Bootstrapping Information Extraction from Field Books

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 827–836, Prague, June 2007. c⃝2007 Association for Computational Linguistics Bootstrapping Information Extraction from Field Books Sander Canisius and Caroline Sporleder ILK / Communication and Information Sciences Tilburg University, P.O. Box 90153, 5000 LE Tilburg, The Netherlands {S.V.M.Canisius,C.Sporleder}@uvt.nl Abstract We present two machine learning ap- proaches to information extraction from semi-structured documents that can be used if no annotated training data are available, but there does exist a database filled with information derived from the type of docu- ments to be processed. One approach em- ploys standard supervised learning for infor- mation extraction by artificially constructing labelled training data from the contents of the database. The second approach com- bines unsupervised Hidden Markov mod- elling with language models. Empirical evaluation of both systems suggests that it is possible to bootstrap a field segmenter from a database alone. The combination of Hid- den Markov and language modelling was found to perform best at this task. 1 Introduction Over the past decades much textual data has be- come available in electronic form. Many text types are inherently more or less structured, for example, classified advertisements for appartments, medical records, or logs of archaeological finds or zoological specimens collected during expeditions. Such doc- uments consist of a number of shorter texts (or en- tries), each describing an individual object (e.g., an appartment, or an archaeological find) or event (e.g., a patient presenting to a health care provider). These descriptions in turn typically consist of different seg- ments (or fields) which contain information of a spe- cific type drawn from a more or less given inven- tory. Example (1), for instance, shows two descrip- tions of zoological specimens (a snake and three frogs) collected during an expedition. The descrip- tions contain different segments giving information about the specimens and the circumstances of their collection. For example, in the first description, Lep- tophis and ahaetulla refer, respectively, to the genus and species of the specimen, road to Overtoom men- tions the place of collection, in bush above water en- codes information about the biotope, in the process of eating Hyla minuta is a remark about the circum- stances of collection, 16-V-1968 gives the collection date and RMNH 15100 the registration number. (1) Leptophis ahaetulla, road to Overtoom, in bush above water in the process of eating Hyla minuta 16-V-1968. RMNH 15100 Hyla minuta 1 ♀ 2 ♂ Las Claritas, 9-VI-1978 quak- ing near water 50 cm above water surface, near sec- ondary vegetation, 200 m, M.S. Hoogmoed, RMNH 27217 27219 Unfortunately, this inherent structure is rarely made explicit. While the different object or event descriptions might be indicated by additional white- space or other formatting means, as in the example above, the individual fields within a description are typically not marked in any way. However, knowl- edge of the inherent structure would be very bene- ficial for information extraction and retrieval. For instance, texts in their raw …

Content truncated to 500 words