'learning paths' instead of parameters

Mon Sep 29 15:58:11 UTC 2014

Dear All:

The subset principle is often mistakenly interpreted as favoring the most
narrow grammar, or being coextensive with conservatism.  Actually this is
incorrect.  The principle has to do with the size of the language (viz
number of sentences generated), not with the restrictiveness of the rules
or piecemeal learning vs. generalization.

In particular, the principle requires that for any two grammars under
consideration by the learner, if grammar A generates the set of sentences
{a}, which includes all of the sentences generated by grammar B, i.e., the
set {b} plus the complementary set {b'} of sentences not generated by
grammar B (In other words, the set {a} = {b + b'}), then {b} is a subset of
{a}.  For example, a grammar that contains an optional rule is a superset
of a grammar that contains a obligatory rule for the same structure.
According to the subset principle, if the learner starts out assuming
grammar A, which generates the larger language, then there will be no
direct evidence that this grammar is incorrect in the language input to the
learner (assuming other idealizations like the no negative evidence
assumption etc.)  However, if one starts out with grammar B, which
generates the subset language, then as soon as a sentence from {b'} is
heard, then this will disconfirm the hypothesis that grammar B is correct.

In the case of the current question about headedness, it seems clear that
the languages generated here are not in a subset-superset relation because
they appear to generate languages of roughly the same size, but that are in
overlapping distribution rather than one being a subset of the other.  The
differences have to do with the underlying structural descriptions, which
can be learned from distributional factors like pronoun substitution,
movement and so on.  So, whether or not you believe in the subset principle
as something that real children actually consider in real life language
learning (which I have doubts about), it is important to get the logic
right.

Peter Gordon

On Sat, Sep 27, 2014 at 9:37 AM, Brian MacWhinney <macw at cmu.edu> wrote:

