Documente Academic
Documente Profesional
Documente Cultură
Chapter
11
Some Evidence
Grammars
for Impaired
Myrna
Gopnik
Over the last few years, specific language impairment (SLI ) has become a
hot topic because it may have the potential to tell us something about the
biological basis of language . I am afraid that comments about this
research, on both sides, have often generated more heat than light . If we
are really interested in the science of it all , then it is important to get the
issues out on the table and seewhich ones we can agree about , which ones
are still outstanding , and how we could resolve them . This chapter is
intended to make a stab at clarifying some of the issues.
The fact that we would even be discussing language and biology in the
same breath would have been unimaginable when I was an undergraduate
in 1955 (though Darwin , himself , did suggest that learning language might
be an instinct ). Over the last 40 years the picture has changed radically .
Research has shown that all languages, despite their seeming differences,
are built on the same general plan and that most children acquire their
native language as easily as they learn to walk upright , without explicit
teaching and with no apparent effort . Newborns can't do it , but it appears
in the first year of life , and the necessary precursors for this achievement
seem to be there at birth . Experiments with very young babies have shown
that humans come equipped with special abilities to selectively pay
attention to and process language ( KuhI1991 ; Kuhl and Meltzoff 1997) .
These data suggest that there is a biological basis to language . It follows
that if language is part of the biological endowment of humans , then
humans must have some genetic properties that build the particular kinds
of brain circuitry that are specialized for human language . If this is true ,
then it would not be surprising to find that some change in this genetic
endowment can interfere with the way brain circuitry is built and thereby
impair the ability to acquire or use language in the normal way . And in
fact this does happen . Though most children do acquire language without
264
Gopnik
any consciouseffort, there are some children who have real trouble with
language, and their linguistic problems persist into adulthood. What
makes this population particularly interesting is that this disorder seems
to be associatedwith some geneticfactors and neurological anomalies. If
this is true, then this situation provides a ready-made natural experiment
to investigatevarious hypothesesabout the biology of language.
This " ready-made" looks easier than it is, however. There are lots of
piecesto the argument that have to be put together, linguistic, psycholinguistic, genetic, and neurological, before we can be sure about what we
are seeing. The first problem is to show that at least some casesof this
disorder are really genetic.
11.1 Genetics
Several epidemiological studies have shown that an individual with a
developmentallanguagedisorder is significantly more likely to have a relative who is also affectedthan is an unaffectedindividual (Tomblin 1989,
1996). Data from the ongoing Genetic Language Impairment Project
conducted at McGill University confirms this pattern of familial clustering of SLI (Palmour 1997). In this study, we looked at subjectswho had a
clinical history of language impairment, but no history of any other
exclusionarycriteria, suchasimpaired auditory acuity, mental retardation,
or autism. Then we took detailed family histories. Of the 95 subjectsin this
initial study, 53 (55.8%) had at least one affected first- or second-degree
relative. In 21 of the families there was a clear pattern of multigenerational impairment. Our researchteam has also found evidenceof familial
aggregation in language-impaired subjects in England (Gopnik 1990),1
Greece(Dalalakis 1994), and Japan (Fukuda and Fukuda 1994).
So we have convergingevidencefrom a number of independentstudies
that show that languageimpairment clustersin families. The question is,
" Is this pattern the result of somegeneticfactors or is it the result of social
or linguistic factors that prevail in this family?" The latter seemsextremely
unlikely becausesome membersof the family have perfectly normal languageand other individuals in the samefamily appear to have grammars
that violate the universal properties that are found in language. Within
genetics, the way to distinguish betweenthe influence of the environment
and the influence of genesis to compare the pattern of impairment in
monozygotic (identical) with that in dizygotic (fraternal) twins. If the
crucial factors for clustering are social, then both kinds of twins should
265
look similar. If , on the other hand, the pattern is due to genetic factors,
then the twins with the samegenesshould be more alike than those who
are merely siblings genetically but have shared a similar social context.
Severalindependentstudieshave shown that this disorder is significantly
more concordant in monozygotic twins than in dizygotic twins (Bishop,
North , and Donlan 1995; Tomblin 1997; Tomblin and Buckwalter 1998).
