Automatic Interpretation of the English Possessive

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 372–381, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Automatic Interpretation of the English Possessive Stephen Tratz † Army Research Laboratory Adelphi Laboratory Center 2800 Powder Mill Road Adelphi, MD 20783 stephen.c.tratz.civ@mail.mil Eduard Hovy † Carnegie Mellon University Language Technologies Institute 5000 Forbes Avenue Pittsburgh, PA 15213 hovy@cmu.edu Abstract The English ’s possessive construction oc- curs frequently in text and can encode several different semantic relations; how- ever, it has received limited attention from the computational linguistics community. This paper describes the creation of a se- mantic relation inventory covering the use of ’s, an inter-annotator agreement study to calculate how well humans can agree on the relations, a large collection of pos- sessives annotated according to the rela- tions, and an accurate automatic annota- tion system for labeling new examples. Our 21,938 example dataset is by far the largest annotated possessives dataset we are aware of, and both our automatic clas- sification system, which achieves 87.4% accuracy in our classification experiment, and our annotation data are publicly avail- able. 1 Introduction The English ’s possessive construction occurs fre- quently in text—approximately 1.8 times for every 100 hundred words in the Penn Treebank1(Marcus et al., 1993)—and can encode a number of different semantic relations including ownership (John’s car), part-of-whole (John’s arm), extent (6 hours’ drive), and location (America’s rivers). Ac- curate automatic possessive interpretation could aid many natural language processing (NLP) ap- plications, especially those that build semantic representations for text understanding, text gener- ation, question answering, or information extrac- tion. These interpretations could be valuable for machine translation to or from languages that al- low different semantic relations to be encoded by †The authors were affiliated with the USC Information Sciences Institute at the time this work was performed. the possessive/genitive. This paper presents an inventory of 17 semantic relations expressed by the English ’s-construction, a large dataset annotated according to the this in- ventory, and an accurate automatic classification system. The final inter-annotator agreement study achieved a strong level of agreement, 0.78 Fleiss’ Kappa (Fleiss, 1971) and the dataset is easily the largest manually annotated dataset of possessive constructions created to date. We show that our automatic classication system is highly accurate, achieving 87.4% accuracy on a held-out test set. 2 Background Although the linguistics field has devoted signif- icant effort to the English possessive (§6.1), the computational linguistics community has given it limited attention. By far the most notable excep- tion to this is the line of work by Moldovan and Badulescu (Moldovan and Badulescu, 2005; Bad- ulescu and Moldovan, 2009), who define a tax- onomy of relations, annotate data, calculate inter- annotator agreement, and perform automatic clas- sification experiments. Badulescu and Moldovan (2009) investigate both ’s-constructions and of constructions in the same context using a list of 36 semantic relations (including OTHER). They take their examples from a collection of 20,000 ran- domly selected sentences from Los Angeles Times news articles …

Content truncated to 500 words