Factuality Annotation and Learning in Spanish Texts

Factuality Annotation and Learning in Spanish Texts Dina Wonsever, Aiala Rosá, Marisa Malcuori Instituto de Computación, Facultad de Ingeniería Universidad de la República Julio Herrera y Reissig 565 Montevideo, Uruguay wonsever@fing.edu.uy, aialar@fing.edu.uy, marisamalcuori@gmail.com Abstract We present a proposal for the annotation of factuality of event mentions in Spanish texts and a free available annotated corpus. Our factuality model aims to capture a pragmatic notion of factuality, trying to reflect a casual reader judgements about the realis / irrealis status of mentioned events. Also, some learning experiments (SVM and CRF) have been held, showing encouraging results. Keywords: Event Factuality, Annotation Scheme, Corpus Annotation, Machine Learning 1. Introduction In automated content extraction from text data, words are a primary source of information. But the presence of a word that denotes an event, for instance a noun such as war or a verb such as eat, does not imply the actual occurrence of a war or an eating act. Factuality is the status of an event mention with regards to its effective occurrence, as perceived by a reader. In this document we describe a model for representing event factuality and a Spanish corpus annotated according to this model. As previously said, the determination of events mention (or event, for short) factuality is necessary for different purposes, like Information Extraction, Information Retrieval, Question Answering or others. Our model of factuality is included in an event annotation scheme (Wonsever et al, 2012), under the form of an attribute with six possible values. The representation and automatic detection of event factuality has been addressed by different authors, the model proposed by Roser Saurí (2008) being the main reference in the area. In Saurí's model factuality is determined as the combination of two elements: modality (CT: certain, PR: probable, PS: possible, U: underspecified) and polarity (+: positive, -: negative, u: underspecified). The combination of these two dimensions results in the following values: CT+, CT-, CTu, PR+, PR-, PS+, PS- and Uu. Another element that Saurí includes in factuality determination is the specification of the source that presents the event. The sources are represented as nested sources, following the model for opinions and emotions from (Wiebe et al, 2005). The scheme from Saurí was applied for the automatic determination of event factuality through an algorithm that uses some lexical ressources like modality and negation markers and source introducing predicates. It was also used for annotating the FactBank corpus (Saurí & Pustejovsky, 2009, 2012). Based on the model defined by Saurí, several proposals have emerged. Van Son et al (2014) incorporated some modifications to the English scheme proposed by Saurí, including a new dimension for temporality to distinguish future from non future events. This new version of the scheme was used to annotate the corpus of the NewsReader Project (Tonelli et al, 2014). Narita et al (2013) and Matsuyoshi et al (2010) worked on the adaptation of the scheme for Japanese. They have annotated a corpus and have worked on the automatic detection of factuality. Glavas et al (2012) …

Content truncated to 500 words