Parsing Time: Learning to Interpret Time Expressions

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 446–455, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Parsing Time: Learning to Interpret Time Expressions Gabor Angeli Stanford University Stanford, CA 94305 angeli@stanford.edu Christopher D. Manning Stanford University Stanford, CA 94305 manning@stanford.edu Daniel Jurafsky Stanford University Stanford, CA 94305 jurafsky@stanford.edu Abstract We present a probabilistic approach for learn- ing to interpret temporal phrases given only a corpus of utterances and the times they ref- erence. While most approaches to the task have used regular expressions and similar lin- ear pattern interpretation rules, the possibil- ity of phrasal embedding and modification in time expressions motivates our use of a com- positional grammar of time expressions. This grammar is used to construct a latent parse which evaluates to the time the phrase would represent, as a logical parse might evaluate to a concrete entity. In this way, we can employ a loosely supervised EM-style bootstrapping approach to learn these latent parses while capturing both syntactic uncertainty and prag- matic ambiguity in a probabilistic framework. We achieve an accuracy of 72% on an adapted TempEval-2 task – comparable to state of the art systems. 1 Introduction Temporal resolution is the task of mapping from a textual phrase describing a potentially complex time, date, or duration to a normalized (grounded) temporal representation. For example, possibly complex phrases such as the week before last are often more useful in their grounded form – e.g., January 1 - January 7. The dominant approach to this problem in previ- ous work has been to use rule-based methods, gen- erally a combination of regular-expression matching followed by hand-written interpretation functions. In general, it is appealing to learn the interpre- tation of temporal expressions, rather than hand- building systems. Moreover, complex hierarchical temporal expressions, such as the Tuesday before last or the third Wednesday of each month, and am- biguous expressions, such as last Friday, are diffi- cult to handle using deterministic rules and would benefit from a recursive and probabilistic phrase structure representation. Therefore, we attempt to learn a temporal interpretation system where tempo- ral phrases are parsed by a grammar, but this gram- mar and its semantic interpretation rules are latent, with only the input phrase and its grounded interpre- tation given to the learning system. Employing probabilistic techniques allows us to capture ambiguity in temporal phrases in two impor- tant respects. In part, it captures syntactic ambigu- ity – e.g., last Friday the 13th bracketing as either [last Friday] [the 13th], or last [Friday the 13th]. This also includes examples of lexical ambiguity – e.g., two meanings of last in last week of November versus last week. In addition, temporal expressions often carry a pragmatic ambiguity. For instance, a speaker may refer to either the next or previous Fri- day when he utters Friday on a Sunday. Similarly, next week can refer to either the coming week or the week thereafter. Probabilistic systems furthermore allow propaga- …

Content truncated to 500 words