Proceedings of the Annual Meeting of the Cognitive Science Society

To acquire the meaning of a verb, language learners not onlyneed to find the correct mapping between a specific verb andan action or event in the world, but also infer the underlyingrelational meaning that the verb encodes. Most verb naminginstances in naturalistic contexts are highly ambiguous as manypossible actions can be embedded in the same scenario andmany possible verbs can be used to describe those actions. Tounderstand whether learners can find the correct verb meaningfrom referentially ambiguous learning situations, we conductedthree experiments using the Human Simulation Paradigm withadult learners. Our results suggest that although finding theright verb meaning from one learning instance is hard, there isa statistical solution to this problem. When provided withmultiple verb learning instances all referring to the same verb,learners are able to aggregate information across situations andgradually converge to the correct semantic space. Even in caseswhere they may not guess the exact target verb, they can stilldiscover the right meaning by guessing a similar verb that issemantically close to the ground truth.