Applying the semantics of negation to SMT through n-best list re-ranking

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 598–606, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Applying the semantics of negation to SMT through n-best list re-ranking Federico Fancellu Centre for Global Intelligent Content School of Computer Science and Statistics Trinity College Dublin ffancellu@cngl.ie Bonnie Webber School of Informatics University of Edinburgh Edinburgh, UK, EH8 9AB bonnie@inf.ed.ac.uk Abstract Although the performance of SMT sys- tems has improved over a range of differ- ent linguistic phenomena, negation has not yet received adequate treatment. Previous works have considered the prob- lem of translating negative data as one of data sparsity (Wetzel and Bond (2012)) or of structural differences between source and target language with respect to the placement of negation (Collins et al. (2005)). This work starts instead from the questions of what is meant by negation and what makes a good translation of negation. These questions have led us to explore the use of semantics of negation in SMT — specifically, identifying core semantic el- ements of negation (cue, event and scope) in a source-side dependency parse and re- ranking hypotheses on the n-best list pro- duced after decoding according to the ex- tent to which an hypothesis realises these elements. The method shows considerable improve- ment over the baseline as measured by BLEU scores and Stanford’s entailment- based MT evaluation metric (Padó et al. (2009)). 1 Introduction Translating negation is a task that involves more than the correct rendering of a negation marker in the target sentence. For instance, translating Italy did not defeat France in 1909 differs from trans- lating Italy defeated France in 1909, or France did not defeat Italy in 1909, or Italy did not con- quer France in 1909. These examples show that translating negation also involves placing in the right position the semantic arguments as well as the event directly negated. Moreover, if the source sentence was uttered in response to the statement I think Italy defeated France in 1911, where the fo- cus is the temporal argument in 1911, one can see that the system should not lose track of the focus of negation when producing the hypothesis trans- lation. Although negation must be appropriately ren- dered to ensure correct representation of the se- mantics of the source sentence in the machine out- put, only some of the efforts to improve the transla- tion of negation-bearing sentences in SMT address the problem. Wetzel and Bond (2012) considered negation as a problem of data sparsity and so attempted to en- rich the training data with negative paraphrases of positive sentences. Collins et al. (2005) and Li et al. (2009) both addressed differences in the placement of negation in source and target texts, by re-ordering negative elements in the source sen- tence to better resemble their position in the corre- sponding target text. Although these approaches show improvement over the baseline, neither con- siders negation as a linguistic phenomenon with specific characteristics. This we do …

Content truncated to 500 words