Therefore, both the epidemiological data and the twin studiesstrongly
suggestthat genetic factors are associatedwith this disorder. But genes
do not code for behavior such as language. They code for proteins that
control developmentthat, in turn, may have consequences
for the way in
which languagedevelops. To really understandwhat is going on, then, we
have to find somelikely consequences
that the geneticfactors could have
for development; and sincelanguagehappensin the brain, the most likely
place to look is neurology.
11.2 Neurology
Previous studieshave documentedneurological anomaliesin the brains of
subjectswith familial languageimpairment (Plante et al. 1991). Our team
has examinedmagneticresonanceimages(MRIs ) from five of the affected
adult membersof the families we have been studying and has found that
" comparedto controls, the CSF (cerebro-spinal fluid)/ grey and CSF/ grey
+ white ratios were significantly high in FLI -adults [familial language
impaiffilent] . . . whereas the grey/white ratio was significantly low. . . .
Thesefindings suggestthat FLI in adults may be associatedwith cortical
atrophy" (Kabani et al. 1997). Though thesedata are all very interesting
and go in the right direction, they are far from the end of the story. The
global patterns documentedin these studies do not provide the kind of
fine-grained picture that we needif we are to really understandthe details
of how the neurological anomalies interact with the linguistic problems.
Our imaging team, headedby Alan Evans, is now developing procedures
that can look at the MRI data in much finer detail, and this has the
potential for clarifying someof thesequestions.
One interesting question is how these neuroanatomical anomalies
arise. We are just beginning to get somehints of how this might happen.
Current evidenceseemsto suggestthat this geneticimpairment interferes
with normal brain developmentin the fetus. Gallagher and Watkin (1997)
used 3D ultrasonic neuroimaging to study in utero brain development
from 24 to 32 weeks of gestational age. Three of the fetuseswere from
266
Gopnik
Neuropsychology
268
Gopnik
11.4 Linguistics
Our research goals in linguistics over the years have been clear: to
describethe grammatical systemthat producesthe pattern of errors that
are observed. After 10 years and almost 100 language-impaired subjects
and well more than 100 controls representingfour different native languages(English, French, Greek, and Japanese
), the data convergeto tell
us that the language-impaired subjects cannot construct normal representations for grammatically complex words and they therefore cannot
use rules that depend on the content of theserepresentations. For example, though they appear to have no problems with the larger grammatical
categoriesof languagelike noun and verb, they are unable to recognize
that words can be composedof roots and affixes. They therefore do not
construct abstract linguistic features such as tense, number, aspect, case,
or gender, and they do not recognizethat there are linguistic rules that
operate on these features. But this does not mean that they cannot use
words that, from the point of view of the nonnal grammar, have these
features. For example, in spontaneousspeechthey usually use the correct
forms such as " books" or " walked." They can do this becausethey
memorize theseforms as unanalyzed chunks that in their lexicon simply
encodethe semanticmeaning of " more than one" or " in the past" rather
than the sublexicalfeatures" plural" or " past." They are more accuratein
269
using the " more than one" forms than the " in the past" forms. This is
probably becausein English the relationship between the form with the
plural marker and the semanticcontext of " more than one" is very regular
and direct and therefore is easy to memorize. However, the relationship
between the past-marked forms and temporal past context is extremely
complicated and not at all regular or direct (Shaer 1996), and therefore it
is more difficult to assign a consistentmeaning relationship between the
inflected foffi1 and the meaning of pastnessby any rote means.
How do we know that theseindividuals do not construct complex representations? We have two sourcesof evidence, linguistic and psycholinguistic. The linguistic evidencecomesboth from spontaneousspeechand
from a wide seriesof linguistic tests. One of the things that makesparents
notice that their children have difficulties with languageis that they make
mistakes in spontaneousspeechin foffi1s like past tensesand pronouns
long after their playmates and siblings have got the systemfigured out.
And these problems persist into adulthood. But as these individuals get
older, they do not make theseerrors very often. Many of them get tense
right over 80% of the time. This is in contrast to the control subjects, who
get tense right virtually all of the time. Marking tense in your native
language is an automatic, unconsciousprocess, not something that you
get wrong. And this leads us to a very interesting problem. Is someone
who gets tenseright 80% of the time impaired, or simply absent-minded?
