Séminaire de Recherche en Linguistique

 

 

The Meaning of Word Embeddings: A Semantic Investigation

 

 

Word embeddings are vectorial semantic representations aimed at capturing shades of meaning from word co-occurrences. Since their introduction, these representations have been criticised for lacking interpretable dimensions. This property of word embeddings limits our understanding of the semantic features they actually encode. Moreover, it contributes to the ``black box'' nature of the tasks in which they are used, since the reasons of word embeddings performance often remain opaque to humans. In this talk, I present the results of a research aiming at exploring the semantic properties encoded in word embeddings by mapping them onto interpretable vectors, consisting of the set of explicit and neurobiologically-motivated features introduced by Binder et al (2016). This analysis addresses different types of embeddings, including the most recent contextualized representations (i.e., ELMo and BERT).