Introduction

A substantial body of experimental evidence has demonstrated that labels have an impact on infant categorisation processes. Yet little is known regarding the nature of the mechanisms by which this effect is achieved. I distinguish two accounts, supervised name-based categorisation and unsupervised feature-based categorisation, and describe a neuro-computational model of infant visual categorisation, based on self-organising maps, that implements the unsupervised feature-based approach. The model successfully reproduces experiments demonstrating the impact of labelling on infant visual categorisation.

The model predicts that the impact of labels on categorisation is influenced by the perceived similarity and the sequence in which the objects are presented to infants and that the observed behaviour in infants is due to a transient form of learning that might lead to the emergence of hierarchically organised categorical structure. New evidence corroborates these predictions. I argue 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 may play a more mundane but nevertheless powerful role as additional features that are processed in a similar fashion to other features that characterise objects and object categories.

Background

Facilitation: In an extensive series of studies, Waxman and colleagues have provided evidence for the view that labels have an impact on category formation in young infants. Using a novelty preference procedure, infants were familiarised with a series of objects taken from the same category and then given a choice between two novel objects, one of which is from the familiarised category. During familiarisation, the objects are accompanied by a novel label such as 'dax' or a neutral carrier phrase such as 'Look at this'. Infants show a preference for the out-of-category object when familiarised with the novel label, but not with the neutral carrier phrase.

These findings are interpreted as demonstrating that `labels facilitate categorisation', that labels 'act as invitations to form categories' and that labels 'highlight the commonalities between objects'. Their findings suggest that the effects are specific to the consistent use of labels that could be words in the infant's language. Tones and buzzers don't achieve the same effect and the same label needs to be used consistently throughout familiarisation. Using different labels doesn't work. These findings have been reported for infants well before their first birthday, indicating that labels have an impact on infant categorisation before they produce their first words and before they have acquired a substantial receptive vocabulary.
Interference: A contrasting set of studies by Sloutsky and Robinson point to a different conclusion: That novel labels overshadow the processing of visual stimuli by young infants and, therefore, that auditory stimuli (including novel labels) interfere with category formation. They base their conclusions on a series of habituation studies in which infants are familiarised with compound auditory-visual stimuli and are then exposed to a dishabituation stimulus that changes either the auditory component or the visual component. Infants notice the change in the auditory component but not the change in the visual component. Failure to dishabituate to a change in the visual stimulus is interpreted as a failure to process the visual stimulus as a result of the auditory stimulus overshadowing the visual information during familiarisation. It should be noted that familiar auditory stimuli, such as well-known names, do not produce such dramatic overshadowing effects. Furthermore, novel labels interfere with visual processing in 10 month olds but not at 16 months.

The finding that novel labels interfere with visual processing in 10 month-old infants does not sit well with the finding that novel labels can facilitate the categorisation of objects: Auditory dominance effects are more likely to impede categorisation than to facilitate infants' attention to the commonalities between objects.
Methodological Differences: There are important differences in the procedures used to test infants in these studies that might readily explain the apparently discrepant findings: The categorisation studies are typically conducted in a one-on-one setting and don't require infant habituation. Auditory dominance studies are infant-controlled and testing does not occur until the infant reaches an habituation criterion.

The stimuli and timing of events are also quite different. The categorisation studies make use of objects or pictures which are likely to be familiar to infants and which are readily interpretable as single whole objects, and perhaps more likely to be members of a category. The auditory dominance studies commonly exploit complex shapes unfamiliar to infants and which are readily segmented into separate objects with no obvious category alignment. In categorisation studies, labels are presented after infants have had an opportunity to examine the visual objects whereas in auditory dominance studies labels are syncronised with the onset (and often the offset) of the visual stimuli. Given the transient character of auditory stimuli compared to visual stimuli in the real world (as opposed to the laboratory), the human cognitive system may have evolved to prioritise auditory stimuli in order ensure that sufficient attentional resources are available for speeded processing.
The observed auditory dominance effects may derive from the syncronisation of visual and auditory onsets, a condition which does not typically hold when objects are labelled for infants.

Interpretation

Although it may be possible to reconcile these apparently disparate findings by appealing to differences in stimuli, timing and experimental procedures, some difficulties of interpretation of the impact of labels on infant categorisation still remain. Waxman and colleagues report that infants show a novelty preference for the out-of-category object, and thereby evidence for categorisation, only when the familiarisation phase includes a novel label, not when the familiarisation phase involves a neutral carrier phrase, such as "Look at that". Hence, the locus of the categorisation effect would appear to be the novel label. However, it is well-established that infants will demonstrate novelty preference effects in experimental situations that indicate category formation, but in the absence of any auditory stimulus. For example, after being familiarised with a sequence of cats, infants will prefer to look at a novel dog over a novel cat.

Interpretation of the role of the novel label as a facilitator of category formation is compromised by the fact that infants exhibit robust category formation in the absence of any auditory stimulation.
An alternative interpretation of the categorisation studies is that the neutral carrier phrase interferes with category formation, whereas novel labels don't block the process. However, this interpretation does not sit well with auditory overshadowing studies, since novel rather than familiar auditory stimuli are likely to interfere with visual processing. Neutral carrier phrases, such as "Look at this one", are likely to be familiar to infants and should not hinder visual processing.

Towards a Resolution

In order to determine the locus of a labelling effect on infant categorisation, we adapted an earlier experiment by Younger. In the original experiment,Younger presented 10-month-old infants with line drawings of animal-like objects which varied in the length of their legs and necks, the size of their tails and the spread of their ears.
Lorem ipsum dolor sit amet, sapien platea morbi dolor lacus nunc, nunc ullamcorper. Felis aliquet egestas vitae, nibh ante quis quis dolor sed mauris. Erat lectus sem ut lobortis, adipiscing ligula eleifend, sodales fringilla mattis dui nullam. Ac massa aliquet.
Lorem ipsum dolor sit amet, sapien platea morbi dolor lacus nunc, nunc ullamcorper. Felis aliquet egestas vitae, nibh ante quis quis dolor sed mauris. Erat lectus sem ut lobortis, adipiscing ligula eleifend, sodales fringilla mattis dui nullam. Ac massa aliquet.
Lorem ipsum dolor sit amet, sapien platea morbi dolor lacus nunc, nunc ullamcorper. Felis aliquet egestas vitae, nibh ante quis quis dolor sed mauris. Erat lectus sem ut lobortis, adipiscing ligula eleifend, sodales fringilla mattis dui nullam. Ac massa aliquet.