Computational models of human word comprehension

Louis ten Bosch

Radboud University, Department of Language and Communication

In this presentation I aim to present an overview of computational models of human speech recognition. First, a number of well-known influential models will be taken as starting point. Special attention will be paid to their architecture and the assumptions of these models, especially the role of the representation of the input of these models (e.g. the actual acoustic representation of speech, or symbolic-phonetic descriptions of the speech signal, or representations in-between). Central theme is the modeling of word activation, word competition, and the (possible) role of morphology.
Next, I will discuss novel trends in these research area such as the introduction of deep learning, end-to-end-modelling. Finally I will touch upon the relation between these models and statistical analyses (primarily mixed effect regression models).