Let's put it another way. Generally speaking, people get their own name
right all of the time. If someonegot his or her name right 80% of the time,
we might begin to wonder. What our model suggestsis that the languageimpaired subjects, in fact, never get tense right . They do not have the
category " tense" in their grammar. What they do 80% of the time is
produce a word that has the samesurfaceform as the marked foffi1 in the
normal grammar. It looks like they get tensemarking right most of the
time, but they get there by a very different route- by memorization or by
explicit rule (Paradis and Gopnik 1997).
The only way to figure out what these individuals really know about
language is to give them linguistically significant tests. The data from
thesediagnostic testsconvergeto tell us that the pattern of what they can
do and what they cannot do is strikingly similar in all of the five populations- speakersof English from Canada and England, and speakersof
French, Greek, and Japanese
- that we have looked at (seefigure 11.1).
These tests show that the language-impaired speakershave significant
problems with grammaticality judgments, tense production, and deriva-
271
Table 11.2
Ability to produce tense marking (% correct). Subjects were given items like
" Every day I walk to school. Just like every day, yesterday I
." This task
requiresthe subject to recognizethat the temporal context specifiedin the second
sentencerequires a particular verb form .
Language-impaired
subjects
Controls
English
(England)
English
(Canada
)
French
(Canada
)
Japanese
Greek
38.3
91.7
52.3
93.5
46.7
96.4
48.1
97.9
20.0
87.1
tional morphology.2 They have fewer problems with a pointing task that
requires them to auditorily distinguish between words like " book" and
" books" or " books" and " cooks." These general diagnostic tests have
beensupplementedby much more detailed testsdesignedto test particular
hypotheses. For example, we have data about tense from spontaneous
speech, grammaticality judgment tasks, grammaticality rating tasks,
storytelling, and tenseproduction; and theseresults are consistentover a
wide range of types of stimuli and responses
- oral, aural, written, pointing, and so on (Gopnik 1994). We also have results that show that these
speakershave problems with plurals and comparatives.
One question was whether theseproblems with tensein English could
be accountedfor by the fact that regular past tensein English is encoded
by a form that has been described as " phonologically vulnerable"
(Fletcher 1990) or by somespecialand particular problems with our first
set of English-speaking subjects, who all came from one particularly
interesting family . In order to resolvethis question, we decidedto look at
more English-speaking subjects and at subjects whose native language
was not English. These data show that the problem with tenseis unique
neither to our first subjectsnor to English. It showsup in every population we have looked at (seetable 11.2; seealso Fukuda 1994; Dalalakis
1996; Gopnik et al. 1996; Royle 1996; Gopnik 1998; Ullman and Gopnik ,
to appear) . All of thesepopulations have similar problems with producing correct tense forms, no matter how tense is encoded in their native
language- in a final stressedsyllable as in French or in a three-syllable
element like " mashita" in Japanese
. Interestingly, both the Greek control and language-impaired subjects have lower scores than the other
subjects; this happens becauseeach verb in Greek has more than 60
272
Table
Gopnik
11 .3
Plural
Nominative
Iik
os
Iik
Genitive
Iik
Iik
on
Accusative
Iik
Iik
us
Vocative
Iik e (!)
Iik i
different inflected forms, as opposed to 4 in English, so there are simply many more ways to go wrong and it is harder to be right just by
chance.
The next question that we had to addresswas whether the subjects'
problem with language was confined to tense, as has been suggested
(Rice, Wexler, and Cleave 1995), or whether they made similar errors in
other parts of the inflectional system. Tests show that across the four
languagesthey make errors in number, case, and gender.3
All of this led us to think that these individuals had a particular
problem with inflectional rules, but Jenny Dalalakis's innovative work on
compounds and diminutives in Greek made it clear that the languageimpaired subjectshave just as much difficulty with finding the root of a
complex word as they have with adding an inflection (Dalalakis 1996). It
is not merely that they cannot add an inflection to a root; they do not
even recognize that there are roots and inflections. There is no way to
check this in English becausethere are words in English like ' 'walk" and
" book" that do not have any overt inflectional marking. In Greek, however, all nouns and verbs are inflected. For example, the word for " wolf "
has eight different forms (seetable 11.3). It would seemto be child's play
to figure out that " lik " is the root and all of the rest are inflectional forms.
