Identifying Functional Relations in Web Text

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 1266–1276, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Identifying Functional Relations in Web Text Thomas Lin, Mausam, Oren Etzioni Turing Center University of Washington Seattle, WA 98195, USA {tlin,mausam,etzioni}@cs.washington.edu Abstract Determining whether a textual phrase denotes a functional relation (i.e., a relation that maps each domain element to a unique range el- ement) is useful for numerous NLP tasks such as synonym resolution and contradic- tion detection. Previous work on this prob- lem has relied on either counting methods or lexico-syntactic patterns. However, determin- ing whether a relation is functional, by ana- lyzing mentions of the relation in a corpus, is challenging due to ambiguity, synonymy, anaphora, and other linguistic phenomena. We present the LEIBNIZ system that over- comes these challenges by exploiting the syn- ergy between the Web corpus and freely- available knowledge resources such as Free- base. It first computes multiple typed function- ality scores, representing functionality of the relation phrase when its arguments are con- strained to specific types. It then aggregates these scores to predict the global functionality for the phrase. LEIBNIZ outperforms previ- ous work, increasing area under the precision- recall curve from 0.61 to 0.88. We utilize LEIBNIZ to generate the first public reposi- tory of automatically-identified functional re- lations. 1 Introduction The paradigm of Open Information Extraction (IE) (Banko et al., 2007; Banko and Etzioni, 2008) has scaled extraction technology to the massive set of relations expressed in Web text. However, additional work is needed to better understand these relations, and to place them in richer semantic structures. A step in that direction is identifying the properties of these relations, e.g., symmetry, transitivity and our focus in this paper – functionality. We refer to this problem as functionality identification. A binary relation is functional if, for a given arg1, there is exactly one unique value for arg2. Exam- ples of functional relations are father, death date, birth city, etc. We define a relation phrase to be functional if all semantic relations commonly ex- pressed by that phrase are functional. For exam- ple, we say that the phrase ‘was born in’ denotes a functional relation, because the different seman- tic relations expressed by the phrase (e.g., birth city, birth year, etc.) are all functional. Knowing that a relation is functional is helpful for numerous NLP inference tasks. Previous work has used functionality for the tasks of contradiction detection (Ritter et al., 2008), quantifier scope dis- ambiguation (Srinivasan and Yates, 2009), and syn- onym resolution (Yates and Etzioni, 2009). It could also aid in other tasks such as ontology generation and information extraction. For example, consider two sentences from a contradiction detection task: (1) “George Washington was born in Virginia.” and (2) “George Washington was born in Texas.” As Ritter et al. (2008) points out, we can only de- termine that the two sentences are contradictory if we know that the semantic relation referred to by the phrase …

Content truncated to 500 words