Lexical Opposition and Discourse Contrast: A Data-driven Investigation

This thesis focuses on the study of both lexical opposition between verbs (e.g. to increase / to decrease) and contrast at the textual level (e.g. `The price of x increased by 5%, while the price of y decreased by 2.3%'). Intuitively, lexical opposition and discourse contrast have some overlap, as they involve the comparison of two elements that are similar in some aspects but different in others. We adopt a data-driven approach and investigate opposition and contrast, as well as their co-occurrence, in the Italian language, observing the two linguistic phenomena in context. We aim at collecting both quantitative and qualitative data on opposition and contrast relations, useful for both linguistic research and computational purposes. We also aim at enriching and creating lexical resources and annotated corpora that encode the two relations. In particular, towards the enrichment of a lexical resource with opposition, we define and test an annotation schema, and discuss methodologies for acquiring reliable data at the verb sense level, focusing on crowd-sourcing. We also provide evidence that different types of opposition are recognized by people. Moreover, adopting a corpus-based approach, we focus on discourse contrast relations both when they are marked by textual connectives and when they are implicitly conveyed. We provide an analysis of the contrast relation in Italian in terms of frequency, explicitness/implicitness of the relation, and type of contrast, as a result of the annotation of a corpus of news (Contrast-Ita Bank). In addition to this, we present an investigation on the connectives of contrast in Italian for the creation of an exhaustive list of these lexical elements, starting from a lexicon of connectives we have created (LICO). Finally, we analyze the connection between lexical opposition and discourse contrast. In particular, we investigate the contribution of opposition when in the context of a contrast relation, by observing their co-occurrence in a corpus. We also propose an exploratory research about how the co-occurrence of the two linguistic phenomena can be formally represented.