LINGUIST Review
Gleitman, Lila and Barbara Landau, eds.
THE ACQUISITION OF THE LEXICON.
Cambridge, Mass.; London, England: A Bradford Book. The MIT
Press. 1994. [Reprinted from LINGUA: INTERNATIONAL REVIEW OF
GENERAL LINGUISTICS, Volume 92, Nos. 1-4, April 1994.]
Reviewed by Elsa Lattey, University of Tuebingen
Before I discuss the content of this interesting volume, I
would like to make a few comments on its form. As mentioned
above, the book is a reprint of Volume 92 of the journal
LINGUA, which contained all of the articles now available here.
All MIT Press did was to put a paperback cover on the LINGUA
volume and add an index (with double entries for Clark, H. and
Clark, H.H., to list just one such example, and entries for
Markman's "whole-object assumption" and "taxonomic assumption"
but not for Bloom's "whole object constraint" or "taxonomic
constraint", although "constraints, subsuming" later in Bloom's
paper and "constraints, syntactic and conceptual" in Fisher et
al do get listed).
The current volume also contains all of the typographical
errors of the original. This is an unfortunate phenomenon of
our times, where publishers turn over the skills they once took
pride in to the authors, who in many cases have not yet ceased
to expect (a mistake as it turns out) the editorial services of
the publisher. Some chapters are relatively error-free, while
others contain numerous errors.
Interpretation difficulties exist especially in the Williams
article, which in addition to assorted typographical mistakes
contains several errors in reference (so, for example,
8c,d should be 8b,c; Ex. 29 should have an a through d
to correspond with the text in the following paragraph --
it has nothing on the first example, followed by a through c;
on p. 27 the discussion of the rules refers to the 3rd and 5th
lines, while these are marked <2n> and <4>).
The articles have individual bibliographies, some of them quite
comprehensive (Markman, Waxman, Landau, Bloom), but this also
led to some inconsistencies, so that we find Disciullo vs. Di
Sciullo references (both also listed back to back in the index)
with a publication date of now 1986 and now 1987. Even the
preface is addressed only to the LINGUA audience, with no
special preface or paragraph added for the MIT book. All this
reflects my disappointment with a publisher of MIT Press's
status. (They can contact me for the typos if they reprint.)
But let me turn now to the content. My criticisms regarding
form notwithstanding, the volume is a useful collection of
papers dealing with a single topic (the acquisition of the
lexicon) from the viewpoints of several disciplines:
linguistics, psychology, computer science. Gleitman/Landau
point out in their preface that "during the past decade, it has
become clear from linguistic inquiry that the lexicon is more
highly structured than heretofore thought; moveover, that much
of grammar turns on critical -- and universal -- links between
syntactic and lexical-semantic phenomena." This is the point
of departure for the papers in this volume. They grew out of a
workshop held at the University of Pennsylvania, in 1992, to
judge from manuscript references in the bibliographies. The
volume is organized into six sections: 1. Nature of the mental
lexicon (papers by Williams and Levin/Rappaport Hovav, the
latter misplaced in this collection insofar as it says nothing
about acquisition, but partially justified in that it provides
some background data for the papers on verb acquisition later
in the volume); 2. Discovering the word units (papers by Cutler
and Kelly/Martin); 3. Categorizing the world (Carey, Keil); 4.
Categories, words, and language (Markman, Waxman, Landau,
Bloom); The case of verbs (Fisher et al, Pinker, Grimshaw); and
6. Procedures for verb learning (Brent, Steedman).
The book thus begins (Section 1) with a linguistic look at
lexical learning, in which Williams ("Remarks on lexical
knowledge") points out that although "the lexicon is the
repository of forms about which something special must be
learned", the abstractness of lexical knowledge is generally
underestimated, including that of phrases and paradigms.
Nevertheless, "intricate structures can be learned, and this
learning is not adequately modelled by either parameter setting
or list learning."(7) W. focuses on 'idioms', the phrases in
the lexicon, speculating that there are more of these than
lexical words, and asking what rules they obey (they obey the
phrase structure rules, e.g., but not the rules of reference,
which can lead to a clash between the syntactic and semantic
argument structures). W. extends the notion 'idiom' to include
syntactic structures and broad typological parameters, in a
graded continuum with various levels of abstractness, and
suggests that the learning process is one in which some device
tracks down statistical correlations (and especially unique
correlations) among various properties of the linguistic units
it has learned to date, which then enables the postulation of
an analysis. Levin/Rappaport Hovav ("A preliminary analysis of
causative verbs in English") deal with a more specific
linguistic phenomenon, namely causative verbs, their paper
being relevant to this volume in that it discusses verbs with
both transitive and intransitive uses (the 'causative
alternation') that are referred to in later papers (Grimshaw,
Fisher et al, Pinker) in terms of learning issues.
