LINGUIST List 5.597

Tue 24 May 1994

Disc: Children's knowledge of binding and reference

Editor for this issue: <>


Directory

  1. ANDREW BARSS, Children's knowledge of binding and reference

Message 1: Children's knowledge of binding and reference

Date: Thu, 12 May 1994 01:34:42 Children's knowledge of binding and reference
From: ANDREW BARSS <BARSSCCIT.ARIZONA.EDU>
Subject: Children's knowledge of binding and reference


 In a recent LINGUIST posting, H. Stephen Straight
presents a review of and commentary upon a recent article I
coauthored ("Children's knowledge of binding and reference:
Evidence from spontaneous speech", by Paul Bloom, Andrew Barss,
Janet Nicol, & Laura Conway, Lg. 70.53-71.). I and my colleagues
are pleased that our article was chosen to inaugurate what we
hope becomes a regular forum for current article discussion.
Since Straight frames his review as a set of questions, and
presents several criticisms of our analysis, I would like
to briefly address Straight's commentary.

 In reviewing our central discussion, Straight writes:

 >"BBNC set out to demonstrate that Principles A and B of binding
 >theory inhere in Universal Grammar and are exempt even from the
 >necessity for parameter-setting language input. In support of
 >this thesis, BBNC present child language-output evidence that
 >flaws in performance rather than competence account for those few
 >cases in which even 2-year-olds violate, either in receptive or
 >expressive language processing, the prohibition on local
 >coindexing (Principle B), which rules out 1, and the requirement
 >that reflexives MUST exhibit such coindexing (Principle A), which
 >rules out 2.
 >
 >(1) Thelma touched her ["Thelma" coreferent with "her"].
 >(2) John hit himself ["John" NOT coreferent with "himself"]."

 This is not a completely accurate summary.

 First, we *do* assume that, whatever one's view is of the
intriguing child/adult differences in comprehension experiments,
there is a "necessity for parameter-setting language input" with
respect to binding. It is a simple fact that the domain for
disjointess of reference of pronouns varies, in a quite limited
fashion, across languages, as is strongly argued by Hestvik 1990
(Brandeis U. dissertation), 1993 (NLLT article), and Mazini and
Wexler 1987 (LI article). It is also a fact, although one
intimately tied up with questions of reflexive morphology, that
the domain for binding of reflexives varies considerably across
languages (refs cited, and many others). And there is no
convincing, explicit theory of the acquisition of these
cross-linguistic differences other than the parameter setting
model, in our opinion; there is no feasible way in which these
differences could be learned from scratch. Since cross-linguistic
acquisition was not our concern, we did not spell out these
assumptions in maximal detail.

 Second, we would hesitate to call the child/adult
differences we attribute to the processing system(s) involved in
comprehension, but not production, "flaws", as Straight puts it.
If there are performance considerations involved in the non-adult
performance the children exhibit, these could range from
confusion, to conflicting task demands (as Grimshaw and Rosen
1990 argue), to computational limitations in the child (as
Grodzinsky and Reinhart 1991 argue), or genuine differences in
the computational processes which the child and adult performance
systems instantiate. None of these are "flaws" in the child's
peformance systems; just differences. And if the last option is
on the right track, it would be an exciting avenue to pursue.

 Rather, we set out to explore the nature of the child's
knowledge of the binding conditions using a methodology
previously unexplored, namely spontaneous production. Since
production and perception are different cognitive acts, one would
expect that a centralized competence knowledge-deficit would
indiscriminately affect production and comprehension, while a
performance-system limitation/difference would not necesarily do
so. So, we sought out to ascertain if the children in our study
(who were significantly younger than those involved in most
comprehension studies) produced sentences which violate the adult
binding conditions. They do not, and so we conclude that the
children "know" the binding conditions. This reduces the logical
options for explaining why the children don't perform as adults
do on comprehension tasks: whatever the explanation, it isn't due
to an across-the-board competence distinction, as some authors
have suggested.

 Straight asks three questions, and presents commentary
with respect to each. I'll take these in turn.

 >"1. Why did BBNC exclude _you_ and _yourself_ from the targeted
 >database?

 >The second-person pronouns do not exhibit the ambiguity that
 >rightly led BBNC to eliminate third-person pronouns."

