Inductive and deductive inferences in a Crowdsourced Lexical-Semantic Network

Proceedings of Recent Advances in Natural Language Processing, pages 740–746, Hissar, Bulgaria, 7-13 September 2013. Inductive and deductive inferences in a Crowdsourced Lexical-Semantic Network Manel ZARROUK LIRMM manel.zarrouk@lirmm.fr Mathieu LAFOURCADE LIRMM lafourcade@lirmm.fr Alain JOUBERT LIRMM alain.joubert@lirmm.fr Abstract In Computational Linguistics, building lexical-semantic networks and validat- ing contained relations are paramount issues as well as adding some reasoning skills in order to enrich these knowledge bases. In this paper we devise an in- ference engine which aims at producing new "potential" relations from already existing ones in the JeuxDeMots network. This network is constructed with the help of a GWAP (game with a purpose) thanks to thousands of players. It handles terms and weighted relations between these terms, and currently contains over 2 mil- lion relation occurences. Polysemous terms may be refined in several senses (bank may be a bank>financial institu- tion or a bank>river) but as the net- work is indefinitely under construction (in the context of a Never Ending Learn- ing approach) some senses may be miss- ing at a given time. The approach we proposed here is founded on the tri- angulation method through two kinds of inference schemes: deduction (top- down from generic to specific terms) and induction (bottom-up from specific to generic terms). A blocking mechanism, whose purpose is to avoid proposing highly dubious new relations, is based on logical and statistical constraints. Auto- matically inferred relations are then pro- posed to human contributors to be vali- dated. In case of invalidation, a reconcili- ation dialog is undertaken to identify the cause of the wrong inference: an excep- tion, an error in the premises or a previ- ously undetected confusion due to pol- ysemy on the central term common to both premises. 1 Introduction Developing resources in NLP is one of the cru- cial issue of the field. Most of the existing lexico- semantic networks have been constructed man- ually, like for instance the famous WordNet. Of course some tools are generally designed for consistency checking, but nevertheless the task remains time consuming and costly. Fully auto- mated approaches are generally limited to term coocurrences as extracting precise semantic re- lations between terms from text remains really difficult. New approaches involving crowdsourc- ing are flowering in NLP especially with the ad- vent of Amazon Mechanical Turk or in a broader scope Wikipedia and Wiktionnary, to cite the most well known examples. WordNet ((?) and (?)) is such a lexical network based on synsets which can be roughly considered as concepts. EuroWordnet (?) a multilingual version of Word- Net and WOLF (?) a French version of Word- Net, were built by automated crossing of Word- Net and other lexical resources along with some manual checking. (?) constructed automati- cally BabelNet a large multilingual lexical net- work from term coocurrences in the Wikipedia encyclopedia. A highly lexicalized lexical-semantic network can contain concepts but also plain words (and multi-word expressions) as entry points (nodes) along with word meanings. The idea itself of word senses in the lexicographic tradition may …

Content truncated to 500 words