CROSSLINGUISTIC RESEARCH IN APHASIA:

AN OVERVIEW
Elizabeth Bates
University of California at San Diego
Beverly Wulfeck
University of California at San Diego
and
Language Research Center, Childrens Hospital Research Center
Brian MacWhinney
Carnegie-Mellon University
(Brain and Language, 41, 123-148, 1991)
Most of us would like to believe that the different
patterns of language breakdown observed in aphasic
patients reflect the way that the human mind and brain
are organized for language. However, because so much
modern research on aphasia has been carried out in
English, it is difficult to separate universal
mechanisms from language-specific content.
Crosslinguistic com-parisons permit us to disentangle
these confounds, while we address one of the most
important issues in cognitive neurobiology, the issue of
behavioral and neural plasticity: How many
different forms can the language processor take under a
range of normal and abnormal conditions? We must
have an answer to this question if we want to understand
what the neural mechanisms responsible for language
really are and really do.
The nine papers presented within this special cross-
linguistic issue of Brain and Language provide
important new information about universal and
language-specific patterns of sparing and impairment, in
nonfluent "agrammatic" Broca's aphasics and in fluent
patients with a diagnosis of Wernicke's aphasia. These
crosslinguistic studies fall into two categories: (1)
research in which language type is treated as an
independent variable, by conducting the same experi-
ment with equivalent materials in two or more different
languages, and (2) research in which language type is
treated as a natural experiment, using the peculiar
characteristics of a single language to answer a question
that would be difficult to ask in (for example) English.
Studies applying one or both of these cross-linguistic
methods have yielded six basic findings, summarized
briefly as follows.
(1) Crosslinguistic variation: First, the papers
in this issue (and related crosslinguistic studies by these
investigators and other research groups see Menn and
Obler, 1990) clearly demonstrate that the "same"
aphasic syndromes look very different from one
language to another. Indeed, language differences
account for more variance than patient group differences
in many of our crosslinguistic experiments to date (e.g.,
Bates, Friederici & Wulfeck, 1987a & b; 1988; Bates,
Friederici, Wulfeck & Juarez, 1988; Wulfeck, Bates,
Juarez, Opie, Friederici, MacWhinney & Zurif, 1989;
Vaid & Pandit, this issue).
(2) Performance deficits: The existence, strength
and nature of the crosslinguistic differences uncovered in
these studies lead to the conclusion that language-
specific knowledge (i.e. competence) is largely preserved
in Broca's and Wernicke's aphasia, requiring an account
of language breakdown based on deficits in the processes
by which this preserved knowledge base is accessed and
deployed (i.e. performance). This conclusion has led, in
turn, to an expanded use of "on-line" or "real-time"
experimental procedures that yield information about
how patients from different language groups arrive at a
correct or incorrect response in receptive and expressive
language use (see especially Wulfeck, Bates & Capasso,
this issue; Friederici & Kilborn, 1989).
(3) Selective vulnerability of morphology:
Overlaid on these language differences, we find some
evidence for a modified version of the Closed-Class
theory of agrammatism, i.e. the idea that grammatical
inflections and function words can be selectively
impaired in aphasia. In these and other papers by the
same research team, we have found evidence for closed-
class impairments in production, comprehension and
error detection although the degree and nature of
those impairments vary greatly from one language to
another. In addition, these crosslinguistic studies have
also helped to distinguish between those aspects of
morphology that are "at risk" (e.g., case contrasts that
are irregular and/or relatively difficult to perceive) and
those that appear to be "protected" (e.g., case contrasts
that are regular and/or relatively easy to perceive)
within and across language types (see especially
Friederici, Weissenborn & Kail, this issue; Mac-
Overview
2
Whinney, Osmn-Sgi & Slobin, this issue).
(4) Patient group similarities: The selective
vulnerability of morphology described above is
apparently not restricted to agrammatic Broca's aphasics.
We have observed equivalent morphological deficits in
the expressive language of fluent Wernicke's aphasics;
receptive deficits appear in an even wider range of
patient groups, including some patients who are
neurologically intact (see especially Bates et al., 1987a;
MacWhinney, Osmn-Sgi & Slobin, this issue). This
suggests that closed-class items might be vulnerable to
global forms of stress that are only indirectly related to
the effects of focal brain injury (e.g., perceptual
degradation; cognitive overload). Such findings point
to the need for experiments that control for the
contribution of a global reduction in perceptual and/or
cognitive resources, in order to isolate those forms of
grammatical impairment that are specific to particular
types of aphasia from those that can be induced in
normals under stressed or nonoptimal processing
conditions (e.g., Kilborn, this issue).
(5) Similarity of lexical & grammatical
symptoms: Although morphology appears to be a
quantitatively vulnerable domain, the grammatical
symptoms displayed by these patients are qualitatively
similar to their lexical (i.e. word-finding) symptoms
(e.g., similar effects of frequency, complexity, semantic
relatedness). These similarities are compatible with
models in which lexical and grammatical forms are
represented in a common format and/or accessed by a
common set of processing mechanisms. This
interpretation is still controversial, but it could be tested
through detailed comparisons of lexical and grammatical
processing, in languages that contrast markedly in the
degree to which they rely on word order, inflections
and/or lexical contrasts to accomplish the same
communicative goals (see, for example, papers by
Bates, Chen, Tzeng, Li & Opie and by Chen & Hung,
this issue).
(6) Patient group differences: Although there are
indeed more similarities than differences in the patterns
of sparing and impairment observed in Broca's and
Wernicke's aphasics, we have uncovered a set of
contrasts that hold up across very different language
types: differential success in the production of nouns
and verbs (Bates, Chen, Tzeng, Li & Opie, this issue),
differences in the ability to exploit both grammatical
and lexical redundancy (Bates et al., 1987a & b), and
differences in the nature of morpheme substitution
errors (Bates, Friederici & Wulfeck, 1988). The papers
presented in this special issue were designed to explore
these proposed "neurolinguistic universals" in greater
detail, bringing us one step closer to a model of
intrahemispheric organization that can handle universal
and language-specific differences between syndromes.
In the next few pages, we will provide a brief
summary of crosslinguistic evidence in support of these
six conclusions, including a discussion of two
competing theories that have guided all this work (i.e.
the Competition Model, and the Closed-Class Theory of
Agrammatism). Then we will end with a discussion of
a serious methodological problem that confronts all
researchers interested in applying the crosslinguistic
method: the problem of patient selection across
languages with radically different structural and
statistical properties. Since this is a problem that can
only be resolved by much more crosslinguistic research,
we hope that this discussion will inspire other
investigators to add to a growing body of comparative
evidence on fluent and nonfluent aphasia across natural
languages.
