Modeling Cognitive Mechanisms in an Embodied System

Twomey, K. E. & Cangelosi A (forthcoming; 2018, July). Modeling Cognitive Mechanisms in an Embodied System. European Society for Cognitive and Affective Neuroscience, Leiden, Netherlands

The nature of the cognitive mechanisms driving development has seen intense debate in psychology and neuroscience. In particular, language acquisition, a fundamental component of cognitive development, has served as a focal point. While children acquire language with apparent ease, the task is complex, since each new word a child hears can refer to a theoretically infinite set of referents (Quine, 1960). Consequently, a range of mechanisms have been proposed by which children could narrow down this referent space. For example, on constraints-based accounts, children possess or learn explicit rules about how words map to the world; on sociocommunicative accounts children capitalize on an understanding of their interlocutor’s communicative intent; and on associative learning accounts the ability to learn low-level associations between words and nonlinguistic cues is sufficient to bootstrap language acquisition.

In this talk we will discuss how work in developmental robotics can shed light on current controversies in the cognitive development field. We present recent research on language acquisition in developmental robots which implements the associative learning account. We discuss examples of developmental experiments illustrating children’s ability to use embodied cues and existing vocabulary knowledge to learn new words. We then consider embodied developmental robotics models using iCub, which capture these empirical results (Morse et al. 2015; Morse & Cangelosi 2017; Twomey, Morse, Cangelosi & Horst, 2017). The implications for theories of cognitive development of such embodied robotic approaches will also be discussed.