Enriching a Lexicon of Discourse Connectives with Corpus-based Data

Enriching a Lexicon of Discourse Connectives with Corpus-based Data Anna Feltracco1,2,3, Elisabetta Jezek2, Bernardo Magnini1 1Fondazione Bruno Kessler, Via Sommarive 18, 38100 Trento, Italy 2University of Pavia, Strada Nuova 65, 27100 Pavia, Italy 3University of Bergamo, Via Salvecchio, 19, 24129 Bergamo, Italy feltracco@fbk.eu, jezek@unipv.it, magnini@fbk.eu Abstract We present the results of the effort of enriching the pre-existing resource LICO, a Lexicon of Italian COnnectives retrieved from lexicographic sources (Feltracco et al., 2016), with real corpus data for connectives marking contrast relations in text. The motivation beyond our effort is that connectives can only be interpreted when they appear in context, that is, in a relation between the two fragments of text that constitute the two arguments of the relation. In this perspective, adding corpus examples annotated with connectives and arguments for the relation allows us to both extend the resource and validate the lexicon. In order to retrieve good corpus examples, we take advantage of the existing Contrast-Ita Bank (Feltracco et al., 2017), a corpus of news annotated with explicit and implicit discourse contrast relations for Italian according to the annotation scheme proposed in the Penn Discourse Tree Bank (PDTB) guidelines (Prasad et al., 2007). We also use an extended -non contrast annotated- version of the same corpus and documents from Wikipedia. The resulting resource represents a valuable tool for both linguistic analyses of discourse relations and the training of a classifier for NLP applications. Keywords: discourse connectives, contrast relation, corpus examples 1. Introduction Discourse relations and the linguistic elements marking them in text, commonly referred to as discourse connec- tives, have recently been at the core of several annota- tion efforts for multiple languages (including English, Ger- man, French, Italian, Portuguese, see Stede and Umbach (1998), Roze et al. (2012) among others). In this paper, we present the results of the effort of enriching the pre- existing resource LICO, a lexicon of Italian connectives re- trieved from lexicographic sources (Feltracco et al., 2016), with real corpus data, thus allowing us to both extend the resource and validate the lexicon. Our goal in this contribu- tion is limited to the class of connectives marking contrast, and the additional relations such connectives might convey, some of them being polysemous. The motivation beyond our effort is that connectives can only be interpreted and disambiguated when they appear in context, that is, in a re- lation between the two fragments of text that constitute the two arguments of the relation. In order to retrieve good ex- amples, we take advantage of the existing Contrast-Ita Bank (Feltracco et al., 2017), a corpus of news annotated with ex- plicit and implicit discourse contrast relations for Italian ac- cording to the annotation scheme proposed in the Penn Dis- course Tree Bank (PDTB) guidelines (Prasad et al., 2007). Contrast-Ita Bank contains corpus annotations for 19 dis- course connectives of contrasts, 14 of them are included in LICO; these provide the starting point for our work. We pick additional examples from the larger news corpus from which Contrast-Ita …

Content truncated to 500 words