I. OVERVIEW OF CROSSLINGUISTIC
EVIDENCE
Our own crosslinguistic aphasia project is built
upon the theoretical, empirical, methodological and
organizational foundations provided by 18 years of
crosslinguistic research on normal adults and children by
Bates and MacWhinney (Bates & MacWhinney, 1987;
MacWhinney & Bates, 1989). The theory of language
processing that has emerged from that work, i.e. the
Competition Model, provides an interactive
activation account of the quantitative and qualitative
variations in sentence processing that are observed
across languages, in normal speakers and aphasic
patients. This model can be contrasted with the
Closed-Class Theory of Agrammatism
(Bradley, Garrett & Zurif, 1980; Friederici, 1986; Kean,
1979 & 1985), a modular account of grammatic
impairment in aphasia that is based in part on Garrett's
multilevel theory of sentence production and sentence
comprehension in normal adults (e.g., Garrett, 1980).
After several years of collaborative research by
investigators in both "camps", we are now close to a
fruitful compromise between these contrasting
approaches to grammatical processing in normals and
aphasics, integrating the relative contributions of
linguistic experience (reflected in the cross-
language contrasts observed in patients from the same
diagnostic category) and neural specialization
(reflected in patient group differences that hold up across
language types).
Taken in their strongest and most interesting form,
modular or disconnection theories of aphasia should
predict broad differences between patient groups, with
relatively little differentiation as a function of language
type. Simply put, if a patient has lost the grammatical
component that handles most if not all of the
significant structural facts that define a natural language,
then the same patient should (presumably) lose most of
the performance characteristics that define a native
speaker of that language. For example, the Closed-
Class theory of agrammatism predicts a selective
Overview
3
impairment of grammatical inflections and function
words in Broca's aphasics, in comprehension and
production. Under a strong interpretation of this model
(i.e. a disconnection view), it is assumed that Broca's
area plays a special role in grammatical representation
and/or processing. Hence we should expect Broca's
aphasia to result in a reduction or indeed a complete loss
of crosslinguistic differences in the use of closed-class
items; such differences should be largely preserved in
Wernicke's aphasia, because the neural regions
responsible for grammar are assumed to be intact in
these patients.
The Competition Model predicts fewer differences
between aphasic syndromes, but more differentiation as
a function of language type. It provides two basic
principles that predict crosslinguistic differences in
the linguistic performance of patients from the "same"
clinical category, and within-language similarities in
the performance of patients with different forms of focal
brain injury. Cue validity refers to the
information value of a given phonological, lexical,
morphological or syntactic form within a particular
language (e.g., the availability and reliability of a
particular word order type as a cue to semantic roles like
agent, action and object). Cue cost refers to the
amount and type of processing associated with
the activation and deployment of a given linguistic
form, when cue validity is held constant (e.g., the
amount of memory required to store and compare
agreement cues across the course of a sentence; the
degree of perceptual difficulty posed by different types of
case markers). These two principles (which can be
quantified with some precision) co-determine the nature
of linguistic representations in a particular language,
and the nature of the dynamic process by which form
and meaning are activated and mapped onto each other in
real time. In this model, linguistic information is
represented as a broadly distributed network of
probabilistic connections among linguistic forms and
the meanings they typically express (see also Hinton &
Shallice, 1989; Seidenberg, McClelland & Patterson,
1987; Rumelhart & McClelland, 1986). Linguistic
rules are treated as form-meaning and form-form
mappings that can vary in strength, so that the "same"
rule may be stronger in one language than it is in
another, as a function of crosslinguistic differences in
the relative cue validity of equivalent linguistic forms
(e.g., basic word order is "stronger" in English than it is
in Italian; subject-verb agreement is "stronger" in Italian
than it is in English). Different kinds of linguistic
information (phonological, lexical, morphological,
syntactic) are represented together in a common format,
and the processes of mapping meaning onto form (in
production), form onto meaning (in comprehension) and
the process of evaluating the internal compatibility of
two or more forms (e.g., grammaticality judgment) all
involve graded activation (excitation and inhibition).
Decisions about what to say or how to interpret the
input emerge through a quantitative process of
competition and conflict resolution within this broadly
distributed and richly interconnected knowledge base.
Applied to language processing in aphasia, the principle
of cue validity makes the following prediction: The
selective impairment of cues i n
comprehension, and the relative accessi bi l i ty
of forms in production, will reflect
quantitative differences in the strength or
probability of form-function and form-form
mappings in the premorbid language of the
patient. Simply put, this means that it should be
very hard to eradicate Italian from Italians, Turkish from
Turks, and so on, because the essential characteristics of
one's native language are broadly represented and deeply
engrained; they do not exist in an isolated "box" that
can be selectively dissociated in aphasia.
The principle of cue cost mitigates and
complements the predictions of cue validity. Because
linguistic items (words, morphemes, phrase structure
frames) can vary in their accessibility (i.e.
perceivability, ease of articulation, degree of
confusability with other items see Bates & Wulfeck,
1989a), a reduction in processing resources along any of
the relevant dimensions (i.e. perception, attention,
memory) may have selective effects on the
speaker/listener's ability to use particular items in real
time. Hence items that are equivalent in information
value (cue validity) may be differentially spared or
impaired, depending on the amount and type of
processing they require (cue cost). The implications for
aphasia are the following: Classes of l i ngui st i c
information that are high in cue cost will be
selectively impaired in all forms of aphasia;
the same pattern of selective sparing and
impairment may result from different forms
of brain damage, and/or from gl obal
processing limitations in subjects who are
neurologically intact. That is, hard things should
be hard for everyone within a particular language;
differences are a matter of degree. As currently
formulated, the Competition Model contains no
principles that would predict a systematic qualitative
difference between patient groups (e.g., Broca's versus
Wernicke's aphasia). To the extent that we find
systematic patient group differences in the deficits
observed within and across language types, the cue cost
principles of the Competition Model must be modified.
Presumably, such modifications would involve
postulating specific rather than general forms of cue
cost, associated with damage to specific brain regions.
A number of processing accounts of differential
language breakdown that are compatible with the basic
architecture of the Competition Model have emerged in
the last few years, in response to findings that contradict
disconnection theories (e.g., sensory vs. motor aphasia;
grammar vs. semantics). Some of these new accounts
include: (1) the suggestion that anterior and posterior
Overview
4
lesions differ in their effect on processing speed (i.e.
anterior lesions have a greater effect on rapid processes
Friederici & Kilborn, 1989; Swinney, Zurif &
Nicol, 1989) and/or types of working memory (e.g.,
anterior lesions result in a "degraded trace" Ostrin &
Schwartz, 1986); (2) the suggestion that anterior lesions
have a selective effect on automatic aspects of
language processing, while posterior lesions have a
greater impact on controlled processing (Milberg &
Blumstein, 1981); (3) Posner's arguments for a
differential anterior-posterior distribution in the basic
components of attention (with differential effects on
those aspects of language processing that are most
dependent on anterior vs. posterior aspects of attention
Posner, Petersen, Fox & Raichle, 1988); (4) the idea
that anterior lesions result in a selectively greater
reduction of excitation while posterior lesions tend to
reduce inhibition an old proposal (e.g., Goldstein,
1948) that has taken on new meaning in an era of neural
network models, combined with an increased
understanding of neural transmitters, their differential
distribution in the brain, and their differential
consequences for computation. Any of these proposed
contrasts might result in qualitative differences in the
performance deficits displayed by Broca's and Wernicke's
aphasics (as defined below), without contradicting the
assumption that linguistic knowledge (competence) is
broadly distributed in the brain and largely preserved in
patients with focal brain injury (see also Hinton &
Shallice, 1989; Linebarger, Schwartz & Saffran, 1983;
Seidenberg et al., 1987). The field of aphasiology
appears to be moving toward a new theory of
intrahemispheric organization. The languages, patient
groups and experimental procedures represented in this
special issue contribute to this effort.
