terms which close meanings have high similari es and terms with unrelated meanings have near-zero similari es.