Managing Multiword Expressions in a Lexicon-Based Sentiment Analysis System for Spanish

This paper describes our approach to managing multiword expressions in Sentitext, a linguistically-motivated, lexicon-based Sentiment Analysis (SA) system for Spanish whose performance is largely determined by its coverage of MWEs. We defend the view that multiword constructions play a fundamental role in lexical Sentiment Analysis, in at least three ways. First, a significant proportion conveys semantic orientation; second, being units of meaning, their relative weight to the calculated overall sentiment rating of texts needs to be accounted for as such, rather than the number of component lexical units; and, third, many MWEs contain individual words that carry a given polarity, which may or may not be that of the phrase as a whole. As a result, successful lexiconbased SA calls for appropriate management of MWEs.