But not for these children. They make serious errors in tasks like compounding and diminutivization that depend upon the manipulation of
roots. They seemto know that they are supposedto take part of the word,
but they are not sensitiveto the boundariesbetweenroots and inflections
so sometimesthey take a part that is shorter than the root and sometimes
they take a part that is longer than the root . The rule for compounding in
Greek requires that the two roots be conjoined (la ), in some caseswith
the insertion of an " 0" that marks the morphemeboundary (lb ); then the
inflection is added to the end of the conjoined roots.
273
-anthrop
man
-os
masc.sing.
nom
b . lik
wolf
- t
titan thropos
werewolf
-+
likofola
-0
- fol
-a
compound
morpheme
nest
fem .sing.
nom .
wolfnest
(A similar rule holds for forming diminutives . For the details , seeDalalakis
1996.) The subjects were shown pictures and told (in Greek ), for example, " This is a mouse (pondikos ) and this is a man (anthropos ) that becomes a mouse. We would call him a 'mouseman ' (pondikanthropos ) ."
They were asked to form similar compounds in response to other pictures .
The language -impaired subjects had significantly more difficulty with this
task than did the controls (see table 11.4) . Sometimes the language impaired subjects produced fonDs in which the first root was shorter
than required (2a), and sometimes they produced forms in which it was
longer (2b).
(2) a. anthrop
man
-fagh
eater
-os
masc.sing.
nom
b. lot
* anthrofaghos
man -eater
-+
* loto ~faghos
-fagh
-os
eater
masc .smg .
perSImmon
-+
nom
persImmon
- eater
at all .
Table
11 .4
(% correct )
Language impaired
sub,jects
Compounding
16
Diminutivization
38
Young
controls
95
100
Age-matched
controls
99
100
274
Gopnik
Evidence
forImpaired
Grammars
275
Table 11.5
Ability to mark novel words grammatically (% correct). In each of these tests
the subjects were given a context, usually in pictures, which required that a
grammatical rule be applied to a novel word: " This pencil is weff. This pencil is
even
."
Controls
Language-impaired
subjects
95.4
93.5
92.6
87.1
38.0
52.3
33.3
20.0
Japanese
Plurals
English (in England)
English (in Canada)
Greek
89.1
37.0
95.7
99.2
79.8
57.0
58~3
42.1
Comparatives
English (in England)
74.0
21.0
Compounds
Greek
93.6
12.8
80.5
20.2
83.9
40.2
Past tense
Japanese
Diminutives
Greek
these speakers produce forms that look " inflected " by very different
means than the normal grammar .
So far , all of our hypotheses about the way that the language-impaired
subjects must be representing words in their grammar have been inferred
from their performance on linguistic tests. But there is another , more direct
way to look at their mental lexicon . Eva Kehayia conducted a series of
on-line psycholinguistic tests to probe the way in which languageimpaired subjects were processing complex words . It might be the case
that these subjects have normal representations for the inflected forms ,
but some processing difficulties make these forms unavailable to them in
spontaneous speech or elicited production tasks. Psycholinguistic research
on aphasics shows that though they make errors in spontaneous speech
and in linguistic tests, their mental representations appear to be intact
( Kehayia and Jarema 1994; Kehayia , in press) . The language-impaired
276
Gopnik
278
Table
Gopnik
11 .7
Language-impaired
subjects(Greek)
Controls (Greek)
817
870
N onword compounds
kigofaghos
(" kigoeater " ; like
English " wugeater " )
Navel compounds
migofaghos
(' 'flyeater " )
Existing compounds
hortofaghos
(" vegetable eater" ==
" vegetarian " )
(3) a. V [+ past]
walked
" move on foot" (+ past)
b. V
walked
" move on foot in the past"
111111111I1
1111111111
The results of the experimentsin this study clearly show that the performance pattern of the language-impaired subjects is significantly different from that of the control subjects on word and nonword targets
for inflection, derivation, and compounding. The controls in this study
behavejust as reported elsewherein the literature (Taft and Forster 1975;
Laudanna, Badecker, and Caramazza 1989; Kehayia and Jarema 1994).