> Dear Maria Teresa,
>
>      In Fritz’s example, positive evidence for the second principle (NPs
> are head-initial) amounts to negative evidence for the first principle (all
> phrases are near initial).   If one assumes a framework such as the
> Competition Model in which cues compete, then there is no problem with
> negative evidence, because the child would just be learning cue strength
> and the cue of NP would dominate over the general cue.  This would reduce
> the problem to a case of a “benign” exception to the logical problem.
>      More importantly, if you compare Fritz’s four-stage pathway with that
> of the learning of the past tense, you will see that the order is largely
> wrong.  For the past tense, it is the high frequency exceptions that are
> learned first, so the fourth step would come before the first in cases such
> as “went” or “fell”.  Such irregularities in areas where we have clear
> evidence about the sequence might lead us to question the notion of
> pathways in general.
>
> —Brian MacWhinney
>
>
>
> n Sep 27, 2014, at 4:19 AM, mariateresa.guasti at unimib.it wrote:
>
>  Hi,
>
> one comment on Brian about the pathway: if the child assumes 1., she can
> move to 2. when she listen to head-initial NPs (and this is positive
> evidence). It is like the past tense (regularization and then learn the
> specific forms based on what one hears). Am I wrong?
>
>
> About the specific examples of head-initial/head final, there are
> proposals in the literature that infants can decide about directionality
> based on prosody. In one of the two papers, it is discussed the case of
> German and Dutch (languages that have a mixed situation)
>
> <22e5549e.jpg>
>
> <22e554ad.jpg>
>
>
>
> Two more papers on the same issues:
>
>
>
> <22e554bd.jpg>
>
>
> <22e554dc.jpg>
>
>
>
>
> Maria Teresa Guasti
>
>
> At 06.04 27/09/2014, Brian MacWhinney wrote:
>
> Fritz and John,
>
>     The general idea that children follow learning pathways seems like a
> reasonable one.  However, the question is whether these pathways arise from
> preexisting universals or from the actual encounter of children with the
> specifics of language.  The example of a pathway that Fritz provided
> involved a gradual retreat from overgeneralization.  It had these four
> steps:
> 1. First s/he will assume that ALL phrases are head-final, even noun
> phrases.
> 2. Next s/he will assume that ALL NPs are head-initial
> 3. Next s/he will learn the class of exceptions to 2.
> 4. Finally, s/he will learn the purely idiosyncratic exceptions.
>
> A pathway of this type is not in accord with the Subset Principle, because
> that principle is fundamentally conservative, always favoring the most
> narrow grammar.  In the pathway Fritz describes, the child starts out with
> a huge generalization and then has to retreat, possibly relying on negative
> evidence.
>
> The idea that the child is fundamentally conservative crops up in many
> accounts.  For example, in my “Mechanisms of Language Acquisition” book
> from 1987, Fodor and Crain argued for conservatism in the form of the
> Subset Principle.  I argued for it in terms of my theory of item-based
> learning, and Berwick assumed it in his learning on error analysis.  The
> literature is full of evidence for conservatism from older work by Maratsos
> and Kuczaj to newer work by Lieven, Rowland, Ambridge, and colleagues.  In
> the area of lexical overgeneralization, there is similarly strong evidence
> for conservatism.
>
> Of course, children are not always conservative.  Often they are missing
> forms and have to resort to overextension.  However, I can’t think of any
> evidence for the type of raw overextension from the very beginning
> suggested by the learning pathway you propose.  Is there any empirical
> evidence for this?
>
> Underspecification, as illustrated by the initial collective
> interpretation of distributive quantifiers such as “each” would seem to be
> a counterexample.  However, the conflict between distributive and
> collective interpretations is often a bit opaque in actual communicative
> contexts.
>
>  Brian MacWhinney
>
> On Sep 25, 2014, at 2:09 PM, Grinstead, John <grinstead.11 at osu.edu> wrote:
>
> Hi Fritz,
>
> It's interesting to think about an alternative to parameters, so thank you
> for starting a conversation.
>
> That said, I can't really think of any syntactic phenomena that have the
> properties you have mentioned. That is, starting with one look and then
> completely switching to the opposite is pretty well unattested, as far as I
> can remember, at least in morphology, syntax, semantics (subjects, DO
> clitics, articles, mass-count, double objects, subject case, subject-aux
> inversion). All of these constructions get studied because there are
> variations from the adult patterns, but most of them are relatively subtle.
>
> So, child English speakers start using accusative case pronouns in subject
> position for a while (e.g. Him cry.), but don't switch to a completely
> Ergative-Absolutive system (not really sure what that would look like,
> anyway, but it would probably involve some interesting forms in object
> position that we don't ever see). Subject-verb agreement can take a while
> to develop in Spanish, but kids don't start marking direct object agreement
> (as in Swahili or Georgian) or indirect object agreement on verbs (as in
> Euskera) all of a sudden.
>
> Most constructions just look adult-like.
>
> It's like they're not willing to take a stab at a construction until they
> have a pretty good idea of what's going on and when they do, it looks
> pretty good. This is Stephen Crain's "conservative learner" (I think that
> was his phrase) and William Snyder's "Grammatical Conservatism".
>
> The exceptions are not that numerous and they are what we spend most of
> our time thinking about.
>
> Starting from something that seems unspecified and moving to something
> that's specified, on the other hand, is maybe more frequently observed. So
> Brooks and Syrett and Musolino and Pagliarini have observed in separate
> studies that children's interpretations of distributive quantifiers like
> "each" seem to allow them in collective situations. This tolerance
> gradually goes away and is replaced by a restrictive interpretation.
>
> I hope that helps.
>
> Best,
>
> John
>
> -------------------------------------------------------------------------