Section 2 addresses the question of how the learner segments
the continuous stream of sound, and the two papers here as well
as Brent's chapter in Section 6 suggest that segmentation
decisions are made based on prosodic and distributional cues in
caretaker speech. Cutler ("Segmentation problems, rhythmic
solutions") assumes that infants are born already armed with a
'periodicity bias', which suggests that "the characteristic
rhythm of speech is incorporated into infants' linguistic
competence before they acquire their first words"(98), and she
sees the procedure developing its language-specific character
via statistical properties of the language input. Bilinguals
generally have only one rhythmically driven segmentation
structure, even if their two languages have different rhythmic
properties. Kelly/Martin ("Domain-general abilities applied to
domain-specific tasks: Sensitivity to probabilities in
perception, cognition and language") suggest that language
learners can learn and exploit probabilistic information. They
use "a conspiracy of cues [syllable cues, prosodic structure,
semantic properties, preferences for certain parsing, etc.] to
identify the grammatical category of a word."(135) These cues
can be language-universal (e.g. the semantic cue to grammatical
class) or language-specific (e.g. stress in English), and they
are mutually reinforcing.
Section 3 explores how children represent their observations of
the world. Carey ("Does learning a language require the child
to reconceptualize the world?") and Keil ("Explanation,
association, and the acquisition of word meaning") take
opposing positions on the question of whether children's mental
representations change over developmental time: Carey argues,
in contrast to the so-called continuity hypothesis, that there
is a major conceptual difference between infants and adults (on
the basis of their appreciation of sortal concepts, count/mass
noun "syntax", and numerical identity), while Keil, in
developing a new model of conceptual structure ("concepts-in-
theories"), finds evidence for an intrinsic mix of domain-
general and domain-specific relations in both young children
and adults, this hybrid incorporating both probabilistic
associations and patterns of explanation. Unfortunately,
Keil's paper is too abstract for readers not familiar with the
cited literature, and would profit from more examples, both in
the historical review (=A72-4) and in the theoretical explanation
sections (=A75-6).
Section 4 focuses on how the child uses both observation and
linguistic principles to decide on a word's meaning. Markman
("Constraints on word meaning in early language acquisition")
attributes children's rapid and successful acquisition of
vocabulary despite limited information-processing abilities in
part to constraints on the kinds of hypotheses they consider
for determining word meaning: e.g. the whole-object assumption,
"which leads children to infer that terms refer to objects as a
whole rather than to their parts, substance, color, or other
properties"; the taxonomic assumption, "which leads children to
extend words to objects or entities of like kind"; and the
mutual exclusivity assumption, "which leads children to avoid
two labels for the same object"(199). There appears to be
evidence that these constraints interact, also for very young
children, and that children may rely more heavily on these
default assumptions when the learning task is more difficult.
Waxman ("The development of an appreciation of specific
linkages between linguistic and conceptual organization")
summarizes evidence from children learning English, Spanish and
French as a first language to substantiate the presence of such
constraints (whole object and mutual exclusivity), adding that
children are also predisposed to "use the grammatical form of a
novel word (e.g., count noun, proper noun, adjective) as a
guide to determining its meaning"(231). W. claims that early
results suggest that the linkage between count nouns and object
categories, being evident already at the onset of language
acquisition, may be universal [though the evidence presented
here is based on closely related languages], whereas the
linkage between adjectives and properties emerges later, may
depend on particular linguistic and conceptual knowledge, and
may vary across languages.
Landau ("Where's what and what's where: The language of objects
in space") continues the discussion of potential constraints,
reporting on work that suggests "that the spatial-cognitive
system may impose significant limitations on the kinds of
hypotheses children will entertain about the meaning of a new
word for an object vs. a place"(291). "They do so because
object [the "what" system, linked primarily to object shape]
and place [the "where" system, linked primarily to principal
axes] are represented separately and distinctly, engaging
different schematizations of the world." Experimental results
suggest "that non-linguistic representational systems may play
a part in shaping some of the character of languages"(259).
