On the one hand, work by such figures as Elman (e.g. 1991 - See also Clark 1993) and
Christiansen and Chater (in preparation) suggest ways of re-evaluating the properties of
the learning involved in coming to behave grammatically. Elman’ s work seeks to
establish what particular connectionist systems are capable of learning, given variations
in their architecture, properties of the training data, and the influence of varying general
cognitive capacities. An example of this is the role of manipulating the capacity of short-
term ‘memory in Elman (1991) which showed that a plausible type of general cognitive
maturation could have the same effects as the kinds of ‘hyper-benevolent’ structuring of
training data otherwise required to enable a network to converge on optimal
generalisations. Christiansen and Chater, on the other hand, urge a kind of Copernican
revolution, in which the vastly greater rate of change of languages as compared to
genotypes is a justification for supposing that, to a significant extent, it is languages that
are adapted to our cognitive peculiarities and limitations, rather than our cognitive
abilities which are specifically and genetically optimised for language.
On the other hand, a range of empirical results concerning the cognitive capacities of
non-human animals indicates that many abilities otherwise easily regarded as being
language specific adaptations are found in species without ‘language’ but with their own
versions of utterance-activity. Chinchillas (Kuhl & Miller 1978) and cotton-top tamarins
(Ramus et al. 2000),4 for example, perform surprisingly well at tasks requiring different
(familiar and unfamiliar) language groups to be distinguished from one another - at least
as well as human infants of certain ages. 5 To the extent that monkeys can do this,
though, it seems reasonable to suppose that the powers of discrimination in question
come for ‘free’ as a consequence of capacities not in any way selected ‘for’ language.
Equally important, although in different ways, are some of the results from ape language
research (ALR), in particular Savage-Rumbaugh’s Sherman, Austin and Kanzi (Savage-
Rumbaugh 1986, Savage-Rumbaugh, Shanker and Taylor 1998). Kanzi’ s comprehension
is roughly equivalent to that of a two and a half year old human child. His production is
more difficult to quantify precisely, partly because it is difficult to determine how much
it is affected by the physical constraints of the lexigram board system. To be interesting
and significant ALR research does not need to produce a non-human ape with levels of
fluency comparable to an educated human adult. The point rather is that every increase in
performance is a blow against the view that to make any headway at all with language
requires specifically human biological endowments.6 For our present purposes what is
especially notable about Sherman, Austin and Kanzi is the lexigram board technology
used for the research and training, and, in Kanzi’ s case, an unusual biography and
learning history.
First, on lexigram boards, recall that chimpanzees and bonobos have, compared to
humans, very limited control over their own vocalisations. Where much other ape
language research turned to manual sign-language, Savage-Rumbaugh’s team used
physical grids of ‘lexigram’ symbols, both in the form of fixed keyboards which
triggered recordings of the relevant spoken term, and as folding boards which could be
carried around and used on the move as well as privately by her subjects (who manifestly
did engage in self-directed lexigram activity). These external, publicly accessible
resources clearly allow some of the memory and other demands of symbolic processing
to be handled by non-neural resources, significantly augmenting the cognitive powers of
their users (See Cowley and Spurrett 2003).
Second, and just as importantly, Kanzi’ s learning biography was unusual. Reared by
Matata, a foster mother, he was present during, and apparently uninterested in, her own
laborious trials with lexigram boards. Matata managed to show facility with only six
different lexigrams, given 30 000 trials over a period of 2 years (Savage-Rumbaugh et al.
1998: 17). When she was taken away to be bred at another site, though, Kanzi soon began
5 / 23