The role of social-pragmatic cues in word learning: a neural network model

Abstract

Human infants begin to produce speech at the beginning of the second year of life. Some theories propose that language is acquired by simply recognising statistical properties in linguistic input. Other hypotheses consider the interactive environments in which humans are raised, looking for links between emerging social skills and word learning.

The social-pragmatic theory of language acquisition suggests that the foundation of word learning is an ability to read the intentions of another, especially the intention to communicate (Akhtar and Tomasello, 2000). The theory posits a special role for the ability to respond to joint attention, a form of intention recognition which has been shown to facilitate word learning (Baldwin, 1993). Social-pragmatic theory argues that the cognitive abilities that develop in the second year of life are not merely coincidental, but are an essential component of language acquisition.

The first goal of this thesis is to develop a neural network model to investigate word learning by implementing key concepts of the social-pragmatic theory. Cognitive skills that develop in the second year of life are shown to facilitate word learning, with the model reproducing characteristics of the ’vocabulary spurt’ that can be seen around 18 months of age (O’Grady, 2005). The second goal is to relate the model to current neurobiological research. The neural correlates of intention recognition and lexical retrieval are tentatively defined, permitting a discussion of the brain regions common to both processes. The prefrontal cortex, in particular, is discussed to investigate how its general functions could be harnessed by mechanisms for word learning and intention recognition. One novel contribution of the model is to tie together joint attention and word learning using a rewardbased learning scheme.