> John Grinstead
> Associate Editor
> Language Acquisition: A Journal of Developmental Linguistics
> Department of Spanish and Portuguese
> The Ohio State University
> 298 Hagerty Hall - 1775 College Road
> Columbus, OH  43210
>
> Tel. 614.292.8856
> Fax. 614.292.7726
> grinstead.11 at osu.edu
>  https://u.osu.edu/langlab/
> -------------------------------------------------------------------------
>
> From: Frederick Newmeyer < fredericknewmeyer at gmail.com>
> Reply-To: " info-childes at googlegroups.com" < info-childes at googlegroups.com
> >
> Date: Thursday, September 25, 2014 1:24 PM
> To: " info-childes at googlegroups.com" < info-childes at googlegroups.com>
> Subject: 'learning paths' instead of parameters
>
> Dear colleagues,
>
> I hope that you don’t mind a question from an outsider who has a very
> small mastery of the acquisition literature.
>
> There is a recently-developed approach to formal syntax that has abandoned
> the idea of innate parameters directing the course of acquisition. In their
> place, it posits universal ‘learning paths’, determined by ‘general
> cognitive optimization strategies’, and whose operation to a considerable
> degree mimics the work once done by parameter hierarchies. In a nutshell,
> it posits that for any structural (or constructional?) domain, the child
> makes the most general hypothesis first, and then gradually over time zeros
> in on the adult grammar.
>
> Let me give a concrete example. Let’s say that a language is consistently
> head-final except in NP, where the noun precedes its complements. However,
> there is a definable class of nouns in this language do follow their
> complements. And a few nouns in this language behave idiosyncratically in
> terms of the positioning of their specifiers and complements (much like the
> English word ‘enough’, which is one of the few degree modifiers that
> follows the adjective).
>
> According to the theory I am describing, the child will go through the
> following stages of acquisition:
> 1. First s/he will assume that ALL phrases are head-final, even noun
> phrases.
> 2. Next s/he will assume that ALL NPs are head-initial
> 3. Next s/he will learn the class of exceptions to 2.
> 4. Finally, s/he will learn the purely idiosyncratic exceptions.
>
> Is there any evidence that acquisition actually proceeds in this ‘orderly’
> manner? I remember from years ago some inconclusive discussion about the
> ‘subset principle’, but I would very very interested to hear what you have
> to say about recent work that bears on the scenario that I have described
> above.
>
> Thanks!
>
> fritz
>
> --
> You received this message because you are subscribed to the Google Groups
> "Info-CHILDES" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to info-childes+unsubscribe at googlegroups.com.
> To post to this group, send email to info-childes at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/info-childes/CANZD-QKs%3Dknk9z4CJ3j5ntHbiQPfkZ8NC1%3DugeBP5jsVapqt9w%40mail.gmail.com
> <https://groups.google.com/d/msgid/info-childes/CANZD-QKs%3Dknk9z4CJ3j5ntHbiQPfkZ8NC1%3DugeBP5jsVapqt9w%40mail.gmail.com?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>
> --
> You received this message because you are subscribed to the Google Groups
> "Info-CHILDES" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to info-childes+unsubscribe at googlegroups.com.
> To post to this group, send email to info-childes at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/info-childes/D049CEFB.3594A%25grinstead.11%40osu.edu
> <https://groups.google.com/d/msgid/info-childes/D049CEFB.3594A%25grinstead.11%40osu.edu?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "Info-CHILDES" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to info-childes+unsubscribe at googlegroups.com.
> To post to this group, send email to info-childes at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/info-childes/BA4DA931-8876-4FCF-AD14-1CF648D07784%40cmu.edu
> <https://groups.google.com/d/msgid/info-childes/BA4DA931-8876-4FCF-AD14-1CF648D07784%40cmu.edu?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>
>  Prof.ssa Maria Teresa Guasti Ph.D.
> Full Professor of Linguistics and Psycholinguistics
> University of Milano-Bicocca
> Department of Psychology
> Piazza dell'Ateneo Nuovo 1
> 20126 Milano
> mariateresa.guasti at unimib.it
>
> http://www.psicologia.unimib.it/03_persone/scheda_personale.php?personId=63
>  www.cladproject.eu
> www.bilinguismoconta.it
>
> --
> You received this message because you are subscribed to the Google Groups
> "Info-CHILDES" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to info-childes+unsubscribe at googlegroups.com.
> To post to this group, send email to info-childes at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/info-childes/auto-000057239662%40pablo2.pablo.unimib.it
> <https://groups.google.com/d/msgid/info-childes/auto-000057239662%40pablo2.pablo.unimib.it?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>
>
>  --
> You received this message because you are subscribed to the Google Groups
> "Info-CHILDES" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to info-childes+unsubscribe at googlegroups.com.
> To post to this group, send email to info-childes at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/info-childes/1D47AEEC-1ADD-4EA3-8EC6-AE0D81BB7FC5%40cmu.edu
> <https://groups.google.com/d/msgid/info-childes/1D47AEEC-1ADD-4EA3-8EC6-AE0D81BB7FC5%40cmu.edu?utm_medium=email&utm_source=footer>
> .
>
> For more options, visit https://groups.google.com/d/optout.
>

-- 
Peter Gordon, Associate Professor
Biobehavioral Sciences Department, Box 180
Teachers College, Columbia University
1152 Thorndike Hall
525 W120th St.
New York, NY 10027
Phone: 212 678-8162
Fax:     212 678-8233
E-mail: pgordon at tc.edu
Web Page:http://www.tc.columbia.edu/faculty/index.htm?facid=pg328

-- 
You received this message because you are subscribed to the Google Groups "Info-CHILDES" group.
To unsubscribe from this group and stop receiving emails from it, send an email to info-childes+unsubscribe at googlegroups.com.
To post to this group, send email to info-childes at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/info-childes/CAJE3P%2B9kzn%3D9aPfv39A6yB0tg9jNKJCB3gvB6_2h-e%3Dd%2BA99yg%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/info-childes/attachments/20140929/054abc3a/attachment.htm>