recent paper (Dunn et al. in Nature)

T. Florian Jaeger tiflo at csli.stanford.edu
Mon Jul 11 21:03:07 UTC 2011


Hi Joan,

I think I understood that point. But it seems to me that any diachronic
change must be created through one of the two possibilities of transmission
I outlined in my reply (biases on acquisition or direct or indirect biases
on production). Actually, a third possibility is that noisy production
creates sufficient variability from which comprehenders/learners than might
generalize in one or the other direction. But in any case, we need to
explain the *cognitive *mechanisms that explain deviation from the input. I
understand that once a certain bias is assumed to operate, we can derive the
direction of change over diachronic times (the subject of many simulation
studies on language change and the subject of work on grammaticalization,
unidirectionality, etc.).

Perhaps I misunderstand your point, but I do believe that the study of
diachronic processes in isolation would not provide sufficiently
constraining evidence to understand what plausible functional biases could
be. Of course, I agree that a lot is known about these processes and that
"no findings about functional utility, noisy channels, etc. are applicable
to explaining language structure unless they correspond to known paths and
mechanisms of change", but the existing data is still compatible with a
large number of hypotheses that vary greatly in their cognitive plausibility
and their compatibility with existing data on what's easy and hard to
process, to produce, and to acquire.

In short, both diachronic pathways of change and what is known about
cognitive mechanisms of language acquisition, production, and comprehension
constraint functional theories. Of course, there is work that acknowledges
this, but part of the motivation for the article with Hal Tily (the WIRE
article) was that we felt that there was much room for further
collaborations and knowledge transfer between research on psycholinguistis
mechanisms and research on diachronic processes.

Maybe we're talking about the same thing, using different terminology?
Sorry, if I am misunderstanding you.

Florian



On Mon, Jul 11, 2011 at 2:12 PM, Joan Bybee <joan.bybee at gmail.com> wrote:

> Dear Florian,
>
> In case you didn't get Tom's point about diachrony: no findings about
> functional utility, noisy channels, etc. are applicable to explaining
> language structure unless they correspond to known paths and mechanisms of
> change. And we do know a lot about how languages change so it is easy enough
> to seek functional explanations that correspond to known changes. These, by
> the way, rarely involve 'language transmission' if by that you mean language
> acquisition. The best way to start is to study change in progress and then
> try to find what cognitive/processing mechanisms have to be involved. That
> is the way American functionalism has been operating since the 1970's.
>
> Joan Bybee
>
>
> On Mon, Jul 11, 2011 at 1:26 PM, T. Florian Jaeger <
> tiflo at csli.stanford.edu> wrote:
>
>> Dear Tom,
>>
>> I recently started drafting a position paper for Language and Cognitive
>> Processes that outlines what I perceive to be the major challenges to what
>> I
>> might loosely call functionalist linguistics (the paper is not on
>> functional
>> linguistics, but rather on language production and to what extent it is
>> driven by considerations about communicative efficiency). I should send it
>> to you for feedback once it's ready (like, 2018 ;). But here's the gist of
>> the introduction to that paper.
>>
>> I think the first big challenge is to define 'utility' in principled
>> terms.
>> That's where, I think, recent work building on information theory and
>> Bayesian models has made much progress (although there is still tons to
>> do).
>> This work picks upon  ideas that have been around for some time and tries
>> to
>> make provide a formal backbone to them. This line of work aims derive what
>> speakers and comprehenders should do *under the assumption that language
>> use
>> is set up to facilitate *efficient and robust communication from basic
>> assumptions about communication through a noisy channel (for example,
>> Genzel
>> and Charniak's 2002 Constant Entropy Rate hypothesis, which is derived
>> from
>> Shannon's noisy channel theorem; van Son and Pols work on the amount of
>> information a segment carries and its phonetic realization; Aylett and
>> Turk's 2004 Smooth Signal Redundancy hypothesis and their test against
>> phonetic reduction; Levy and Jaeger's 2007 proof that a uniform
>> distribution
>> of Shannon information across the linguistic signal minimizes processing
>> costs under certain assumption, etc.; that work is summarized briefly in
>> Jaeger and Tily, 2011-WIRE and in much more detail in Jaeger,
>> 2010-Cognitive
>> Psychology, both refs were given in my previous email).
>>
>> As you said, notions of utility (which I am using as a placeholder term
>> for
>> all kinds of ideas as to what's good for language usage) also need to be
>> supported empirically, e.g. by psycho-linguistic studies (the Jaeger and
>> Tily 2011 paper aims to provide a 7 page summary of work on sentence
>> processing over the last four decades that, we think, linguists working on
>> language usage would benefit from knowing about).
>>
>> The second big challenge is to identify how functional (and perhaps also
>> non-functional) biases affect the transmission of language from generation
>> to generation. There's two basic logical possibilities that are mutually
>> compatible. Biases can operate during language acquisition and they can
>> operate during language production (cf. Bates and MacWhinney 1982),
>> possibly
>> involving long lasting changes due to implicit learning over previous
>> productions. The artificial language learning and iterated artificial
>> language learning studies I mentioned in my previous email provide a great
>> (though definitely not perfect!) way to study the first possibility and I
>> very much hope that researchers with training in linguistics and, in
>> particular, typology will have a strong presence in this line of work.
>> Crucially, I don't just mean iterated language learning simulations, but
>> learning experiments with actual people (or better, actual infants).
>> Today,
>> we received the reviews on our summary on this line of work from LT, so we
>> should soon have the final version with additional references up at the
>> address I mentioned in my previous email.
>>
>> There's also a rather active line of work on language adaptation in adults
>> that can be seen as addressing the second possible transmission route.
>> Most
>> adaptation work has been conduction on perception (there's, of course, a
>> long tradition of this work on phonetic perception going back at least
>> until
>> the late 60s; more recently, we have also started to show that similar
>> effects are observed during syntactic processing). However, somewhat
>> unsurprisingly to anyone who ever tried to learn another language ;), this
>> work has found that changes in perception do not necessarily affect
>> production. To the best of my knowledge, there's relatively few studies
>> that
>> investigate in a controlled way how production changes through exposure.
>> Most of them seems to be focused on phonetic production (actually, I'd be
>> curious to hear references, if people don't mind sending them to me). Of
>> course, there's tons of evidence for syntactic priming - but almost all of
>> that has focused on rapid effects, where "long lived" means that the
>> effects
>> of priming can survive for a few minutes (e.g. Bock and Griffin, 2000;
>> Chang
>> et al., 2006; Reitter et al., 2011 - searches for these names with the
>> keyword "syntactic priming" will give you the relevant references). I
>> think
>> it's only recently that folks started to look at longer-lasting changes in
>> morpho-syntactic productions as a function of exposure. For example,
>> Kaschak
>> and Glenberg (2004) showed how repeated exposure to novel structures
>> (needs
>> washed) actually increases the probability that speakers later use the
>> structure themselves. These studies do, however, not yet show that there*
>> *are
>> *functional* biases at work during such adaptive changes to one's
>> productions. But maybe readers of this list know of other work (e.g. in
>> sociolinguistics) that addresses this question? This is something I am
>> very
>> much interested in and we have several studies running in the lab that try
>> to get at this question.
>>
>> apologies for the long email.
>>
>> Florian
>>
>
>



More information about the Funknet mailing list