UCM-2: a Rule-Based Approach to Infer the Scope of Negation via Dependency Parsing

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 288–293, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics UCM-2: a Rule-Based Approach to Infer the Scope of Negation via Dependency Parsing Miguel Ballesteros, Alberto D´ıaz, Virginia Francisco, Pablo Gerv´as, Jorge Carrillo de Albornoz and Laura Plaza Natural Interaction Based on Language Group Complutense University of Madrid Spain {miballes, albertodiaz, virginia}@fdi.ucm.es, pgervas@sip.ucm.es, {jcalbornoz, lplazam}@fdi.ucm.es Abstract UCM-2 infers the words that are affected by negations by browsing dependency syntactic structures. It first makes use of an algo- rithm that detects negation cues, like no, not or nothing, and the words affected by them by traversing Minipar dependency structures. Second, the scope of these negation cues is computed by using a post-processing rule- based approach that takes into account the in- formation provided by the first algorithm and simple linguistic clause boundaries. An initial version of the system was developed to handle the annotations of the Bioscope corpus. For the present version, we have changed, omitted or extended the rules and the lexicon of cues (allowing prefix and suffix negation cues, such as impossible or meaningless), to make it suit- able for the present task. 1 Introduction One of the challenges of the *SEM Shared Task (Morante and Blanco, 2012) is to infer and classify the scope and event associated to negations, given a training and a development corpus based on Co- nan Doyle stories (Morante and Daelemans, 2012). Negation, simple in concept, is a complex but essen- tial phenomenon in any language. It turns an affir- mative statement into a negative one, changing the meaning completely. We believe therefore that be- ing able to handle and classify negations we would be able to improve several text mining applications. Previous to this Shared Task, we can find several systems that handle the scope of negation in the state of the art. This is a complex problem, because it re- quires, first, to find and capture the negation cues, and second, based on either syntactic or semantic representations, to identify the words that are di- rectly (or indirectly) affected by these negation cues. One of the main works that started this trend in natu- ral language processing was published by Morante’s team (2008; 2009), in which they presented a ma- chine learning approach for the biomedical domain evaluating it on the Bioscope corpus. In 2010, a Workshop on Negation and Spec- ulation in Natural Language Processing (Morante and Sporleder, 2010) was held in Uppsala, Swe- den. Most of the approaches presented worked in the biomedical domain, which is the most studied in negation detection. The system presented in this paper is a modifica- tion of the one published in Ballesteros et al. (2012). This system was developed in order to replicate (as far as possible) the annotations given in the Bio- scope corpus (Vincze et al., 2008). Therefore, for the one presented in the task we needed to modify most of the rules to make it able to handle the more …

Content truncated to 500 words