Identifying Temporal Orientation of Word Senses

Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning (CoNLL), pages 22–30, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Identifying Temporality of Word Senses Based on Minimum Cuts M. Hasanuzzaman,1,3 G. Dias,1,2 S. Ferrari,1,2 Y. Mathet,1,2 and A. Way3 1Université de Caen Normandie, Caen, France 2GREYC UMR 6072, Caen, France 3ADAPT Centre, School of Computing, Dublin City University, Dublin, Ireland Abstract The ability to capture time information is essential to many natural language pro- cessing and information retrieval applica- tions. Therefore, a lexical resource asso- ciating word senses to their temporal ori- entation might be crucial for the computa- tional tasks aiming at the interpretation of language of time in texts. In this paper, we propose a semi-supervised minimum cuts strategy that makes use of WordNet glosses and semantic relations to supple- ment WordNet entries with temporal infor- mation. Intrinsic and extrinsic evaluations show that our approach outperforms prior semi-supervised non-graph classifiers. 1 Introduction Recognizing temporal information can signifi- cantly improve the functionality of information re- trieval (Campos et al., 2014) and natural language processing (Mani et al., 2005) applications. Most text applications have been relying on rule-based time taggers such as HeidelTime (Ströt- gen and Gertz, 2015) or SUTime (Chang and Man- ning, 2012) to identify and normalize time men- tions in texts. Although interesting levels of per- formance have been seen (UzZaman et al., 2013), their coverage is limited to the finite number of rules they implement. Let’s take the following sentence: “Apple’s iPhone is currently one of the most popular smartphone”. When labeled by SU- Time1 or HeidelTime2, the adverb currently is cor- rectly tagged with the PRESENT_REF value. How- ever, if we change the sentence to “Apple’s iPhone 1http://nlp.stanford.edu:8080/sutime/ process 2http://heideltime.ifi.uni-heidelberg. de/heideltime/ is one of the most popular smartphones at the present day”, no temporal mention is found, al- though one may expect that within this context currently and present day share some equivalent temporal dimension. Such systems would cer- tainly benefit from the existence of a temporal re- source enumerating a large set of possible time variants (Kuzey et al., 2016). In parallel, new trends have emerged in the con- text of human temporal orientation (Schwartz et al., 2015). The underlying idea is to understand how past, present, and future emphasis in text may affect people’s finances, health, and happi- ness. For that purpose, temporal classifiers are built to detect the overall temporal dimension of a given sentence. For instance, the following Face- book post “can’t wait to get a pint tonight” would be tagged as FUTURE. Successful features include timexes, specific temporal (past, present, future) words from a commercial dictionary, but also n- grams, thus indicating that temporality may be embodied by multi-word terms, whose temporal orientation is unknown. As a consequence, discovering the temporal ori- entation of words is a challenging issue that may benefit many text applications. Whereas most prior studies have focused on temporal expres- sions and events, there has been a lack of work looking …

Content truncated to 500 words