Conclusion and Perspectives for Future Research

Abstract

In this book we summarized the more relevant results of five years of research on Semantic Domains in Computational inguistics. As a linguistic work, we concentrated on studying the basic properties of Semantic Domains, a concept that we derived from the notion of Semantic Field, widely known in structural linguistics.We have shown that Semantic Domains are characterized by lexical coherence, a fundamental property that we corroborated by means of a corpusbased analysis. Lexical coherence is a key concept to understand the basics of the domain-driven approach to Natural Language Processing proposed here: it allows us to automatically induce Domain Models from corpora and to define the Domain-Driven Disambiguation methodology.

In addition, we remarked on the connections between lexical ambiguity and Semantic Domains, identifying a fundamental aspect of sense discrimination that we have called domain polysemy. Domain polysemy emerges when the same word frequently occurs in texts belonging to different domains. We also observed that Semantic Domains are shallow models for lexical variability: words belonging to the same domain denote very closely related concepts. For this reason Semantic Domains provide a precious knowledge source to estimate topic similarity among texts. Another very important property of Semantic Domains is their multilinguality. We have shown that the lexicon of different languages is structured into a common domain set that can be learned from comparable corpora in a totally unsupervised way.