lg as lists vs lg as skill

[Joyce Tang Boyland]

The current discussion, despite the theoretical leanings of the
participants, seems to retain the presupposition that language consists of
lists of words, lists of constructions, lists of sentences.
This is a useful presupposition for the purpose of
discussing language as a formal system, which has its place;
but I would like to put up for consideration the idea
that language be thought of not as a set of lists, but as a skill.

Empirical language acquisition researchers may not be thrilled by the
simple theory from skill acquisition that all skills start out as computed
and then are stored, but ACT-R (the current version of John Anderson's
theory of skill acquisition) is set up to handle more complex transitions
between stored and computed representations.
For example one can learn a rule from stored examples, but one also
then turns instantiations of rules into stored examples.

A form like `brokeded' would be straightforwardly explained in ACT-R,
as a case of a rule being applied to a form that got stored after
being used previously.

A lesson that ACT-R teaches us, I think, is that, well, sometimes you
need to compute a form and sometimes you need to retrieve it from storage,
but where these forms come from is not what we really care about;
what we are really trying to explain is the skill (language using)
that these forms serve.  And as you apply the skill you just happen to do
many things by rote and some things by computation.

Something this view buys us is that language competence (as in accepting
vs. rejecting a string) isn't the main thing, with performance being
the by-product; rather, the performance is the main thing and the competence
(the lists of accepted vs. rejected strings) is the by-product.

I'm just beginning a project (with Eric Scott and perhaps John Anderson)
to model the creation of collocations as a product of skill learning in ACT-R,.
to complement my recent work on long-term syntactic priming (evidence that
stored forms are used in production).

Joyce Tang Boyland

Joyce.Tang.Boyland at alverno.edu
Alverno College
Milwaukee, WI  53234-3922

>>
>> On the other hand, the simple, old-fashioned idea of "regularization",
>> replacing an exception with a form based on the application of a regular
>> rule, doesn't predict attested forms of the sort: _It got brokeded_
>> (meaning "It got broken"). You probably wouldn't want to say that the child
>> uttering _brokeded_ has added an uninflected root _broked_ to her lexicon
>> to which she is now adding the suffix -ed. (I won't vouch for the exact
>> example, but I've seen things like that--maybe it was "tookeded".) This
>> really looks more like the result of the convergence on a particular form
>> in response to various pressures in the system. This could be called
>> "computation" but may also hint that the dichotomy of storage vs.
>> computation is already a bit off track.
>>
>> Greg Thomson
>>

>> Barsalou via Tomasello:
>> Finally, this tradeoff between computation and storage is manifest in many
>> current theories of skill.  Essentially, novices are viewed as having
>> stored few cases, and so have to compute, whereas experts are viewed as
>> having stored many cases, and so don't have to compute (they just
>> retrieve).  This goes back to Chase and Simon's work on chunking and chess,
>> and it can be found in the modern theories of John Anderson (ACT*), Gordon
>> Logan (exemplar-based skill model), and Alan Newell (SOAR).  Each includes
>> two ways of producing a behavior--computation vs. retrieval--and assumes
>> that novices mostly compute but increasinly retrieve as they become
>> expert.
>>