An Experiment of Use and Reuse of Verb Valency in Morphosyntactic Disambiguation and Machine Translation for Euskara and North Sámi

There are a number of well known resources dealing with verb valency including PropBank (Palmer et al., 2005), VerbNet (Kipper et al., 2006) and VALLEX (Hajič et al., 2003). These include thematic roles, morpho-syntactic specifcations and selection preferences. A comparatively wide definition of valency including subcategorization information on all mentioned linguistic levels is applied here. However, these resources have not often been used in rule-based NLP tasks such as machine translation or disambiguation. Bick (2000) uses syntactic verb valency tags specifying e.g. transitivity and selection preferences for various NLP tasks. The use of verb valency is on a high level of grammatical analysis and requires other elaborated linguistic resources. Bick (2000) uses tags specifying transitivity preferences such as "preferably transitive, but potentially intransitive" but also selection preferences, e.g. specifying a human accusative. Agirre et al. (2009) successfully apply valency information, i.e. case subcategorization information, to the Spanish->Euskara MT system Matxin in order to improve NP/PP translation. They present different kinds of tests enriching their machine translation system with different techniques. In all cases, the combinations of techniques that include valency information produce the best results especially in recall and F-score. This paper describes an experiment for the application of verb valency in Euskara and North Sámi rule-based NLP applications, i.e. morphosyntactic disambiguation and machine translation. 10 frequent verbs each are annotated to improve the analysis, and later the effects on the application is evaluated. The main objective of the experiment is improving linguistic resources for North Sámi and Euskara taking advantage of pre-existing existing resources in one language and transfering them to the other language. Other works (Antonsen et al., 2010) have shown that the reuse of grammatical resources between both related and unrelated endangered languages is possible and provides useful results, especially on a high level of linguistic analysis. In Antonsen et al. (2010) especially the reuse of the dependency grammar is described. A number of problems that syntax alone cannot handle can be resolved by semantically richer information included in verb valency. Verb valency annotation is applied on a high level of linguistic analysis and is therefore useful for reuse even for unrelated languages.