UConcordia: CLaC Negation Focus Detection at *Sem 2012

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 294–300, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics UConcordia: CLaC Negation Focus Detection at *Sem 2012 Sabine Rosenberg and Sabine Bergler CLaC Lab, Concordia University 1455 de Maisonneuve Blvd West, Montr´eal, QC, Canada, H3W 2B3 sabin ro@cse.concordia.ca, bergler@cse.concordia.ca Abstract Simply detecting negation cues is not suffi- cient to determine the semantics of negation, scope and focus must be taken into account. While scope detection has recently seen re- peated attention, the linguistic notion of focus is only now being introduced into computa- tional work. The *Sem2012 Shared Task is pioneering this effort by introducing a suitable dataset and annotation guidelines. CLaC’s NegFocus system is a solid baseline approach to the task. 1 Introduction Negation has attracted the attention of the NLP com- munity and we have seen an increased advance in sophistication of processing tools. In order to assess factual information as asserted or not, it is important to distinguish the difference between (1) (a) Newt Gingrich Not Conceding Race After Losing Florida Primary (b) Newt Gingrich Conceding Race Af- ter Losing Florida Primary This distinction is important and cannot be properly inferred from the surrounding context, not conced- ing a race after losing is in fact contrary to expecta- tion in the original headline (1a), and the constructed (1b) is more likely in isolation. Negation has been addressed as a task in itself, rather than as a component of other tasks in recent shared tasks and workshops. Detection of negation cues and negation scope at CoNLL (Farkas et al., 2010), BioNLP (Kim et al., 2011) and the Negation and Speculation in NLP Workshop (Morante and Sporleder, 2010) laid the foundation for the *Sem 2012 Shared Task. While the scope detection has been extended to fictional text in this task, an impor- tant progression from the newspaper and biomedi- cal genres, the newly defined Focus Detection for Negation introduces the important question: what is the intended opposition in (1a)? The negation trig- ger is not, the scope of the negation is the entire verb phrase, but which aspect of the verb phrase is underscored as being at variance with reality, that is, which of the following possible (for the sake of linguistic argument only) continuations is the more likely one: (2) i ..., Santorum does. (¬Newt Gingrich) ii ..., Doubling Efforts (¬concede) iii ..., Demanding Recount (¬race) iv ..., Texas redistricting at fault (¬Florida) This notion of focus of negation is thus a prag- matic one, chosen by the author and encoded with various means. Usually, context is necessary to de- termine focus. Often, different possible interpreta- tions of focus do not change the factual meaning of the overall text, but rather its coherence. In (1 a) the imagined possible contexts (2 ii) and (2 iii) closely correspond to a simple negation of (1 b), (2 i) and (2 iv) do not feel properly represented by simply negating (1 b). This level of interpretation is con- tentious …

Content truncated to 500 words