What’s the Date? High Accuracy Interpretation of Weekday Names

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 553–560 Manchester, August 2008 What’s the Date? High Accuracy Interpretation of Weekday Names Paweł Mazur1,2 1Institute of Applied Informatics, Wrocław University of Technology Wyb. Wyspia´nskiego 27, 50-370 Wrocław, Poland Pawel.Mazur@pwr.wroc.pl Robert Dale2 2Centre for Language Technology, Macquarie University, NSW 2109, Sydney, Australia {mpawel,rdale}@ics.mq.edu.au Abstract In this paper we present a study on the interpretation of weekday names in texts. Our algorithm for assigning a date to a weekday name achieves 95.91% accuracy on a test data set based on the ACE 2005 Training Corpus, outperforming pre- viously reported techniques run against this same data. We also provide the first detailed comparison of various approaches to the problem using this test data set, em- ploying re-implementations of key tech- niques from the literature and a range of additional heuristic-based approaches. 1 Introduction Many temporal expressions in text are underspeci- fied, requiring contextually-sourced information in order to determine their correct interpretation. In some cases, it is sufficient to determine what is sometimes called the temporal focus, so that the precise location of a relative temporal expression on a timeline can be determined with respect to this ‘time of speaking’. Consider, for example, ex- pressions like the following: (1) three days ago (2) last Monday (3) in two weeks time Once we know the temporal focus, calculation of the temporal location referred to in each of these cases is straightforward, since the temporal ex- pressions themselves explicitly indicate what we c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. will call the direction of offset (here, respectively, past, past and future). However, in other cases there is no explicit indication of the direction of offset from the temporal focus. This is most ob- viously the case when bare expressions based on calendar cycles—i.e., weekday names and month names—are used, as in the following example: (4) Jones met with Defense Minister Paulo Portas on Tuesday and will meet Foreign Minis- ter Antonio Martins da Cruz before leav- ing Portugal Wednesday. Here, the proper interpretation of the references to Tuesday and Wednesday requires at the least a correct syntactic analysis of the sentence, in order to locate the controlling verb for each weekday name. The tense of this verb can then be used to determine the direction—either in the past or in the future—in which we need to look to establish the fully specified date referred to. In the case of example (4), this means determining that Tuesday is in the scope of the verb met, and that Wednes- day is in the scope of the verb group will meet. As we note below, it turns out that there are cases where even the controlling verb does not provide sufficient information to determine the direction of offset. But even in those cases where the tense of the verb does provide the relevant information, there are two problems. First, especially when the sentences considered are complex, …

Content truncated to 500 words