They processproductively formed complex words by meansof decomposition, and they processsimple and idiosyncratic complex words by means
of whole-word access
. The language-impaired subjectsappear to process
all of the items by means of whole-word access(Kehayia 1994, in press;
Dalalakis 1996). The results of these on-line studies are consistent with
what the off-line linguistic tests have shown: the language-impaired subjects do not build complex representationsfor words; instead, they treat
both complex and simple words in the same way- as if they were
unanalyzedchunks.
The question then is, " What do their representationslook like?" There
are two possibilities for representingwords with no internal structure.
Evidence
for ImpairedGrammars
279
The difference betweentheserepresentationsis that in (3a) the information that this word refers to something that happened in the past is
expressedby a morphological feature (+ past) and in (3b) the concept of
pastnessis expressedas part of the meaning of the word.
As I have argued, the data from a wide range of grammaticality judgment tasks, production tasks, novel-form tasks, and reaction time experiments indicate that these subjects show no sensitivity to the internal
structure of words. These data strongly suggestthere is no evidencethat
the lexical entries of language-impaired speakers contain sublexical
(morphological) featuresfor tense, number, and perhapsother inflectional
properties. Sincethere is no evidencefor such featuresfrom the linguistic
behavior of thesespeakers, we have no grounds for introducing such features into their representations. Given that regular pastsend in a coronal
stop in English, and given that language-impaired individuals can sometimes add a pastlike ending to novel words, we assumethat these individuals store such forms in family resemblanceclassesand that they can
thus perform " wug" -type tasksthrough analogy (Gopnik and Goad 1997).
We therefore have a broad range of crosslinguisticevidence, both on-line
and off-line, that all tells the samestory: that the language-impaired subjects, though they get the same linguistic input as their siblings, do not
build the samesort of grammar. The grammarsthat they do build appear
to rely on memory and analogy and not productive rules for handling the
morphological aspectsof language. The spontaneousspeechof the older
subjects, and even their performance on some linguistic tests, demonstrates that a reasonablesimulacrum of morphology can be constructed
by using memory and analogy instead of productive rules. They do not
perform perfectly, but they are not terrible either. It is only when the
subjectsare given teststhat dependon productivity or when their internal
processesare monitored that the full extent of their inabilities can be
detected.
All of this would be interesting evenif it only provided a description of
the difficulties that a small group of language-impaired subjectshave. But
if it is all true, and if this disorder is related to somegeneticfactors that in
turn have consequencesfor neurological development, then apparently
there are some genetic and neurological precursorsthat are required for
the normal courseof theseaspectsof languagedevelopmentto take place.
This does not mean that we think that there is a gene for morphology!
From the point of view of linguistics, there is good evidencethat though
morphology is one of the striking aspectsof languagethat seemsto stump
280
Gopnik
these individuals
The question
ficulties
are directly
causes their
problems
their prosodic
in morphology
are an independent
or whether
are . And
there is abundant
not follow
of neurologi -
this disorder
language
would
dif -
system that
these other
and syntactic
problems
with .
necessarily
linguistic
in the normal
with
system to develop ;
case might
not be
the same as those that can cause the system to break down . Of course ,
there is much
intellectual
more
to find
out . Weare
only
at the beginning
of this
journey , but it does appear that it will turn out that Darwin
nature of humans .
Notes
1 . This family has been the subject of much debate ( Fletcher 1990; Vargha Khadem et al . 1995) . For a d'etailed discussion of some of these disputes , see
Gopnik and Goad 1997. From the data , it is clear that the language impair ment seen in this family is not caused by oral apraxia or low performance IQ .
Moreover , as I will show , the same pattern of impairment is found in English speaking subjects in Canada as well as in French -, Greek -, and Japanese-speaking
subjects.
