For decades, controversy has swirled around the question of whether some
kind of innate capacity is required for language to develop normally.
Ostensibly, there has been progress over the years on the old ‘nature-
nurture’ question. Only the most die-hard theorists would now publicly
argue for anything other than some sort of interactionist position. Nonetheless,
as Sabbagh & Gelman (S&G) note in their review of The emergence
of language, there is little clarity about the nature of that interaction or
about what constitutes an emergentist position on it. As good scientists, S&G
propose to evaluate emergentism, despite its vagueness, by describing a
‘strong’ emergentist position as one that must include the following tenets:
domain-general cognitive processes (‘buzzsaws’ as they call them) like
attention and memory are sufficient to account for the elegance of language
as well as the ease and speed of its acquisition by children; these domain-
general processes do so without explicit rule-based knowledge representations;
and the same processes account for all aspects of language. There is
much to agree with in the review by S&G. They cite the importance of
development and performance as pluses for emergentist views, but they also
raise serious questions about the limitations of emergentist accounts: these
often either leave unanalyzed crucial concepts like similarity or hide what
look suspiciously like rule-governed knowledge representations. Despite
their concerns, S&G are gracious in their conclusions; they seem cautiously
optimistic about the enterprise of accounting for language development in an
emergentist framework. Such optimism, however, seems appropriate only
insofar as the framework is not taken in its strong form. S&G's arguments
(made by others as well) about the vague and hidden aspects of some
emergentist models give the lie to the claim that all aspects of language can
be accounted for by the same domain-general processes. Those arguments
also raise questions about whether domain-specific rule-based representations can be completely done away with.
A Domain-specific Rule Generation Using Model-Driven Architecture in Controlled Variability Model