In our crosslinguistic studies across the last 8-10
years, the Competition Model and the Closed-Class
theory of agrammatism have been tested against basic
comprehension and production data for Broca's aphasics,
Wernicke's aphasics and a range of other patient groups,
in Indo-European languages (English, Italian, German,
Serbo-Croatian & Spanish) that vary in the relative cue
validity of word order and grammatical morphology.
More recently, these results have been replicated and
extended in three directions:
a larger array of language types (including
Hungarian, Chinese, Turkish, Hindi & Kannada), from
four different language families (Finno-Ugric, Sino-
Tibetan, Ural-Altaic, Indo-European);
a wider array of experimental techniques (e.g., real-
time studies of grammaticality judgment Wulfeck,
Bates & Capasso, this issue);
a range of new and interesting control populations
that help us to sort out global and specific cue cost
factors (e.g., congenitally deaf individuals who are
neurologically intact Volterra & Bates, 1989;
normals under conditions of perceptual degradation or
cognitive overload Kilborn, this issue).
Given the array of contrasts studied across the last
decade, we are now confident that our six basic
conclusions about the nature of language breakdown are
correct. These include (1) strong evidence that language-
specific lexical and grammatical knowledge is preserved
in aphasia, interacting with (2) evidence for a selective
"softening" of the patients' ability to make use of this
knowledge, a processing deficit that is (3) most evident
in the comprehension and/or production of grammatical
inflections and function words. The first three findings
are entirely compatible with the Competition Model,
although they are also compatible with a probabilistic
variant of the Closed-Class theory. In the same period
we have also obtained important new information about
the nature and extent of these processing deficits,
leading to a compromise view of intrahemispheric
organization for language. (4) The vulnerability of
morphology is not restricted to Broca's aphasia (against
the Closed-Class theory). (5) There are qualitative
similarities between the morphological and lexical
symptoms displayed by aphasic patients (against the
Closed-Class theory). (6) There are subtle processing
differences between Broca's and Wernicke's aphasia that
hold up across language types: in noun vs. verb
production; in the effects of cue convergence; in the
nature of morpheme substitution errors (i.e. more use of
high-frequency or unmarked forms in Broca's aphasia;
more low-frequency or highly marked substitutions in
Wernicke's aphasia). These differences are not
compatible with the Competition Model in its original
form, nor are they compatible with either of the major
disconnection theories (sensory vs. motor; grammar vs.
semantics); however, they are compatible with one or
more of the processing accounts described above.
To provide background for the nine papers reported
in this issue, let us briefly review crosslinguistic
evidence in support of these six conclusions.
(1) Cross-language contrasts
The major rationale for crosslinguistic research is
the search for quantitative and qualitative variations in
the symptom patterns displayed by fluent and nonfluent
aphasic patients. We have indeed found robust evidence
for cross-language variation in both patient groups, in
production, comprehension and grammaticality
judgment.
Sentence product i on. Three published
studies of sentence production in English, Italian and
German patients provide background for new studies of
sentence production using the same method in Turkish
(Slobin, this issue), Hungarian (MacWhinney &
Osmn-Sgi, this issue), and Chinese (Tzeng, Chen &
Hung, this issue). All these studies are based on a
picture-description situation called the Given-New Task
(Bates, Hamby & Zurif, 1983; MacWhinney & Bates,
Overview
5
1978), which involves a series of three-picture cartoons
in which one element varies while the remaining
elements remain constant (e.g., a little girl is pictured
eating an apple, then an ice cream, then a cookie). The
picture triplets are designed to elicit intransitive,
transitive, dative and locative structures, in a situation
that restricts the range of semantic and pragmatic targets
the patient might have in mind. In several of these
studies, the Given-New data have been supplemented by
free-speech results from a biographical interview. Both
data sets have been transcribed and coded in the format
specified by the Child Language Data Exchange System
(CHILDES, MacWhinney & Snow, 1985), modified to
handle the special problems posed by adult aphasia data.
We have used these data as the cornerstone for a new
system called ALDES (Aphasic Language Data
Exchange System, announced in Bates & Wulfeck,
1989b), available to qualified researchers around the
world (with proper controls to insure patient
confidentiality).
Pragmatic effects on lexical and
grammatical f orm. Building on an earlier study
with English patients (Bates, Hamby & Zurif, 1983),
we have examined the effect of the given-new contrast
on several aspects of linguistic expression in English,
Italian and German patients (Wulfeck et al., 1989).
Briefly stated, results suggest that Broca's and
Wernicke's aphasics both retain sensitivity to the given-
new contrast: in the decision about which elements in
the picture to lexicalize (new) or omit (old), in the use
of indefinite (new) vs. definite (old) articles, in the use
of pronouns (which tend to be used to express old
information, when they are used at all). Combining
given-new data with biographical interviews, we have
also shown that German and Italian patients are
sensitive to an important crosslinguistic contrast called
"the null subject parameter" (Rizzi, 1980). That is,
German patients appear to know that subjects are
obligatory in free-standing declarative sentences
(producing subject pronouns most of the time even
when the identity of the subject can be taken for
granted); Italian patients appear to know that subjects
can be omitted from free-standing declarative sentences
in their language (when the identity of the subject is
obvious from the context). These cross-language
differences in subject omission are very large,
transcending the smaller difference in subject omission
that characterizes Broca's vs. Wernicke's aphasics within
each language.
Word order. The first study of word order that
we completed within our own crosslinguistic project
(Bates, Friederici, Wulfeck & Juarez, 1988) focussed on
the order of basic sentence constituents in English,
Italian and German patients, showing that canonical
sentence order (Subject-Verb-Object, or SVO) appears to
be preserved in both Broca's and Wernicke's aphasics.
Indeed, some patients (particularly Broca's) appear to
overuse basic SVO, as though this word order type
provided a kind of "safe harbor" for sentence planning.
Such overuse is only evident in languages that permit
pragmatic word order variation; it could not be detected
in a rigid word order language like English.
