Distributional Semantics (DS) is a rich family of computational models that build semantic representations of lexical items from their statistical distribution in linguistic contexts. DS is currently experiencing an unprecedented fortune with a growing attention not only in computational linguistics, but also in cognitive science and theoretical linguistics. This is proved by the wide range of DS models that have appeared (e.g., vector spaces, Bayesian models, neural networks, etc.), but even more by the increased number of semantic tasks that these models have been applied to.
DS was born to address a specific issue, that is measuring the semantic similarity of lexical items to be used for thesaurus construction or synonym identification. The Distributional Hypothesis, the main theoretical foundation of DS, is in fact a statement about lexical semantic similarity, which is defined in terms of similarity of linguistic contexts. However, human semantic competence well exceeds the ability to judge lexical similarity. Polysemy, compositionality, inference, semantic creativity are only some of the main phenomena that must be part of the agenda of any full-fledged semantic theory. DS aims at becoming a general model for semantic representation and processing, and therefore it must be evaluated with respect to its ability to explain semantic facts like these. What is the current ability of DS to address these issues? To what extent semantic properties can be modeled in terms of distributional semantic similarity, or alternatively, can DS go beyond the mere notion of semantic similarity? What lies beyond its possibilities? Recently, DS has begun to address issues such as compositionality, polysemy, and semantic relations, but lots of questions remain open. The purpose of this talk is to explore the current boundaries of DS and the chances to enlarge them, in particular by finding new synergies with other types of semantic models.
Alessandro Lenci, PhD, is Associate Professor in Computational Linguistics at the University of Pisa, and the director of the Computational Linguistics Laboratory (http//colinglab.humnet/unipi.it) at the Dept. of Philology, Literature, and Linguistics. He received a PhD in linguistics in 1999 from the Scuola Normale Superiore in Pisa. He has extensively published on lexical semantics, computational lexicons and natural language processing. Dr. Lenci's main research areas are: distributional semantics (with particular interest to its application in linguistic and cognitive research), computational lexical semantics, computational models of verb argument structure, event types and verb aspect, tools and resources for Natural Language Processing.
Computational Linguistics Laboratory
Dept. of Philology, Literature, and Linguistics
University of Pisa (Italy)