FrameNet on the Way to Babel: Creating a Bilingual FrameNet Using Wiktionary as Interlingual Connection

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 1363–1373, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics FrameNet on the Way to Babel: Creating a Bilingual FrameNet Using Wiktionary as Interlingual Connection Silvana Hartmann† and Iryna Gurevych†‡ † Ubiquitous Knowledge Processing Lap (UKP-TUDA) Department of Computer Science, Technische Universit¨at Darmstadt ‡ Ubiquitous Knowledge Processing Lap (UKP-DIPF) German Institute for Educational Research and Educational Information www.ukp.tu-darmstadt.de Abstract We present a new bilingual FrameNet lex- icon for English and German. It is cre- ated through a simple, but powerful ap- proach to construct a FrameNet in any language using Wiktionary as an inter- lingual representation. Our approach is based on a sense alignment of FrameNet and Wiktionary, and subsequent transla- tion disambiguation into the target lan- guage. We perform a detailed evaluation of the created resource and a discussion of Wiktionary as an interlingual connection for the cross-language transfer of lexical- semantic resources. The created resource is publicly available at http://www. ukp.tu-darmstadt.de/fnwkde/. 1 Introduction FrameNet is a valuable resource for natural lan- guage processing (NLP): semantic role labeling (SRL) systems based on FrameNet provide se- mantic analysis for NLP applications, such as question answering (Narayanan and Harabagiu, 2004; Shi and Mihalcea, 2005) and information extraction (Mohit and Narayanan, 2003). How- ever, their wide deployment has been prohibited by the poor coverage and limited availability of a similar resource in many languages. Expert-built lexical-semantic resources are ex- pensive to create. Previous cross-lingual trans- fer of FrameNet used corpus-based approaches, or resource alignment with multilingual expert-built resources, such as EuroWordNet. The latter in- directly also suffers from the high cost and con- strained coverage of expert-built resources. Recently, collaboratively created resources have been investigated for the multilingual extension of resources in NLP, beginning with Wikipedia (Nav- igli and Ponzetto, 2010). They rely on the so- called “Wisdom of the Crowds”, contributions by a large number of volunteers, which results in a continuously updated high-quality resource avail- able in hundreds of languages. Due to the ency- clopedic nature of Wikipedia, previous work fo- cused on encyclopedic information for Wikipedia entries, i.e., almost exclusively on nouns. This is not enough for resources like FrameNet. Such resources need lexical-semantic information on various POS. For FrameNet, information on the predicates associated with a semantic frame – mostly verbs, nouns, and adjectives – is crucial, for instance gloss or syntactic subcategorization. A solution for the problem of multilingual ex- tension of lexical semantic resources is to use Wiktionary, a collaboratively created dictionary, as connection between languages. It provides high-quality lexical information on all POS, for in- stance glosses, sense relations, syntactic subcate- gorization, etc. Like Wikipedia, it is continuously extended and contains translations to hundreds of languages, including low-resource ones. To our knowledge, Wiktionary has not been evaluated as an interlingual index for the cross-lingual exten- sion of lexical-semantic resources. In this paper, we present a novel method for the creation of bilingual FrameNet lexicons based on an alignment to Wiktionary. We …

Content truncated to 500 words