Noncanonical word order patterns may be slightly
impaired in languages that permit such options, but the
degree to which this is true seems to depend on the
frequency and utility of each word order variant. For
example, Italian patients (including Broca's aphasics) do
produce a number of pragmatically appropriate subject-
final constructions (the most frequent noncanonical
word order type in the language); a less frequent form of
word order variation (with the object placed before the
verb) appears in the speech of Wernicke's and normal
controls, but seems to be avoided by Broca's aphasics.
We conclude that canonical word order is preserved in
aphasia, but there are variations in the "accessibility" of
noncanonical word order options that vary by language
and patient group.
Subsequent studies in other languages have
replicated and extended this word order finding. The
basic SOV word order of Turkish is clearly preserved in
fluent and nonfluent aphasics (Slobin, this issue), and
indeed may be overused by aphasic patients (especially
Broca's) when these data are compared with the word
order variations produced by normal controls. Further
information comes from Hungarian (MacWhinney &
Osmn-Sgi, this issue). Although it is sometimes
argued that Hungarian is an SOV language, the
situation is actually more complex. SOV order is used
when the object of the verb is indefinite, while SVO
order is preferred when the object is definite. Object
definiteness is marked on the verb itself, creating an
interesting interplay of pragmatic, morphological and
syntactic factors. Hungarian patients produced a high
proportion of both SOV and SVO in their picture
descriptions, suggesting that canonical word order is
preserved in aphasia even when there are two canonical
options available. Furthermore, patients appear to
retain the ability to coordinate word order and
definiteness, in accordance with the rules of Hungarian
(although there are some interesting strategies that
patients adopt to "avoid" this situation, making use of
legal options including article omission see
morphological production, below). Finally, SVO
biases are also evident in the speech of our Chinese
patients, although many patients (including Broca's)
attempt to produce one or more of the legal word order
alternatives in Chinese, struggling to provide the topic
markers that are required for these alternative word order
types (Chen, 1989; Chen, Bates & Tzeng, 1990; Tzeng,
Bates & Wong, 1990; see discussion in Bates, Chen,
Tzeng, Li & Opie, this issue).
Grammatical morphol ogy. We find
consistent evidence that grammatical morphemes (bound
and free) are selectively vulnerable in aphasia (see
below), but this finding must be interpreted together
with overwhelming evidence that Broca's and Wernicke's
Overview
6
aphasics retain detailed and specific knowledge of the
system of grammatical morphology in their language.
For example, in our study of English, German and
Italian patients (Bates, Friederici &Wulfeck, 1988) we
found a significant main effect of language in the
proportion of function words to total words produced
(i.e. higher proportions of function word use in German
and Italian, lower proportions in English); this cross-
linguistic pattern reached significance when Broca's,
Wernicke's and normal controls were analyzed separately
or together. We also found a large and consistent cross-
language difference in the production of definite and
indefinite articles. There are only three forms of the
article in English, compared with nine forms marked for
gender and number in Italian, and a much larger array of
options marked for gender, number and case in German.
On grounds of relative difficulty, we should expect
article omission rates to be highest in German patients
and lowest in English patients. However, on grounds
of cue validity (i.e. the amount of information encoded
in the article), we should expect the opposite finding:
German < Italian < English on a measure of article
omission. Results strongly favor the cue validity
prediction: Even among the nonfluent Broca's aphasics,
article omission averaged less than 15% in German and
25% in Italian, compared with an average rate of 70%
omission in English. Patients did occasionally produce
the wrong form of the article, but these error rates were
quite low (e.g., 7% in Italian and 16% in German, with
similar error rates in both patient groups see below);
this means that patients produce the correct form in the
overwhelming majority of cases, far more than we
would expect if morphemes were generated randomly (as
predicted by Grodzinsky, 1986).
Further evidence for the preservation of
morphological knowledge (as opposed to performance
see below) is provided in our studies of the
production of case inflections in Turkish (Slobin, this
issue) and Hungarian (MacWhinney & Osmn-Sgi,
this issue), and in a study of one small but intricate
domain of function word production in Chinese, the
system of nominal classifiers (Tzeng, Chen & Hung,
this issue). Errors do occur, but the probability of an
error is greatly conditioned by cue validity (i.e.
information value) and cue cost (i.e. relative frequency,
salience, pronounceability), and patients are right more
often than they are wrong. These findings are difficult
to reconcile with a disconnection view of closed-class
impairments in aphasia.
We should also point out some systematic findings
regarding the nature of those morphological errors that
do occur in expressive language. First, although
omission errors are more common in Broca's aphasics
compared with Wernicke's, morpheme substitution
errors occur in both forms of aphasia. It is not the case
that Broca's aphasics regularly omit free-standing
function words; rather, the patient's decision to omit an
item or try to find the correct form appears to be a
partial function of the cue validity or importance of that
form in the patient's native language (e.g., the above
data on article omission in English, German and
Italian). In other words, the patterns o f
omission and substitution observed i n
Broca's and Wernicke's aphasia reflect the
patients' detailed knowledge of legal opti ons
and information demands in their language.
Second, certain types of logically possible errors are rare
or nonexistent in our data. There is virtually no
evidence for errors that violate principles governing the
order of bound morphemes within a word, or the order
of function words within a noun phrase. Patients
simply do not make mistakes like "Dog the" or "ing-
walk", even though errors of this kind are logically
possible (particularly in a language like Hungarian,
where a noun can take up to 3-4 suffixes in a required
order). There is also very little evidence in our Turkish
or Hungarian data for errors of vowel harmony, an
aspect of morphology that is conditioned by
phonological factors spanning more than one word.
Although these findings are drawn from sentence
production data, they can be used to guide studies of on-
line grammaticality judgment, to determine the extent
to which sensitivity to errors in receptive processing
mirrors the error patterns that occur in spontaneous
speech (see Wulfeck et al., this issue).
Sentence comprehensi on. Three previous
studies of sentence comprehension provide background
for the papers in this issue: one on Broca's and
Wernicke's aphasics in English, German and Italian
(Bates et al.,1987a), and two examining the interaction
of grammatical cues in Serbo-Croatian Broca's and
anomic aphasics (Smith & Mimica, 1984; Smith &
Bates, 1987). All of these experiments were based on
the same simple off-line procedure (i.e. the procedure
applied by MacWhinney et al., this issue, and by Vaid
& Pandit, this issue; see also Chen et al., 1990, for
Chinese). In this task, patients are asked to interpret
simple sentences by acting them out with small toy
objects (e.g., "Show me THE COW IS KICKING THE
PENCIL"). The sentence stimuli represent competing
and converging combinations of semantic cues (i.e. the
contrast between animate and inanimate objects),
syntactic cues (i.e. standard and nonstandard word orders)
and morphological cues (presence/absence of case
markers on the first or second noun in case-inflected
languages; presence/absence of subject-verb agreement
or object marking on the first or second noun in
languages with agreement markers). This design
permits us to assess the hierarchy of importance of
lexical, morphological and syntactic cues to
agent/object relations in each language, a direct test of
the cue validity predictions of the Competition Model.
