Connectionist Perspectives on "Late Talkers"

A slightly different view of the late-talkers’ predicament comes from connectionist perspectives on language development (nativist and interactionist views were covered previously). As a framework, connectionism emphasizes the graded, domain-general, and input-sensitive nature of cognition (McClelland & Patterson, 2002).

Connectionist simulations of language acquisition do not explicitly posit any innate mechanisms except those common to biological neural networks, nor do they ascribe any particular role to social interaction except simply providing linguistic input to the network. Nonetheless, such simulations accurately model a variety of detailed linguistic phenomena, including aspects of sentence reading, speech segmentation, and speech production (Christiansen & Chater, 2001).

Because connectionist principles are based on neural computation, and because language clearly results in some sense from neural computation, both nativist and interactionist predictions for effective intervention can usually be recast in the connectionist framework.

For example, reward conditioning of social interaction might benefit vocabulary acquisition by increasing the salience of linguistic input – one need not posit a special role for social cognition. Likewise, some developmental delays may simply “iron themselves out” not due to additional environmental triggers or the belated expression of a “shape-bias gene,” but rather because of individual differences in learning rate or the diversity of their early-life experiences.

Finally, children may appear to be sensitive to “intent” when in fact they are discriminating designed from non-designed objects on the basis of simple perceptual features (Colunga & Smith, 2005). Thus, predictions motivated by nativist and interactionist accounts can typically also be explained in terms of connectionist principles.

However, connectionism does suggest a few unique predictions for the absence of a shape bias among late-talkers. For example, attentional-learning accounts of word acquisition suggest that the associative links between names and perceptual features may become weighted with experience. Object features that more reliably correlate with naming patterns (such as shape for solid objects) ultimately become more salient with experience.

According to this view, the experience of late-talkers may have included an abnormally large proportion of objects whose names cannot be differentiated on the basis of shape; the corollary of this view is that certain types of words may be over-represented in the small vocabulary of late-talkers. One would expect these over-represented words to relate to living things and nonsolid objects, since many living things and nonsolid objects can be very similar in shape but have different words (Colunga & Smith, 2005).

Other word types may be under-represented in late-talkers' vocabulary, such as items where shape is a diagnostic feature (for example, solids and possibly non-living things). These predictions could be verified through analysis of the words that late-talkers did know, as measured by the MacArthur Communicative Development Inventory (Jones, 2003) or carefully selected stimuli from the Peabody Picture Vocabulary Test.

If such words were over-represented, late-talkers might show a more pronounced material bias for nonsolid and for simply shaped solid objects (Colunga & Smith, 2005). There is already tentative support for this prediction: late talkers have shown nonsignificant trends towards the use of a material bias (Jones, 2003). Cross-linguistic differences in shape bias usage - and related connectionist simulations - have also shown that syntax can also influence the shape bias (Colunga & Smith, 2005). Based on this work, late-talkers may also be less proficient in discriminating count and mass nouns.

If these predictions were verified, successful intervention techniques would likely involve training on categories well-organized by shape in an attempt to cultivate a shape bias, which has been shown to increase the rate of word learning (Smith et al., 2002). Another possible intervention is training on count vs. mass noun distinctions. It is even possible that extensive training on simple (and even non-verbal) shape matching tasks could instill a habit for late-talkers to attend to shape, which might ultimately translate into an acceleration of word learning. If verified, this latter prediction would be particularly compelling, because it would reflect an instance of far transfer of learning, one of the “holy grails” of training research.

References:

Colunga E, & Smith LB. (2005). From the lexicon to expectations about kinds: a role for associative learning. Psychol Rev. 2005 Apr;112(2):347-82.

Jones, S. S. (2003). Late talkers show no shape bias in a novel name extension task. Developmental Science 6(5):477-483.

McClelland, J. L. & Patterson, K. (2002). Rules or Connections in Past-Tense inflections: What does the evidence rule out? Trends in Cognitive Sciences. 6:11 (2002), pp 465-472

Smith LB, Jones SS, Landau B, Gershkoff-Stowe L, & Samuelson L. (2002). Object name learning provides on-the-job training for attention. Psychol Sci.;13(1):13-9.