A Multilingual Predicate Matrix

A Multilingual Predicate Matrix Maddalen Lopez de Lacalle3, Egoitz Laparra⋆, Itziar Aldabe⋆, German Rigau⋆ 3 Elhuyar Foundation ⋆ IXA group, University of the Basque Country, UPV/EHU m.lopezdelacalle@elhuyar.com,{egoitz.laparra, itziar.aldabe, german.rigau}@ehu.eus Abstract This paper presents the Predicate Matrix 1.3, a lexical resource resulting from the integration of multiple sources of predicate information including FrameNet, VerbNet, PropBank and WordNet. This new version of the Predicate Matrix has been extended to cover nominal predicates by adding mappings to NomBank. Similarly, we have integrated resources in Spanish, Catalan and Basque. As a result, the Predicate Matrix 1.3 provides a multilingual lexicon to allow interoperable semantic analysis in multiple languages. Keywords: Lexicon, Semantics, Multilinguality 1. Introduction Predicate resources such as VerbNet (Kipper, 2005), FrameNet (Baker et al., 1997), PropBank (Palmer et al., 2005) and WordNet (Fellbaum, 1998) offer individually in- teresting characteristics not provided by their alternatives. Unfortunately, these semantic resources are developed in- dependently and they are not fully integrated in a com- mon framework. Thus, a common semantic infraestructure would allow the interoperability among all these lexicons. With this aim, we developed the Predicate Matrix (L´opez de Lacalle et al., 2014b; L´opez de Lacalle et al., 2014a; L´opez de Lacalle et al., 2016 fc), a lexical resource resulting from the automatic integration of multiple sources of predicate information including FrameNet, VerbNet, PropBank and WordNet. One of the few projects working on the integration of pred- icate information is SemLink (Palmer, 2009). The aim of this project is to link together different predicate resources via manual mappings. In contrast, the Predicate Matrix is built by automatic methods. Following different methods and techniques to automatically integrate different knowl- edge bases that contain predicate and role information, we have completed and extended the existing manual map- pings provided by SemLink. In (L´opez de Lacalle et al., 2016 fc), we demonstrate that these methods increase the number of mappings that are included in SemLink. Further- more, our automatic methodology makes easier to main- tain updated the set of mappings when new versions of the knowledge resources (each one developed independently) are released. As a result, in (L´opez de Lacalle et al., 2016 fc), we present the Predicate Matrix 1.2 which integrates SemLink and a new set of mappings obtained by automatic methods. Section 3. presents the main steps to build the Predicate Matrix 1.2 as well as its main characteristics. The semantic resources that are part of the Predicate Ma- trix 1.2 only contain an English verbal lexicon. But, if any other semantic resource is linked to any of the resources included in the Predicate Matrix the integration of the new resource can be done straightforwardly. In this paper we demonstrate this feature by extending the Predicate Ma- trix to English nominal predicates and to Spanish, Basque and Catalan languages (cf. Section 4.) and we also present the Predicate Matrix 1.31 that includes nominalizations and multilingual predicates. In the case of the English nominal predicates, we use Nom- Bank (Meyers et al., 2004). The projection to Spanish and …

Content truncated to 500 words