(Because the notion "percent correct" is meaningless in
such a competition design, the dependent variable for all
analyses is "percent choice of the first noun as agent").
In the Bates et al. study, Broca's and Wernicke's
Overview
7
aphasics within each language group displayed
language-specific patterns in their use of word order
(English > German > Italian), semantic contrasts
(German > Italian > English) and agreement
morphology (Italian > German > English). Although
use of grammatical morphology in sentence
comprehension appeared to be impaired in all aphasic
groups relative to normal listeners in that language (see
below), patients were still performing above chance in
the use of morphological information (particularly in
richly inflected languages). Use of canonical word order
information showed no deterioration in any language or
patient group; in fact, many of our German and Italian
patients (fluent and nonfluent) used canonical word order
more than normal controls, compensating for reductions
in the use of grammatical morphology a finding that
mirrors our results for sentence production (above). Use
of semantic information is also in the normal range (and
varies from one language to another for aphasic
patients, as it does for normals). In short, the
hierarchy of importance of cues to sentence
meaning from one language to another
reflects differences in cue validity; these
differences are preserved in the data for
fluent and nonfluent aphasics, although the
accessibility of some cues (especi al l y
grammatical morphology) appears to be
reduced.
Grammaticality j udgment . An influential
study by Linebarger et al. (1983) showed that so-called
agrammatic Broca's aphasics can make subtle judgments
of grammaticality. This important finding has now
been replicated in several laboratories, including our
own, and it appears to hold up even when patients are
asked to make their judgments "on line" (Wulfeck,
1987; Wulfeck, 1988; Wulfeck & Bates, 1990; Wulfeck
et al., this issue; Lukatela, Crain & Shankweiler, 1988;
Shankweiler, Crain, Gorrell & Tuller, 1989). In
Wulfeck et al. (this issue), we have explored cross-
linguistic differences in the detection of grammatical
violations as a new domain in which to test the basic
cue validity and cue cost assumptions of the
Competition Model. Italian and English Broca's
aphasics and normal controls were presented with
sentences in which verb auxiliaries and determiners are
manipulated in two ways to produce an ungrammatical
sentence: Either the element is substituted to produce an
incorrect form of agreement (e.g., "The girl is selling
books..." "The girl are selling books..."), or the
same element is moved downstream ("The girl is selling
books..." "The girl selling is books...."). Although
both English and Italian normals show sensitivity to
both types of violations (responding at ceiling), their
decision times are quite different on this task, in
keeping with other evidence on the relative importance
of word order vs. morphology in these two languages:
English normals are particularly quick and efficient at
detecting errors of item ordering, while Italian normals
are relatively slow in detecting order violations but
perform optimally in the detection of errors involving
morphological substitution. This crosslinguistic
difference is preserved in the error and reaction time data
of English and Italian Broca's aphasics, but to a much
smaller degree (in line with our argument that
morphology is selectively vulnerable in aphasia see
below).
(2) Performance deficits and activation
models
The detailed crosslinguistic variations that we just
described constitute ipso facto evidence in favor of the
view that grammatical and lexical knowledge (i.e.
competence) is largely preserved in Broca's and
Wernicke's aphasia. These facts argue in favor of a
performance account, i.e. a theory that attributes
selective patterns of sparing and impairment to deficits
in the processes by which linguistic knowledge is
accessed and deployed. In addition to these facts, we
have found other lines of evidence to support a
particular kind of processing model, an interactive
activation approach in which cues of varying strength
combine to determine the probability of a correct or
incorrect response.
First, this approach is supported by the
systematically graded nature of the deficits we see in
both production and comprehension (i.e. a direct
relationship between probability of error and strength of
a given linguistic form in a particular language).
Second, it is supported by evidence on the way that
cues combine in receptive processing tasks. For
example, Smith and Bates (1987) have examined the
interactions among three different grammatical cues to
sentence meaning in Serbo-Croatian: word order (a
small but significant tendency for normal listeners to
choose the first noun as agent), accusative case marking
(to indicate the object role), and gender agreement (to
indicate the subject role). In morphologically
ambiguous sentences (no case or gender sentences
which do occur in this language), patients showed a
weak tendency to choose the first noun; when this bias
was reinforced by the presence of an agreement cue,
performance did not improve (as though the patients
were unable to detect gender marking); when word order
and case are presented together (with no gender
marking), performance was significantly better but still
far below normal levels. However, when all three
grammatical cues were presented together (ORDER +
CASE + GENDER), Broca's aphasics were able to
interpret sentences at close to normal levels. In other
words, these three cues are combining in a nonlinear
fashion, as though all three were needed to boost
response above some performance threshold a
threshold that has been raised to abnormally high levels
as a result of focal brain injury.
We have similar results on cue convergence in our
study of sentence comprehension in English, Italian and
Overview
8
German aphasics (Bates et al., 1987a), although these
convergence effects are particularly marked in Broca's
aphasics (a point to which we will return below).
These results place previous studies of receptive
agrammatism in a new light. The existence of receptive
agrammatism in patients who are "clinically normal" in
comprehension has usually been explained by
suggesting that these patients use pragmatic and
semantic information to circumvent their grammatical
limitations in a real-life situation. Our studies
suggest instead that Broca's aphasics can use
any form of redundancy (including
grammatical redundancy) to overcome their
performance deficits. This conclusion is
compatible with interactive activation models, but it is
difficult to reconcile with theories that explain
agrammatism by postulating a disconnection between
linguistic modules.
(3) Selective vulnerability of morphology
We have already described evidence suggesting that
grammatical morphology is selectively vulnerable in
aphasia, compared with pragmatic or syntactic aspects
of linguistic performance. This finding is also
confirmed in the Wulfeck et al. study of error detection
(i.e. selectively greater loss of sensitivity to agreement
vs. ordering errors in both Italian and English aphasics,
although agreement sensitivity was still significantly
greater among the Italians). Given the degree to which
morphological knowledge is preserved in the same
patients, it is all the more surprising that
morphological processing deficits are so pervasive
across different language types. We were particularly
surprised to find severe deficits in the use of case
morphology in our study of sentence comprehension in
Turkish and Hungarian (MacWhinney et al., this issue),
because case morphemes are extremely high in
information value in both these languages, and because
errors of case marking are relatively rare in the
expressive language of Turkish and Hungarian aphasics
(in line with predictions based on cue validity
Slobin, this issue; MacWhinney & Osmn-Sgi, this
issue). To account for this pattern, we suggest that the
processing costs associated with grammatical
morphology may be particularly high in receptive
language a conclusion that is bolstered by findings in
the next category.
