Sense inventory alignment using lexical substitutions and crowdsourcing

Sense inventory induction is a topical problem of deriving a set of synsets representing concepts nsing various automatic or human-assisted methods. There might be, and actually are, mistakes in such synsets. Here we are focused on the problem of eliminating potentially duplicate synsets having exactly two words in common as the broader lntersectlon is known to be successfully addressed by beoristics. We exploit the pbenomena of lexical substitotions and microtaskbased crowdsourcing for aligning the synsets to the individual word senses. We also present an open source mobile application implementing oor approach. Our experiments on the Russian language show that the approach scales well aad dramatically reduces the number of duplicate synsets in the inventory.