How to make words with vectors: Phrase generation in distributional semantics

Distributional methods for semantics approximate the meaning of linguistic expressions with vectors that summarize the contexts in which they occur in large samples of text. Recently these methods have been extended beyond the lexical domain, by composing word vector representations into vectors for phrases and sentences. However, the link between language and meaning is, obviously, bidirectional and in this talk I will address the opposite problem of going from vectors to words: Given a distributional vector representing some meaning, how can we generate the phrase that best expresses that meaning? I will describe our proposal for a simple data-driven approach to the estimation of generation functions and present experiments in a monolingual setting, on paraphrase generation, as well as cross-lingually, on a task of translating adjective-noun phrases. Finally, source vectors may come from a different modality and I will discuss recent results on using vector representations of images to generate corresponding phrasal descriptions.