Tata Consultancy Services Linguistic Data Consortium for Indian Languages (LDC-IL)

Lexical networks can be used with benefit for semantic analysis of texts, word sensedisambiguation (WSD) and in general for graph-based Natural Language Processing.Usually strong relations between terms (e.g.: cat --> animal) are sufficient to help for thetask, but quite often, weak relations (e.g.: cat --> ball of wool) are necessary. Ourpurpose here is to acquire such relations by means of online serious games as otherclassical approaches seems impractical. Indeed, it is difficult to ask the users (nonexperts) to define a proper weighting for the relations they propose, and then we decidedto relate weights with the frequency of their propositions. It allows us to acquire first thestrongest relations, but also to populate the long tail of an already existing network.Furthermore, trying to get an estimation of our network by the very users thanks to a tipof the tongue (TOT) software, we realized that they rather tend to favor the relations ofthe long tail and thus promote their emergence. Developing the long tail of a lexicalnetwork with standard and non-standard relations of low weight can be of advantage fortasks such that words retrieval from clues or WSD in texts.