Parallel Entity and Treebank Annotation

Proceedings of the Workshop on Frontiers in Corpus Annotation II: Pie in the Sky, pages 21–28, Ann Arbor, June 2005. c⃝2005 Association for Computational Linguistics Parallel Entity and Treebank Annotation Ann Bies Linguistic Data Consortium 3600 Market Street, 810 Philadelphia, PA 19104 bies@ldc.upenn.edu Seth Kulick Institute for Research in Cognitive Science 3401 Walnut Street Suite 400A Philadelphia, PA 19104 skulick@linc.cis.upenn.edu Mark Mandel Linguistic Data Consortium 3600 Market Street, 810 Philadelphia, PA 19104 mamandel@ldc.upenn.edu Abstract We describe a parallel annotation ap- proach for PubMed abstracts. It includes both entity/relation annotation and a tree- bank containing syntactic structure, with a goal of mapping entities to constituents in the treebank. Crucial to this approach is a modification of the Penn Treebank guide- lines and the characterization of entities as relation components, which allows the in- tegration of the entity annotation with the syntactic structure while retaining the ca- pacity to annotate and extract more com- plex events. 1 Introduction A great deal of annotation effort for many different corpora has been devoted to annotation for entities and syntactic structure (treebanks). However, pre- vious efforts at treebanking have largely been inde- pendent of the constituency of entities, and previous efforts at entity annotation have likewise been inde- pendent of corresponding layers of syntactic struc- ture. We describe here a corpus being developed for biomedical information extraction with levels of both entity annotation and treebank annotation, with a goal that entities can be mapped to constituents in the treebank. We are collaborating with researchers in the Di- vision of Oncology at The Children’s Hospital of Philadelphia, for the purpose of automatically min- ing the corpus of cancer literature for those as- sociations that link specified variations in individ- ual genes with known malignancies. In particular, we are interested in extracting three entities (Gene, Variation event, and Malignancy) in the following relationship: Gene X with genomic Variation event Y is correlated with Malignancy Z. For example, WT1 is deleted in Wilms Tumor #5. In addition, Vari- ation events are themselves relations, consisting of entities representing different aspects of a Variation event. Mapping entities to treebank constituents is a de- sirable goal since the entities can then be viewed as semantic types associated with syntactic con- stituents, and we expect that automated analyses of these related levels will interact in a mutually rein- forcing and beneficial way for development of sta- tistical taggers. In this paper we describe aspects of the entity and treebank annotation that allow this mapping to be largely successful. Potentially large enti- ties that would otherwise cut across syntactic con- stituents are decomposed into components of a re- lation. While this is worthwhile by itself on con- ceptual grounds for entity definition, and was in fact not done for reasons of mapping to syntactic con- stituents, it makes such a mapping easier. The tree- bank annotation has been modified from the Penn Treebank guidelines in various ways, such as greater structure for prenominal modifiers. Again, while this would have been done regardless of …

Content truncated to 500 words