Transforming Examples into Patterns for Information Extraction

Transforming Examples into Patterns for Information Extraction Roman Yangarber and Ralph Grishman Dept. of Computer Science New York University 715 Broadway, 7th Floor New York, NY 10003, USA roman, grishman~cs.nyu.edu Abstract Information Extraction (IE) systems today are com- monly based on pattern matching. The patterns are regular expressions stored in a customizable knowl- edge base. Adapting an IE system to a new subject domain entails the construction of a new pattern base - - a time-consuming and expensive task. We describe a strategy for building patterns from examples. To adapt the IE system to a new domain quickly, the user chooses a set of examples in a training text, and for each example gives the logical form entries which the example induces. The system transforms these examples into patterns and then applies meta-rules to generalize these patterns. 1 Introduction The task of Information Extraction (IE) as under- stood in this paper is the selective extraction of mean- ing from free natural language text. 1 This kind of text analysis is distinguished from others in Natu- ral Language Processing in that "meaning" is under- stood in a narrow sense - in terms of a fixed set of se- mantic objects, namely, entities, relationships among these entities, and events in which these entities par- ticipate. These objects belong to a small number of types, all having fixed regular structure, within a fixed and closely circumscribed subject domain, which per- mits the objects to be stored in a database (e.g., a relational data base). These characteristics make the IE task both simpler and more tractable than the more ambitious problem of general text understand- ing. They allow us to define the notion of a "correct answer", and, therefore, to conduct quantitative eval- uation of performance of an IE system. A series of for- mal evaluations -- the Message Understanding Con- 1For a review of a range of extraction systems, the reader is referred to [9]. ferences (MUCs), 2 conducted over the last decade -- are described in [8, 6]. The MUCs have yielded some widely (if not uni- versally) accepted wisdom regarding IE: • Customization and portability is an important problem: to be considered a useful tool, an IE system must be able to perform in a variety of domains. • Systems have modular design: Control is encap- sulated in immutable core engines, which draw upon domain- or scenario-specific information stored in knowledge bases (KB) which are cus- tomized for each new domain and scenario. • Text analysis is based on pattern matching: regu- lar expression pattern matching is a widely used strategy in the IE community. Pattern match- ing is a form of deterministic bottom-up par- tial parsing. This approach has gained consid- erable popularity due to limitations on the accu- racy of full syntactic parsers, and the adequacy of partial, semantically-constrained, parsing for this task [2, 1, 5]. • Building effective patterns for a new domain is the most complex and time-consuming part of the customization process; it is highly error- …

Content truncated to 500 words