(4) Morphology is vulnerable across patient
groups
First, we have found no evidence to support the
idea that Broca's aphasics are unique in the inability to
process closed-class morphemes. In richly inflected
languages, rates of morpheme substitution errors are
remarkably similar for Broca's and Wernicke's aphasics
(Bates, Friederici & Wulfeck, 1988; MacWhinney &
Osmn-Sgi, this issue; Slobin, this issue; Tzeng,
Chen & Hung, this issue). For example, German
Broca's aphasics produce the wrong form of the article
16% of the time, compared with a 17% substitution rate
for German Wernicke's. The same is also true in
Chinese: Broca's and Wernicke's aphasics both make
substitution errors on noun classifiers (a complex class
of function words in a language with relatively few
closed-class items, and no inflectional morphology of
any kind). We conclude that the contrast between
agrammatism (attributed to Broca's aphasia) and
paragrammatism (attributed to Wernicke's aphasia) has
been greatly exaggerated, due to the fact that morpheme
substitution errors are relatively ambiguous and difficult
to detect in English (for a detailed discussion, see Bates
& Wulfeck, 1989a & b). There are subtle differences in
the errors produced by Broca's and Wernicke's aphasics
(see below), but morphology is the domain that is most
prone to error in both these patient groups.
Second, we have found evidence for receptive
agrammatism in patients who show no signs of
grammatical impairment in their spontaneous speech.
For example, we have administered the sentence
comprehension task described above to a range of Italian
patient groups including anomics, right-hemisphere
patients, neurological patients without focal brain
injury, and some nonneurological patients from the
orthopedic ward (Bates et al., 1987a; Bates & Wulfeck,
1989b; Smith & Bates, 1987); all showed significant
and selective impairments in the ability to use subject-
verb agreement as a cue to sentence meaning deficits
that were in many cases just as severe as the deficits
shown by Broca's and Wernicke's aphasics. The same
receptive/expressive dissociation occurs in our studies of
Turkish and Hungarian (MacWhinney et al., this issue),
showing up in patients who were classified as simple
anomics, and in a subset of nonneurological patient
controls despite the fact that cue validity is a
powerful cue to meaning for normal adults in both these
languages, used reliably by small children between 2-3
years of age. This cannot be a simple effect of age or
education, because healthy controls matched for age and
education perform like college students on the same
sentence comprehension task. Instead, it suggests that
selective impairments of grammatical morphology may
result from global perceptual and/or cognitive
limitations above and beyond the specific effects of
focal brain injury.
We have tried to disentangle global vs. specific
causes of morphological breakdown, in several ways.
For example, to study the effect of perceptual
limitations on production and comprehension of
morphology, we have carried out studies of congenitally
deaf but neurologically intact speakers of Italian
(Volterra & Bates, 1989). These subjects also display a
selective impairment in their ability to comprehend and
produce grammatical inflections and function words. In
act, we have found a striking impairment in the use of
grammatical morphemes by a highly educated and
successful deaf woman who is indistinguishable from
Overview
9
normal in lexical and syntactic aspects of her written
Italian, demonstrating near-perfect knowledge of her
grammar in off-line metalinguistic tasks. At the same
time, we have also uncovered some interesting
differences between deaf subjects and aphasic patients in
the kinds of errors produced in written and spoken
Italian. For example, a much higher proportion of
errors in the deaf involve free-standing function words
(rather than bound inflections). Also, gender errors are
extremely rare in the deaf but quite common in aphasic
patients, suggesting that perceptual limitations may
have less of an effect on morphemes that are an inherent
part of the word (e.g., gender). Taken together, these
results support the idea that perceptual limitations have
a selectively greater impact on the use of grammatical
morphology one factor (but probably not the only
factor) in receptive and expressive agrammatism.
Additional evidence for the selective impact of
global perceptual and/or cognitive factors on
morphology comes from our pilot studies testing
normals under adverse processing conditions. For
example, Kilborn (this issue) has conducted a ground-
breaking study of German- and English-speaking college
students in an on-line version of our sentence
interpretation task, using auditory stimuli that are
partially masked by "pink noise" (i.e. random noise
restricted to the speech band). The noise manipulation
has virtually no effect on English subjects: There is no
reduction in the strength of word order as a cue to
agent/object relations (although reaction times are
slowed overall), and little effect on agreement
morphology (which is largely ignored by English
subjects even under optimal conditions). However,
German subjects display a marked and very specific
reduction in the use of agreement morphology (a very
strong cue for Germans under normal conditions), with
virtually no effect on animacy or word order
mirroring our results with German aphasics.
We are certainly not trying to claim that all aphasic
symptoms can be reduced to normal processing under
some form of stress. However, in order to separate the
effects of global stress from specific effects of focal
brain injury, we believe that the usual normal control
conditions used in neuropsychological research must be
supplemented by observations of normals operating
under stressed conditions. Global stress controls should
include what Norman & Bobrow (1975) call data-
limited processes (e.g., auditory stimuli degraded by
a partial noise mask) and resource-limited
processes (e.g., cognitive overload in a dual task
paradigm). Indeed, such manipulations may become
standard in the field as aphasia researchers make
increased use of on-line processing tasks of the sort
described in this special issue.
(5) Similarity between lexical and
grammatical symptoms
Although the closed class appears to be
quantitatively more vulnerable than other aspects of
language processing, we have also found evidence for
qualitative similarities in the lexical (word-finding) and
grammatical symptoms displayed by aphasic patients
similarities that are particularly clear in spontaneous
speech by speakers of a richly inflected language (Bates,
Friederici & Wulfeck, 1988; MacWhinney & Osmn-
Sgi, this issue; Slobin, this issue). Examples include
the following.
(a) There are frequency and/or markedness
effects on the production of grammatical morphemes
(e.g., a bias toward the nominative in case-inflected
languages), and in the production of whole sentence
frames (e.g., a bias toward canonical word order in all
the languages studied, and a bias toward more frequent
noncanonical word order variations); these findings
resemble well-known effects of frequency in the
production of content words.
(b) Patients who are struggling to produce the right
function word in an obligatory slot (e.g., production of
articles in German) often go through a process of
successive approximation and self-correction that is
highly reminiscent of the word-finding episodes that are
so often reported for both fluent and nonfluent aphasics
(e.g., "(die...der...das...die...den)..den Hund").
(c) There are effects of semantic relatedness in
morpheme production errors, in those languages in
which it is possible to rank morphemes along a
semantic gradient (e.g., substitution of closely related
locative suffixes in Hungarian MacWhinney &
Osmn-Sgi, this issue). This result is similar to the
effects of semantic relatedness that are so often seen
when a patient substitutes one content word for another
(i.e. semantic paraphasias).
(d) Grammatical structures appear to be affected by
"priming" effects (i.e. a build-up of contextual
information), similar at some level to the priming
effects that are known to occur in lexical access (see
also Bock, 1986; Lukatela, Kostic, Feldman & Turvey,
1983).
Although these findings are certainly not
conclusive, they are compatible with the predictions of
the Competition Model (and other interactive activation
models), and constitute a major area for exploration in
future crosslinguistic research, through detailed
comparisons of lexical and grammatical processing,
varying predictability (contextual buildup), frequency
and semantic relatedness in both domains.