 We had two reasons. The first, primary one was that an
early, cursory inspection of the data (all drawn from the CHILDES
database, a rich and valuable source of spontaneous production
data which we highly recommend) indicated that there were more
tokens of first- than second-person proforms, and so many of the
former that our limited time might be best spent given close
consideration just to those. This decision was made prior to any
statistical evaluation of the data; it just seemed a practical
move, since focusing on just first-person proforms gave us
thousands of sentences to work through. Our second reason was
that second-person proforms are unmarked for number; thus we
could not be sure which of four logically possible propositions
e.g. "you wash yourself" expresses for the child, and this seemed
a potential confound.

 >"2. Why did BBNC reject the possibility that the children's
 >output is based simply on the strong positive input evidence that
 >reflexive pronouns are locally bound while non-reflexive pronouns
 >are not?

 >Knowledge of the principles of intra-clausal coreference would
 >appear to be acquirable from knowledge of what a clause is plus
 >what coreference is plus what a reflexive looks like (myself,
 >ourselves, yourself, yourselves, himself, herself,
 >theirselves--not very challenging, especially if we acknowledge
 >early errors like "ourself", "hisself", and "theirself"). What
 >does binding theory add to this?"

 Binding theory adds a precise characterization of how the
proforms *cannot* be used as well. Straight seems to suggest a
simple hypothesis-formation theory of acquisition; the child
listens to how certain words are used to express certain
thoughts, and then encodes useful aspects of this for further
use. If language acquisition proceeds in this loose way, it is
very unclear how the child will correctly converge on just that
grammar which produced the input sentences; somehow, the
hypothesis space must be highly constrained from the outset, with
certain options rigidly fixed, and others partially fixed with
certain, finite, options made available (e.g. the domains for
binding). If Straight's general theory is made more explicit, it
will, we suspect, strongly resemble the theory we assume.

 >"3. Why did BBNC dismiss the evidence for dissociation between
 >receptive and expressive language processes (p. 69)?

 >Although BBNC state that "there may be SOME dissociation between
 >production and comprehension," they leave themselves no way to
 >account for this possibility when they ask the dismissive
 >rhetorical question, "if the input grammar is separate from the
 >output grammar, what type of data modifies the output grammar?" "

 Straight raises a good point here. What we meant is that
there cannot be a complete dissociation; the production system
(e.g. the syntactic-structure-building operations performed when
a person asks a question, for example) HAS to be tied to the
input-processing system in an intimate way. They are, of course,
different psychological computational systems, and they are
not identical, nor are their computational subroutines likely to
be matched up one-to-one. Since we showed that children do not,
in any sense, violate the adult anaphora conditions in
spontaneous production, and the comprehension studies show that
they do seem to violate these conditions, one seems to be left
with two options for explanation.

(a) As some have suggested, the comprehension experiments place
the child in a position where she must violate the binding
conditions (Grimshaw and Rosen) or cannot perform the memory-rich
computations necessary to make the appropriate judgment
(Grodzinsky and Reinhart). If this explanation is adopted, the
spontaneous speech experiment we reported gives a truer index of
what the child actually knows.

(b) The child knows, by age 2 or 3, that e.g. pronouns are
characteristically disjoint from local NPs, and her production
system incorporates this; but for some reason the comprehension
system(s) don't incorporate this knowledge for another four or
five years. If true, this is exceedingly puzzling, and fully
explaining it will lead to exciting insights into the
experience-driven process of adjusting the imput processor.

 Finally, Straight comments:

 >"Their own study supports the implication that the formulating
 >processes that a child can readily infer from the reflexive vs.
 >nonreflexive input data will indeed resist modification in
 >response to a rather confusing array of exceptions (under
 >"special pragmatic conditions") to the usual coindexing
 >attributes of these two sets of pronouns. Receptive processes, on
 >the other hand, apparently respond to all manner of input
 >variation, much of which gets incorporated into receptive
 >processing without any effect on expressive processing. "

 If I am reading this correctly, Straight is proposing
that the comprehension ("formulating") system will learn
(whatever actually needs to be *learned* of) the adult system
more easily, and earlier, than the production ("receptive")
system will. But this is inconsistent with the facts of our and
other researcher's experiments: it is the comprehension
experiments which show the puzzling non-adult effect; the
production experiment we did, which examined maximally natural
uterances, showed completely adult-like performance. So, if the
comprehension experiments give an accurate measure of how the
child comprehends pronouns and reflexives, then it is the
comprehension system which lags behind for a number of years.


 --Andrew Barss
 University of Arizona
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue