A Preliminary Work on Classifying Time Granularities of Temporal Questions

R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 414 – 425, 2005. © Springer-Verlag Berlin Heidelberg 2005 A Preliminary Work on Classifying Time Granularities of Temporal Questions Wei Li1, Wenjie Li1, Qin Lu1, and Kam-Fai Wong2 1 Department of Computing, The Hong Kong Polytechnic University, Hung Hom, Hong Kong {cswli, cswjli, csluqin}@comp.polyu.edu.hk 2 Department of Systems Engineering, the Chinese University of Hong Kong, Shatin, Hong Kong kfwong@se.cuhk.edu.hk Abstract. Temporal question classification assigns time granularities to tempo- ral questions ac-cording to their anticipated answers. It is very important for an- swer extraction and verification in the literature of temporal question answer- ing. Other than simply distinguishing between "date" and "period", a more fine- grained classification hierarchy scaling down from "millions of years" to "sec- ond" is proposed in this paper. Based on it, a SNoW-based classifier, combining user preference, word N-grams, granularity of time expressions, special patterns as well as event types, is built to choose appropriate time granularities for the ambiguous temporal questions, such as When- and How long-like questions. Evaluation on 194 such questions achieves 83.5% accuracy, almost close to manually tagging accuracy 86.2%. Experiments reveal that user preferences make significant contributions to time granularity classification. 1 Introduction Temporal questions, such as the questions with the interrogatives “when”, “how long” and “which year”, seek for the occurrence time of the events or the temporal attributes of the entities. Temporal question classification plays an important role in the litera- ture of question answering and temporal information processing. In the evaluation of TREC 10 Question-Answering (QA) track [1], more than 10% of questions in the test question corpus are temporal questions. Different from TREC QA track, Workshop TERQAS (http://www.timeml.org/terqas/) particularly investigated on temporal ques- tion answering instead of a general one. It focused on temporal and event recognition in question answering systems and paid great attention to temporal relations among states, events and time expressions in temporal questions. TimeML (http://www.ti- meml.org), a temporal information (e.g. time expression, tense & aspect) annotation standard, has also been used for temporal question answering in this workshop [2]. Correct understanding of a temporal question will greatly help extracting and verify- ing its answers and certainly improve the performance of any question answering system. Look at the following examples. [Ea]. What is the birthday of Abraham Lincoln? [Eb]. When did the Neanderthal man live? A Preliminary Work on Classifying Time Granularities of Temporal Questions 415 In a general question answering system, the question classifier commonly classifies temporal questions into two classes, i.e. “date” and “period”. With such a system, the above two questions are both assigned a “date”. Whereas it is natural for the question [Ea] to be answered with a particular data (e.g. “12/02/1809”), it is not the case for question [Eb], because a proper answer could be “35,000 years ago”. However, if it is known that the time granularity concerned is “thousands of years”, answer extraction turn to be more targeted. The need for a more fine-grained classification is obvious. Although there …

Content truncated to 500 words