(6) Patient group differences
Finally, there are several areas in which we find
consistent differences between Broca's and Wernicke's
aphasics in our studies to date, beyond the diagnostic
criteria that we used to define these groups in the first
place (see below). These differences are subtle, but they
appear to hold up across different language types,
serving as "neurolinguistic universals" that may be
particularly informative as our field moves toward a new
Overview
10
theory of intrahemispheric organization for language.
(a) Redundancy. There are differences in the
degree to which Broca's and Wernicke's aphasics are able
to exploit converging sources of information in a
receptive language task, i.e. the redundancy effects
described above in Serbo-Croatian (Smith & Bates,
1987) and in English, Italian and German (Bates,
Friederici & Wulfeck, 1987a). This pattern is
compatible with the idea that Broca's aphasics suffer
from "under-excitation" (requiring more information to
reach threshold), while Wernicke's suffer from "under-
inhibition" (so that additional information activates too
many associates, effectively doing more harm than
good).
(b) Frequency/Markedness. There are
differences in the nature of the morphological
substitution errors produced by each group, reflecting
what may be a conscious or unconscious difference in
the underlying strategies that govern sentence planning
in these patients. Nonfluent patients tend to avoid
contexts that require production of a difficult
morphosyntactic structure; when they do make
mistakes, they tend to substitute a simpler, more
frequent and/or less marked form. Fluent patients do
not seem to have this kind of control over their own
speech; they barge ahead and attempt complex
constructions, making a less systematic array of
substitution errors (including cases in which a low-
frequency item is substituted for the base form). These
differences are compatible with (for example) a proposal
by Milberg, Blumstein & colleagues that controlled
aspects of processing are spared in Broca's aphasia, with
automatic processes spared to a greater extent in
Wernicke's (e.g., Blumstein & Milberg, 1983; Milberg
& Blumstein, 1981).
(c) Form class. In Bates, Chen, Tzeng, Li &
Opie (this issue), we have replicated a selective
dissociation between action naming (impaired in Broca's
aphasia) and object naming (impaired in Wernicke's
aphasia). Although this difference has been reported by
other investigators (e.g., Miceli, Silveri, Romani &
Caramazza, 1989; Saffran, Berndt & Schwartz, 1989),
crosslinguistic studies have helped to eliminate the
hypothesis that verb problems are a byproduct of the
fact that verbs carry more grammatical marking, because
the same noun/verb dissociation between Broca's and
Wernicke's aphasics appears in Chinese (where there are
no inflections of any kind on nouns or verbs) and
Hungarian (where nouns and verbs both require
extensive morphological marking Osmn-Sgi,
1990). Furthermore, a particularly interesting and
informative variant of the noun/verb problem appears in
our data for Chinese, a language in which many words
are compounds made up of a verbal element and a
nominal element (e.g., the verb "to read", which can be
translated literally as LOOK-BOOK). In their attempts
to produce such compound words, Chinese Broca's tend
to omit or substitute the verbal component, while
Wernicke's err more often on the nominal component.
As elaborated in more detail by Bates et al. (this issue),
these facts suggest that the noun/verb dissociation must
be explained in lexical and/or semantic terms, a finding
with implications for localization of function that
transcend either a comprehension/production or a
grammar/semantics account of the two aphasias.
To summarize, crosslinguistic research has added
immeasurably to our understanding of universal and
language-specific symptom patterns in aphasia. These
results have forced a compromise between two
contrasting theories of grammatical impairment, and
they have opened up a whole new list of questions that
would be difficult to answer within any single language
(for example, English). We believe that the cross-
linguistic method holds great promise for the future of
aphasiology. However, it also presents some serious
methodological problems that must be considered before
we proceed.
II. THE PROBLEM OF PATIENT
SELECTION AND COMPARABILITY OF
PATIENTS ACROSS LANGUAGE TYPES
In the studies presented within this special issue,
we have focussed on Broca's and Wernicke's aphasia, in
comparison with a range of control populations. These
two groups were chosen because (1) they both display
forms of grammatical impairment that are of major
interest in a crosslinguistic study, and (2) differences
between these groups are correlated (albeit imperfectly)
with anterior and posterior lesion sites, yielding
information that is relevant to a characterization of
intrahemispheric organization for language. However,
because we have elected a Patient Group by Language
Group design
1
, we must ask ourselves whether it is

1
Obviously we have taken a stand in favor of group
studies. For reasons that we have outlined in considerable
detail in two methodological papers (Bates, McDonald,
MacWhinney & Appelbaum, 1991; Bates, Appelbaum &
Allard, 1991), we disagree with arguments against group
studies offered by Caramazza and his colleagues (e.g.
Caramazza, 1986). Although individual case studies play
an important role in aphasia research, these studies can
only inform us about those patterns of language sparing
and impairment that are possible within a given language.
Case studies cannot tell us which patterns are likely or
typical within and between language types. To test the
competing hypotheses that we have described here against
a cross-linguistic data base, we need information of the
latter type. Indeed, we are quite concerned that our own
group studies are not large enough, since results are often
based on samples of only 4 - 10 patients within a particular
Language X Patient cell. This is one reason why we are
interested in the establishment of a computerized data
exchange system that will permit researchers in different
language communities to share free-speech data.
Overview
11
possible to select equivalent or even vaguely
comparable groups of patients when the languages in
question vary radically along structural and statistical
dimensions that are central to a definition of
"agrammatism", "paragrammatism" or (for that matter)
grammatical impairment in any form.
To clarify the point, consider the contrasts that hold
between just two of the languages represented in this
special issue, Chinese and Turkish.
Chinese has what may be the most austere system
of grammatical morphology in the world. There is
essentially no bound morphology in this language, i.e.
no marking for gender, number or tense, no agreement
phenomena, indeed no verb conjugations or noun
declensions of any kind. The language does provide a
set of free-standing function words, but almost all of
these words are homophonous with single-syllable
content words that have a related meaning (e.g., to
indicate that an act has been completed, one adds a
particle after the verb that is equivalent in form and
meaning to the word for "finished", as in "EAT-
FINISHED"). Hence the line between closed-class and
open-class words is much less obvious in Chinese than
it is in other language families. Furthermore, those
function words that do exist are obligatory only in
particular discourse contexts. As a result, sentences
with no function words and (of course) no bound
morphemes of any kind can be completely grammatical
in this language. It is fair to say that normal Chinese
speech has many of the properties that characterize the
"telegraphic" speech of so-called agrammatic aphasics in
English!
Turkish stands at the opposite extreme: a case-
inflected language with an extremely rich and regular set
of bound morphemes for verbs, nouns, pronouns and
several other elements. An entire sentence may consist
of no more than one or two richly inflected words, each
inflection standing in a particular position within an
ordered string of prefixes or suffixes around a root word.
Slobin (this issue) demonstrates that these inflections
are remarkably well preserved in the speech of both
fluent and nonfluent Turkish aphasics. In other words,
Turkish Broca's aphasics simply do not produce
telegraphic speech. For that matter, as Slobin notes,
Turkish 2-year-olds do not produce telegraphic speech
either. Indeed, telegraphic speech is not even used in
Turkish telegrams!