2. Since Japanese does not have a rich , productive system of derivational
phology , we did not test the Japanese-speaking subjects on this task .
mor -
3. They also make errors in syntax and in some phonological processes such as
prosody , but discussing these in detail would take us too far afield here (van der
Lely and Harris 1990; van der Lely 1997; Clahsen 1992; Piggott and Kessler 1994;
Rice , Wexler , and Cleave 1995; Goad , in press) .
References
Bishop , D . V ., North , T ., and Donlan , C . ( 1995) . Genetic basis of specific
language impairment : Evidence from a twin study . Developmental Medicine and
Child Neurology , 37, 56- 71.
Clahsen , H . ( 1992) . Linguistic perspectives on specific language impairment
( Theorie des Lexikons Arbeitspapier 37) . Dusseldorf : Heinrich Heine Universitat .
Dalalakis , J. ( 1994). Familial language impairment
Papers in Linguistics , 10, 216- 227.
in Greek . McGill
Working
281
282
Gopnik
Kuhl
effect
"
for
1991
Kuhl
. ,
of
Palmour
. ,
10
and
. ,
Plante
16
with
. ,
Rice
. ,
850
Royle
thesis
Shaer
863
1996
1996
. ,
Verb
ni
. ,
. ,
language
Mahwah
Tomblin
Gopnik
NJ
in
der
SLI
boy
Perception
&
Ed
28
. )
Press
531
specific
learning
in
the
inheritance
and
1989
546
and
The
Priming
homographic
linguistic
impairment
Journal
of
Compensatory
strategies
Neurolinguistics
1994
in
10
173
Prosodic
genetic
185
dyspha
constraints
McGill
in
Working
Papers
familial
in
Linguis
. ,
and
Rapcsak
Brain
Cleave
and
1991
Language
1995
Specific
Journal
41
MRI
findings
52
66
language
of
Speech
in
boys
impairment
and
as
Hearing
Research
in
of
tense
1996
Tense
subjects
Unpublished
Master
' s
In
of
and
and
14
linking
638
of
647
54
287
Ed
. )
prefixed
language
295
impair
to
Toward
contributions
Rice
theory
retrieval
developmental
environmental
storage
Behavior
Disorders
and
refel
Lexical
Verbal
Hearing
Genetic
time
University
concentration
and
and
Familial
DLI
McGill
1975
Learning
French
the
genetics
risk
of
for
language
1997
The
Epidemiology
of
inheritance
and
specific
language
innateness
of
impairment
grammars
New
York
In
Oxford
. ,
and
Buckwalter
twins
Journal
of
1998
Speech
Heritability
Language
of
&
poor
Hearing
language
Research
41
phology
van
magnet
among
Ullman
sense
Erlbaum
. )
achievement
199
1997
production
ty
Speech
Press
infinitive
ersi
1989
Ed
report
impairment
University
Tomblin
perceptual
not
nativism
Gopnik
University
of
and
Verbal
of
specific
Vance
Forster
Journal
Tomblin
"
do
Oxford
preliminary
Making
of
Evolution
in
dissertation
and
Journal
Tomblin
ment
optional
In
Caramazza
Robb
doctoral
words
and
impairment
Wexler
Unpublished
Taft
monkeys
Journal
Kessler
McGill
show
extended
1997
Language
230
Swisher
of
language
infants
specific
period
studies
215
and
York
. ,
and
23
Genetic
human
categories
speech
mory
impairment
lO
New
Gopnik
and
memory
Piggott
188
1997
tics
and
speech
Badecker
Declarative
107
. ,
is
Meltzoff
ofM
language
38
lingu
Paradis
90
language
Journal
Neuro
adults
grammars
stems
Human
of
of
innateness
tics
50
and
development
Laudanna
prototypes
Psychophysics
sia
the
. ,
Lely
and
Gopnik
hereditary
Modularity
specific
1997
and
to
appear
language
Language
innateness
The
impairment
and
cognitive
Journal
production
of
Applied
Ps
inflectional
development
oj
Neurolinguistics
mor
' cholinguistics
in
10
grammatical
75
107
20
283
, 92 , 930 - 933 .