While the three preceding papers discuss the link between
perceptual-conceptual and linguistic facts, they make no claim
about direction of influence.
Bloom ("Possible names: The role of syntax-semantics mappings
in the acquisition of nominals") defends a theory that speakers
have "mappings from grammatical categories [+/- "a" before a
noun] to abstract semantic categories [count/mass N]" that
"serve to constrain hypotheses about word meaning"(297), thus
making special word-learning constraints unnecessary. The
mappings have the advantage, according to B., of having
"independent linguistic and psychological support" and
providing "a framework to explain the acquisition of all
nominals, including those such as 'dog', 'water', 'wood',
'forest', and 'Fred', and also including those count nouns that
are not names for material entities."(306) They are, however,
"not by themselves sufficient to explain children`s success at
word learning", as children are exposed to words in the absence
of the entities they describe and have to sort out a multitude
of possible meanings of the new word. Consequently, "not only
does a complete account of the acquisition of word meaning
require an explanation of how people understand the intended
reference of others, it also requires a theory of conceptual
representation."(324) Bloom thus makes a case for viewing word
learning and the nature and development of grammatical
knowledge as related phenomena.
Section 5 shifts the focus of word learning from nouns to
verbs, where it is more difficult to see the link between the
utterance context and word meaning. All authors are agreed that
as with nouns there must be constraints on children's
hypotheses about possible verb meanings. In an active debate
about the implication of the link between verb argument
structure and subcategorization frames in verb learning,
Fisher/Hall/Rakowitz/Gleitman ("When it is better to receive
than to give: Syntactic and conceptual constraints on
vocabulary growth") focus on the mediating function of the
syntactic subcategorization structures, Pinker ("How could a
child use verb syntax to learn verb semantics?") argues that
logically the process must go from experience to syntax, and
Grimshaw ("Lexical reconciliation") suggests a reconciliation
of the two approaches.
Fisher et al describe an experiment in which children were
shown scenes in which either of two related verbs could have
been uttered (e.g. giving vs. getting, chasing vs. fleeing).
They were asked to guess which novel meaning was intended, the
verbs being presented in various syntactic contexts. Although
children were influenced by both syntactic and conceptual
constraints, the syntactic ones dominated, and this was taken
to support a syntax-mediated procedure for verb acquisition.
Pinker takes issue with Gleitman (1990), but in fact also with
Fisher et al, suggesting that their arguments speak only
against a narrow associationist view in which the child would
require temporal contiguity of sensory features in the context
of the spoken verb. Since the child is assumed (also by
Gleitman) to be capable of inductive hypothesis building that
goes beyond the immediate situation, these arguments do not
hold. As for the dependence of the child on a verb's syntactic
subcategorization frame to determine its meaning, P. claims
that the subcategorization structure gives the child only very
general semantic information (such as number and type of
arguments), and so may help in determining the semantic
perspective of the utterance context. It cannot provide
information regarding the root meaning of the verb, however.
While P. agrees with G. that learning mechanisms, some
syntactically driven and others semantically, complement each
other, he objects to the use of the term "syntactic
bootstrapping" to refer to the process of inferring a verb's
meaning from its set of subcategorization frames because it
suggests a real opposition to his "semantic bootstrapping"
while in fact "the theories are theories about different
things"(385). (P.'s "semantic bootstrapping" is a theory about
how the child begins learning syntax, not about how it learns
word meanings.)
In the last paper in this section, Grimshaw proceeds from the
two controversial positions in Pinker 1989 and Gleitman 1990
"that the semantics of a word is critically involved in the
acquisition of its syntax" and "that the syntax of the word is
critically involved in the acquisition of its semantics"(411).
She critically examinines the positive features and the
limitations of each, basing her discussion on pairs of verbs
such as kill/die and melt/melt, verbs which are equally
compatible with a given context, considering that "events
typically have multiple construals" (415). G. proposes a
reconciliation of the two positions in which the learner takes
a hypothesized meaning based on observation and checks it
against linguistic mapping principles of universal grammar in
their language-specific instantiation, and then compares the s-
structure predicted by this step with the observed s-structure.
When they match, learning follows in that the verb "is entered
into the lexicon with the hypothesized lexical conceptual
structure."
