Labels as Features (Not Names) for Infant Categorization: A Neurocomputational Approach

Abstract

A substantial body of experimental evidence has demonstrated that labels have an impact on infant categorization processes. Yet little is known regarding the nature of the mechanisms by which this effect is achieved. We distinguish between two competing accounts: supervised name-based categorization and unsupervised feature-based categorization. We describe a neurocomputational model of infant visual categorization, based on self-organizing maps, that implements the unsupervised feature-based approach. The model successfully reproduces experiments demonstrating the impact of labeling on infant visual categorization reported in Plunkett, Hu, and Cohen (2008). It mimics infant behavior in both the familiarization and testing phases of the procedure, using a training regime that involves only single presentations of each stimulus and using just 24 participant networks per experiment. The model predicts that the observed behavior in infants is due to a transient form of learning that might lead to the emergence of hierarchically organized categorical structure and that the impact of labels on categorization is influenced by the perceived similarity and the sequence in which the objects are presented. The results suggest that early in development, say before 12 months old, labels need not act as invitations to form categories nor highlight the commonalities between objects, but they may play a more mundane but nevertheless powerful role as additional features that are processed in the same fashion as other features that characterize objects and object categories.

Learned Labels Shape Pre‐speech Infants’ Object Representations

“…Alternatively, the observed looking time differences could reflect a novelty preference. Specifically, several “labels‐as‐features” accounts of early representational development assume that labels initially serve as one among multiple nonreferential features in object representations (Gliozzi, Mayor, Hu, & Plunkett, 2009; Sloutsky & Fisher, 2004; Sloutsky & Lo, 1999); for example, the word strawberry and the color red will have the same status in a speaker's representation of the fruit. Thus, if a stored representation incorporates a label, then encountering the object without the label results in an incongruent online representation (Lupyan, 2008).…”

Electrophysiological evidence of early word learning

“…Note, however, that our findings are also compatible with association-based accounts of early word learning. These assume that infants use domain-general mechanisms to learn the meaning of words and extend them to novel exemplars by keeping track of cross-situational regularities, again without requiring infants to understand the communicational intentions (Gliozzi, Mayor, Hu, & Plunkett, 2009;Sloutsky & Robinson, 2008;Smith & Yu, 2008). Nevertheless, whether infants learn the meanings of words through increased associations between words and their referents, or through lexical principles such as that of extendibility, the present study shows that infants as young as 9 months already can perform word-to-world mappings in an adult-like manner, even for exemplars they have never seen before.…”

Children’s referent selection and word learning

“…Recent interdisciplinary research has begun to address this issue by integrating insights from developmental psychology with computational and robotic techniques to explore the perceptual and cognitive processes underlying empirically observed behaviour (Cangelosi, Schlesinger, & Smith, 2015;Gliozzi, Mayor, Hu, & Plunkett, 2009;McMurray et al, 2012;Morse & Cangelosi, in press;Samuelson, Smith, Perry, & Spencer, 2011;Westermann & Mareschal, 2014). Computational models of word learning simulate how children behave (e.g., pointing to the flamingo and not the dog or the fish) based on what they see and hear (e.g., one novel object, two known competitor objects and the novel word flamingo).…”