A Synchronous Context Free Grammar for Time Normalization

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 821–826, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics A synchronous context free grammar for time normalization Steven Bethard University of Alabama at Birmingham Birmingham, Alabama, USA bethard@cis.uab.edu Abstract We present an approach to time normalization (e.g. the day before yesterday⇒2013-04-12) based on a synchronous context free grammar. Synchronous rules map the source language to formally defined operators for manipulat- ing times (FINDENCLOSED, STARTATENDOF, etc.). Time expressions are then parsed using an extended CYK+ algorithm, and converted to a normalized form by applying the opera- tors recursively. For evaluation, a small set of synchronous rules for English time expres- sions were developed. Our model outperforms HeidelTime, the best time normalization sys- tem in TempEval 2013, on four different time normalization corpora. 1 Introduction Time normalization is the task of converting a natural language expression of time into a formal representa- tion of a time on a timeline. For example, the expres- sion the day before yesterday would be normalized to the formal representation 2013-04-12 (assuming that today is 2013-04-14) in the ISO-TimeML rep- resentation language (Pustejovsky et al., 2010). Time normalization is a crucial part of almost any infor- mation extraction task that needs to place entities or events along a timeline. And research into methods for time normalization has been growing since the ACE1 and TempEval (Verhagen et al., 2010; UzZa- man et al., 2013) challenges began to include time normalization as a shared task. 1http://www.itl.nist.gov/iad/mig/tests/ace/ Most prior work on time normalization has taken a rule-based, string-to-string translation approach. That is, each word in a time expression is looked up in a normalization lexicon, and then rules map this sequence of lexical entries directly to the normalized form. HeidelTime (Str¨otgen and Gertz, 2012), which had the highest performance in TempEval 2010 and 2013, and TIMEN (Llorens et al., 2012), which re- ported slightly higher performance in its own experi- ments, both follow this approach. A drawback of this approach though is that there is no nesting of rules: for example, in HeidelTime the rules for yesterday and the day before yesterday are completely separate, despite the compositional nature of the latter. A notable exception to the string-to-string ap- proach is the work of (Angeli et al., 2012). They de- fine a target grammar of typed pre-terminals, such as YESTERDAY (a SEQUENCE) or DAY (a DURATION), and compositional operations, such as SHIFTLEFT (a (RANGE, DURATION) → RANGE). They apply an expectation-maximization approach to learn how words align to elements of the target grammar, and achieve performance close to that of the rule-based systems. However, their grammar does not allow for non-binary or partially lexicalized rules (e.g. SE- QUENCE → DURATION before SEQUENCE would be impossible), and some of their primitive elements could naturally be expressed using other primitives (e.g. YESTERDAY as SHIFTLEFT(TODAY, 1 DAY)). We present a synchronous grammar for time nor- malization that addresses these shortcomings. We first define a grammar of formal …

Content truncated to 500 words