Section 6 of the volume begins with a description of a computer
simulation of steps a child might take in verbal learning,
using string-local surface cues rather than global constraints.
Brent ("Surface cues and robust inference as a basis for the
early acquisition of subcategorization frames") suggests a
possible set of cues for English subcategorization frames that
operate on very limited assumptions, assuming only "the ability
to detect the ends of utterances and knowledge of a few
function morphemes and proper names"(433). B. hypothesizes
that "young children can learn lexical syntax on the basis of
partial and uncertain syntactic analyses of input utterances,
and that they can cope with the resulting misconstruals using
statistical inference"(436). Using transcripts from the
CHILDES corpus of speech directed at children between the ages
of 1;0 and 2;6, he checked out his implementation and the
hypothesis that surface functional cues and statistical
inference constitute an effective strategy by which 2-year-olds
could learn subcategorization frames in a computer simulation.
The results suggest that it is possible. The final paper in
the volume, by Steedman ("Acquisition of verb categories"),
began as a commentary on Brent's presentation, but goes beyond
that to present a good case for using statistical techniques
like Brent's to minimise the effect of contamination in the
data available to children learning language. He finds that
Brent has demonstrated "a practical technique that actually can
be used to automatically build lexicons." Nevertheless,
Steedman maintains that "the case for believing that children
acquire subcategorization and other aspects of syntax on the
basis of semantic and contextual cues remains strong"(471).
The value of this volume lies a) in its bringing together in
one place articles on a particular research topic from
different disciplinary perspectives written in such a way that
representatives from the disciplines can understand each other,
and b)in the fact that these papers, building on a joint
background and with many cross references, report on related
and often cooperative research that explores the acquisition of
the lexicon from many vantage points.
In effect, we have here a body of research that might have
evolved out of a joint research project among all participants.
The in-depth understanding thus obtained of a portion of
language is invaluable as a basis for further research.
Despite my critical remarks regarding the form of this volume,
I find the research reported on worth reading and generally
well presented. To single out just one paper: Grimshaw's
contribution is particularly helpful and informative,
organizing in an easily understandable way the issues discussed
also in other papers in the volume and giving the reader good
access to the questions involved.
What remains is the question of who should buy this book. If
you are a subscriber to LINGUA you already have it. No
additions or improvements have been made (the index isn't worth
buying the book for). If your library subscribes to LINGUA you
have access to it. If you are interested in the issues
discussed herein and want to get involved in the detail of the
arguments (and there are many individual comments to be made
and discussed that a review of 15 papers at one go must omit
because of space considerations), then you may want to have a
copy you can make notes and comments in. There are certainly
more notes in my copy than this review would suggest -- open
questions, differences of opinion, other grammaticality
judgments, other ways of approaching the issues that I would
like to be able to discuss with some of the authors. The
subject matter is certainly worthy of further debate. I
append below a few questions to particular authors (cf.
Appendix), in case they care to respond.
One general question to all: Aren't some of the "problems"
that are discussed here the result of insisting on a separation
of semantics and syntax (for example, the presence or absence
of "a" before a noun is not basically a syntactic question but
a semantic one -- does what I want to say call for the meaning
of "a" or not). I find the direction taken by Williams in
extending the concept of idiom to include so-called syntactic
phenomena a promising one.
Elsa Lattey teaches linguistics and English as a foreign
language in the English Dept. at the University of Tuebingen,
Germany. Her areas of special interest include idioms,
sociolinguistics, language contact, and second-language
acquisition.
Appendix of questions and comments:
For Williams:
p. 12: (10a,b): I think the issue here is not whether plural is
conceivable or not. In 10b "bridges" =3D "problems", but in 10a
"bucket" does not =3D "death".
p.13: How does the child distinguish when it can assign
semantic argument structure and when it cannot?
(14): English does have the possibility: cf. (for people)
killjoy, cutthroat, pickpocket, spoilsport; (for non-people)
breakfast and dreadnought (examples from Bauer, Laurie. English
Word-formation. [Cambridge Univ. Press, 1983.]
p.22 bottom: But cf. dove-dived, hung-hanged (this one with a
meaning difference).
p.24: Doesn't the schema in (35) have a superfluous asterisk
at the perfect node, given the example verb "run"? Had you
used a verb with a differentiated perfect, like "speak", this
would have justified the assignment of a form to that node --
have I understood you correctly?
p.30: You say "whatever paradigm is learned first will embody
the most distinctions." How can this be ensured?
