Modelling the co-emergence of linguistic constructions and action concepts: the case of action verbs

In this paper, we are concerned with understanding how linguistic and conceptual structures co-emerge, shaping and influencing each other. Most theories and models of language acquisition so far have adopted a ‘mapping’ paradigm according to which novel words or constructions are ‘mapped’ onto existing, priorly acquired or innate concepts. Departing from this mapping approach, we present a computational model of the co-emergence of linguistic and conceptual structures. We focus in particular on the case of action verbs and develop a model by which a system can learn the grounded meaning of a verbal construction without assuming the prior existence of a corresponding sensomotorically grounded action concept. Our model spells out how a learner can distill the essence of the meaning of a verbal construction as a process of incremental generalization of the meaning of action verbs, starting from a meaning that is specific to a certain situation in which the verb has been encountered. We understand the meaning of verbs as evoking a grounded simulation rather than a static concept and propose to capture the meaning of verbs via generative statistical models that support simulation, in our case Hidden Markov Models. Statistical models can represent the essence of a verb’s meaning while modelling uncertainty and thus variation at the surface level of (observed) action performances. We show that by extending an existing framework for construction learning, our approach can account for the co-emergence of linguistic and conceptual structures. We provide proof-of-concept for our model by experimentally evaluating it on matching, choice and generation tasks, showing that our model can not only understand but also produce language.