Semantic Links for Portuguese

Semantic Links for Portuguese Fabricio Chalub1, Livy Real1, Alexandre Rademaker1,2, Valeria de paiva3 1 IBM Research Brazil, 2 FGV/EMAp Brazil, 3 Nuance Communications USA fchalub@br.ibm.com, livym@br.ibm.com, alexrad@br.ibm.com, valeria.depaiva@gmail.com Abstract This paper describes work on incorporating Princenton’s WordNet morphosemantics links to the fabric of the Portuguese OpenWordNet- PT. Morphosemantic links are relations between verbs and derivationally related nouns that are semantically typed (such as for instance “tune-tuner” — in Portuguese “afinar-afinador” – linked through an agent link). Morphosemantic links have been discussed for Prince- ton’s WordNet for a while, but have not been added to the official database. These links are very useful, they help us to improve our Portuguese WordNet. Thus we discuss the integration of these links in our base and the issues we encountered with the integration. Keywords: Wordnet, Morphosemantics, Portuguese 1. Introduction Fellbaum, Osherson and Clark (Fellbaum and Clark, 2007) discuss how, to aid automatic reasoning using WordNet, it would be useful to classify and label the relations between nouns and verbs that are derivationally and semantically re- lated. Many traditional paper dictionaries include morpho- logical derivations between verb and noun pairs, but they tend to simply list them as run-ons without any information on their meaning. For example, the large-scale database of categorial variations of English lexemes, known as Cat- Var (Habash and Dorr, 2003) contains some 100,000 unique English word forms; however, no information is given on the meanings of these words. Similarly, for Portuguese, the “Portal da L´ıngua Portuguesa”1 lists more than 5,400 dever- bals with their respective verbs, but again has no meanings associated. By contrast, the work on WordNet itself pays special at- tention to meanings as logical constructs. They say “We are currently working to transform WordNet into a Knowl- edge Base that better supports such reasoning and inferenc- ing” (Fellbaum and Clark, 2007, p.02). Thus it is surprising to us that for several years they have provided a collection of files, manually checked, with almost 17k typed links, re- lating verbs and nouns, but that these links have not been made part of the official distribution of WordNet, nor of the online version, available for users’ consultation. For many NLP tasks and also for people trying to construct semi-automatically wordnets in other languages such a re- source would seem very useful. We started our own version of morphosemantic links, not knowing about the Princeton Wordnet ones to begin with. The method we used to pro- duce our original morphosemantic links in Portuguese was somewhat tortuous, as deciding what semantics types are relevant and how to label them are not trivial tasks. When we realized the existence of the Princeton morphoseman- tics links we decided to integrate our previous work with theirs, as described below. 2. OpenWordNet-PT and NomLex-PT Our main reason for producing lexical resources is to cap- ture the meaning of natural language expressions in rep- 1http://www.portaldalinguaportuguesa.org/. resentations suitable for performing inferences, that is, we want to do computational semantics. Semantic information in the shape of lexical resources is …

Content truncated to 500 words