For Levin/Rappaport Hovav:
fn.14: "toughen" may not be able to mean 'make difficult' but
it can mean 'make more difficult'. Cf. "I think we need to
toughen/simplify the exam."
p.62: re the claim that "cut" implies a volitional agent: What
about: "The sheet of paper cut my finger."?
fn.31: Why shouldn't 95b have the accompaniment interpretation,
especially if the riders are on separate horses?
Cf. also my request to Grimshaw (below) about your 1992 paper.
For Cutler:
p.94: Did the French listeners who segmented English and
Japanese speech by syllables know English and Japanese? If so,
this preference for a segmentation strategy might be a measure
of dominance in bilinguals, rather than something to be
predicted by "a decision as to which of their two languages the
bilinguals would be most sorry to lose" -- how was this
ascertained, by asking the subjects?
For Kelly/Martin:
p.120: In Zajonc's experiments, how is "the more it is liked"
measured?
p.128: The syllable transcribed with theta + schwa should be
transcribed with edh + schwa, shouldn't it --i.e. the initial
sound is voiced (on this page and the next)?
For Carey:
p.160: You mention an experiment by Baillargeon in which
babies used the existence of two numerically distinct objects
to make sense of what would be an impossible event if only one
object were involved. Please tell us what the event was
(especially since you cited an oral paper).
p.164: I think there is a third possibility: baby looks
longer at (expected) elephant/truck or elephant + cup because
it is trying to establish a link between the two.
For Markman:
p.208: Aren't you conditioning the child in the no label
situation to assign two meanings to the phrase "another one",
namely the correct taxonomic and an incorrect thematic
interpretation?
p.212: Here I asked myself how the adjectives were used (but
fortunately I found the answer in Waxman's article: "These are
the ak-ish ones").
For Waxman:
p.236: Are the children in the novel adjective condition given
ak-ish and dob-ish without any previous information regarding
what akas or dobus are?
p.240: Aren't you really testing for a knowledge of "another
one" and "-ish" rather than categorization here?
p.242: I didn't understand how the superordinate level trials
differed from basic level trials.
p.248: For Spanish consider that the tested forms are not
different grammatical categories: to my knowledge "el rojo"
'the red one' has all the qualities of a so-called count noun
like "el caj=F3n" 'the large box'.
For Bloom:
p.316: What is an "unlearned" mapping and how can a child
possess it?
For Fisher et al:
p.361: *Want eat the apple is just as out as *Make eat the
apple.
p.361: How could the child rule out "touch" in the context of
"Let's see if there's cheese in the refrigerator" if she's
blind? Her mother could have touched the cheese, which is what
the blind child would have to have done.
For Pinker:
3.1.2: You say that pairs that referred to exactly the same
set of situations would be exact synonyms. But in the kinds of
pairing you are making here they wouldn't mean the same thing
because of the shift in perspective, isn't that so? Also
"winning without beating" is different from the rest, as it's
not a matter of perspective with them but of the nature of the
second argument. By the way, the Celtics and the Nets did win
something: some of the games they played. In the Coke example,
isn't the operator selling it to you by machine proxy?
p.397: (last line) Wouldn't syntax be rather helping to
distinguish between the meaning of "find" and that of "find
that", i.e. the contribution of "that" (why hypothesize two
different meanings for "find")?
p.398: What are "psych-verbs" with ambiguous roles? "Fear"
and "frighten" seem to have a clear perspective.
For Grimshaw:
p.414: Has Levin and Rappaport Hovav (1992) been published? I
find it difficult to imagine why "arrossire" should be a change
of state predicate and "blush" not. Change of state is a
semantic property, isn't it?
For Brent:
p.434: Is (2d) really ungrammatical for you? Sounds o.k to me
(even better with "that").
p.449: "Some ambiguous words occur as nouns more often than
they occur as verbs" -- please give some examples.
p.461: "as off we trot to play" -- analysable as a purpose
adjunct?
p.462: Something missing in this sentence?: "Because the cues
are check agreement, John eat ... is mistaken for a tensed
clause."
PS: The importance of function elements demonstrated here
should motivate speakers not to leave them out in foreigner
talk.
|
