What are currently used biologically plausible computational models/frameworks of early learning in children?

Personally, I have used cascade correlation neural nets to model pronoun acquisition in children, but in hindsight, I consider this framework to be not neurobiologically plausible.

Are there existing computational frameworks that have been used for early-learning and development in children that have a reasonable biologically justification? For example: just being a neural net model is not a reasonable justification, but a neural net model witch has local update rules (like Hebbian learning) and models neurons more realistically (instead of just saying they are a sigmoid function without any further biological justification) would be fine. General non-connectionist frameworks with empirical justification are also of interest, even if they don't have a solid reductionist account.


I have asked an alternate version of this question on Linguistics.SE that focuses on language acquisition instead of general learning, but considers both children and adults.

  • $\begingroup$ What does 'early learning' mean is this context? It sounds way too broad, but perhaps that is due to my relative ignorance of developmental psychology. $\endgroup$
    – zergylord
    Commented Jan 31, 2012 at 4:19
  • $\begingroup$ @zergylord I guess by early learning I would refer mostly to the learning when the child's brain is still undergoing massive neurogenesis... i.e. before it settles down to adult levels of plasticity. $\endgroup$ Commented Jan 31, 2012 at 4:21
  • $\begingroup$ In that case I'd specify a domain of interest (e.g. Computational models of pronoun acquisition early learning in children), since a general model is currently quite out of reach. $\endgroup$
    – zergylord
    Commented Jan 31, 2012 at 4:27
  • 1
    $\begingroup$ Alternatively, you could the term 'plausible algorithms' instead of 'computational models', since the latter tends to be task dependent whilst the former is a paradigm in which to build models. $\endgroup$
    – zergylord
    Commented Jan 31, 2012 at 4:30
  • 1
    $\begingroup$ then I will get my own paper (and the ones it references) back as an answer. I don't want to restrict to a single task, because then the models you get back have very little meaning. The goal is to have models and frameworks that generalize across many related tasks (like CC NN is used across all parts of developmental psychology, for instance). I was hoping that by leaving the domain more open ended, I could get answers from areas I am not as familiar with. However, if it is too broad I could restrict it to models to do with linguistic/language-acquisition. $\endgroup$ Commented Jan 31, 2012 at 4:31

1 Answer 1


Here's something I dug up for language: a Computer Science Thesis from Boulder:

The Sensorimotor Foundations of Phonology: A Computational Model of Early Childhood Articulatory and Phonetic Development (1994)

it discusses what it calls HABLAR (Hierarchical Articulatory Based Language Acquisition by Reinforcement learning).

From the reductionist/biology side, developing brains have a much higher population of gap junctions. I would speculate that this would create a much more analog brain, as gap junctions are modeled diffusively, unlike synaptic connections, which can be modeled in a more binary way. I would speculate that the analog driving of the environmental and genetic signals leads to the formation and orgnaization of the synaptic connectivity.

I wonder if one could find a correlation between plasticity and gap junctions. They are found to connect axons laterally in hippocampal cells (a region of brain that remains plastic through adulthood)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.