Mapping Verbs in Different Languages to Knowledge Base Relations using Web Text as Interlingua

Proceedings of NAACL-HLT 2016, pages 818–827, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Mapping Verbs In Different Languages to Knowledge Base Relations using Web Text as Interlingua Derry Tanti Wijaya Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA, 15213 dwijaya@cs.cmu.edu Tom M. Mitchell Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA, 15213 tom.mitchell@cs.cmu.edu Abstract In recent years many knowledge bases (KBs) have been constructed, yet there is not yet a verb resource that maps to these growing KB resources. A resource that maps verbs in different languages to KB relations would be useful for extracting facts from text into the KBs, and to aid alignment and integration of knowledge across different KBs and lan- guages. Such a multi-lingual verb resource would also be useful for tasks such as machine translation and machine reading. In this pa- per, we present a scalable approach to auto- matically construct such a verb resource us- ing a very large web text corpus as a kind of interlingua to relate verb phrases to KB rela- tions. Given a text corpus in any language and any KB, it can produce a mapping of that language’s verb phrases to the KB relations. Experiments with the English NELL KB and ClueWeb corpus show that the learned English verb-to-relation mapping is effective for ex- tracting relation instances from English text. When applied to a Portuguese NELL KB and a Portuguese text corpus, the same method au- tomatically constructs a verb resource in Por- tuguese that is effective for extracting relation instances from Portuguese text. 1 Introduction In recent years a variety of large knowledge bases (KBs) have been constructed e.g., Freebase (Bol- lacker et al., 2008), DBpedia (Auer et al., 2007), NELL (Carlson et al., 2010), and Yago (Suchanek et al., 2007). These KBs consist of (1) an on- tology that defines a set of categories (e.g., Sport- sTeam, City), (2) another part of the ontology that defines relations with these categories as argument types (e.g., teamPlaysInCity(SportsTeam, City)), (3) KB entities which instantiate these categories (e.g., Steelers ∈ SportsTeam), and (4) KB entity pairs which instantiate these relations (e.g., (Steel- ers, Pittsburgh) ∈ teamPlaysInCity). The KB on- tology also specifies constraints (e.g., mutual exclu- sion, subset) among KB categories and relations. Despite recent progress in KB construction, there is not yet a verb resource that maps to these KBs: one that contains verb phrases1 that identify KB re- lations. Such a verb resource can be useful to aid KB relation extraction. A distribution of verb phrases associated with any given KB relation is also a KB- independent representation of that relation’s seman- tics which can form the basis of aligning ontologies across arbitrary KBs (Wijaya et al., 2013). Given a KB and verb resources in different languages that map to the KB, we can also begin to align knowl- edge expressed in different languages. We introduce here an approach to mapping verb phrases to KB relations using a very large ClueWeb corpus (Callan et al., …

Content truncated to 500 words