Self Organizing Notes 9-1-06
Computational and evolutionary aspects of language
Martin A. Nowak, Natalia L. Komarova & Partha Niyogi
Nature, v417, 6 June 2002
Nowak et al. provide a coherent though understandably cursory introduction to some important aspects of language evolution; language theory, learning theory, and evolutionary dynamics. Starting with something like a set of first principles, they present working definitions of language and grammar along with descriptions of the computational machinery able to construct and evaluate these forms. The article build from Gold's learning theory framework where a learner:
The authors close out with a discussion of the role of Unviersal Grammars and their role in language acquisition and evolution.
A few points for discussion:
The authors suggest a parallel between how a child learns language, and how language is learned in general. Does this hold for the situations we are going to study? Does anthropomorphizing computational elements in this way help us by providing convenient metaphors, or perhaps do more harm by rooting us too firmly in familiar models?
How well do Gold’s firm assumptions from learning theory hold if we assume malicious/imperfect agents? Or even the relaxed statistical learning approaches discussed on p. 612? For example in Angluin’s models of learning with counterexamples, how might an agent be able to cope when learning a language from an agent intent on confusion?
The issue of vocabulary seemed a little overlooked. There was some discussion of what meanings were important to convey, and the resultant problems with poverty and ambiguity. But what about issues where learners disagree on what concepts are important to include in a language from the start. Is this just a particularly harsh version of poverty, where whole regions of the language are absent, or is it another problem altogether?
In a related issue, consider the evolution of the Universal Grammar. If the evolution of a UG within the generations of a population is determined by the environmental input, might the UG ever become too highly tuned? Can a UG be designed to leave in a degree of adaptability?
Regarding the paradox of language acquisition – and given our conversations in class last week... Language is being described as a social process building from small interactions between learners. I can't really argue against this social constructivist view, but it seems to highlight the problem with the UG as a rationalist or innate element in the mind of a child/learner.
If we can understand language as technology, and a social technology at that, how well do models like Adaptive Structuration Theory apply to language acquisition and use?