How on earth could we possibly hope to "match"
patients for severity or type of aphasia across languages
that differ so radically in their basic structure?
Obviously it would be sheer folly to match Chinese and
Turkish patients for mean length of utterance, number
of words per minute, or any other straightforward
metric. For the same reasons, it would be most unwise
to rely on standardized scores from the "same" aphasia
test. To be sure, standard tests like the Boston

Diagnostic Aphasia Examination (Goodglass and
Kaplan, 1983) have been translated and adapted in many
of the languages represented within our crosslinguistic
research. But adaptation and norming are not the same
thing. We have no way of knowing at this time
whether the "same" score means the "same" thing in
English, Turkish or Chinese, because these powerful
structural and statistical differences between languages
are not taken into account when standardized tests are
translated from one language to another.
To illustrate the same point with a less exotic
example, we know from our previous work on English,
German and Italian (Bates, Friederici and Wulfeck,
1988) that Italian and German Broca's aphasics tend to
produce a much larger number of function words than
their English counterparts. As we noted earlier, German
patients typically omit the article before the noun no
more than 5-20% of the time, compared with an average
omission rate of 70% in our English sample. Bates et
al. attribute this crosslinguistic difference to an effect
of cue validity, i.e. to the fact that the case-marked
article carries crucial information of agent-object
relations in German, while the article carries much less
information and hence can be omitted with relatively
little communicative loss in English. Of course, there
is an alternative interpretation: It is possible that this
difference derives from a failure to match subjects over
languages, i.e. we have accidentally selected mild
German aphasics and compared them with English
patients who are more seriously impaired. However, if
that were the case, shouldn't we expect similar
differences in severity along all linguistic dimensions?
And yet our English patients were not more impaired
than their German counterparts in the use of canonical
word order a much more important aspect of
grammar in English.
In the same vein, our Italian patients stand between
their German and English counterparts in rate of article
omission. Again, we could attribute this to an
accidental difference in severity level in our two
language groups. And yet a very different profile is
observed in omission of the sentence subject: Subject
omission is clearly much more common in Italian
patients than it is in German or in English (Wulfeck et
al., 1989). This difference is best explained by the fact
that subject omission is a legal option in Italian (e.g.,
it is possible to say "Is going to the store" or "Is
raining"); the same option is not permitted in German
or in English. In short, if all the crosslinguistic
differences that we have observed to date were due to
random differences in severity over language groups,
then it would be difficult to explain why these
differences invariably line up in the predicted language-
specific direction.
This trend is comforting, but it does not make the
problem of patient-matching disappear. In principle, it
should be possible to match patients for their relative
severity, on a scale that is calibrated differently for each
Overview
12
individual language. With a sufficiently large body of
norming information in each language group, we could
match patients for their degree of deviation from normal
in their particular language a procedure that takes
crosslinguistic differences in amount and type of
grammatical marking into account. For example, it
may turn out to be the case that a severely impaired
Italian Broca's aphasic (i.e. a patient in the bottom tenth
percentile for Italian) omits bound morphemes at a 50%
level, comparable to omission rates for a mildly
impaired English Broca's aphasic (i.e. a patient at the
50th percentile for English). We believe that percentile
matches of this sort will ultimately prove to be the
correct solution to the problem of patient selection and
patient matching in crosslinguistic research.
Unfortunately, the data base that is necessary to develop
such percentile scores still does not exist (but see
Paradis, 1987, for some important steps in this
direction). Until a large crosslinguistic data base
becomes available, we must be satisfied with an interim
solution to the problem.
In our own crosslinguistic project, we have
proceeded as follows. Patients are defined on behavioral
grounds, according to a set of classic definitions that are
recognized by clinicians in all the participating language
communities (despite wide variations in the way these
symptoms are realized from one language to another).
Broca's aphasics are defined as nonfluent patients who
display an abnormal reduction in utterance length and
sentence complexity, with marked errors of omission
and/or substitution in grammatical morphology;
comprehension abilities appear to be normal in free
conversation. By contrast, Wernicke's aphasics are
defined as patients who produce fluent or hyperfluent
speech with an apparently normal melodic line; these
patients typically display serious word-finding diffi-
culties, with semantic and/or phonological paraphasias;
comprehension abilities are clearly impaired in free
conversation. In all the languages under study,
Wernicke's aphasics can be distinguished from (for
example) anomics, i.e. fluent patients who display
word-finding problems in free-speech and confrontation
naming, in the absence of severe paraphasias or
paragrammatism, and in the presence of normal
comprehension abilities in free conversation. Because
there are still no reliable grounds for a percentile-based
match, we define patient groups within each language,
according to their fit to one of two prototypes used by
neurologists and speech pathologists in that com-
munity. For example, a prototypic Broca's aphasic
would show reduced fluency and phrase length, and a
tendency toward omission of functors (relative to
normals in that language). Hence patients are matched
across languages in the sense that they represent
different degrees of deviation from a prototype
developed out of observed variation within each
language group. This permits us to compare the "best"
and the "worst" patients across languages, as well as
those who fit the mean. This is, of course, the logic
behind percentile scores and may eventually lead to
cross-language percentile matching when enough cases
have accu-mulated within each language group to permit
the development of comparable aphasia norms.
Note also that we do not attempt to match patients
by lesion type, for two reasons: (1) neural imaging is
not always available, nor are the radiological facilities
comparable from one site to another (e.g., San Diego,
Rome, Taiwan and Budapest); (2) there are imperfect
correlations among lesion site, lesion size and
behavioral syndrome, exceptions that might (for all we
know right now) interact in unknown ways with
language type. However, we include all available
neurological information in our patient archives; within
some language and patient categories, we will soon
have enough cases to permit post hoc analyses of the
relationship between lesion type and symptom patterns,
within and across language groups.
At each of the four research sites, neurologists and
speech pathologists throughout the community refer
patients who (in their clinical judgment) fit one of the
above diagnoses. These diagnoses are accompanied by
neurological records (including CT scans in many
cases), together with the results of those standard
aphasia batteries that are used at each research site. To
eliminate the possibility that a patient has changed
status since the diagnosis provided at referral, we screen
all patients in a biographical interview administered and
recorded prior to testing. In addition, we exclude all
patients with one or more of the following conditions:
history of multiple strokes; significant hearing and/or
visual disabilities; severe gross motor disabilities;
severe motor-speech involvement such that less than
50% of subject's speech attempts are intelligible;
evidence that subject is neurologically or physically
unstable and/or less than 3 months post onset.
This is our solution for the moment. Meanwhile,
as we noted earlier, members of our research team have
begun to transcribe and code their free-speech data in
accordance with a modification of the coding scheme
developed by the Child Language Data Exchange
System (CHILDES, MacWhinney & Snow, 1985).
These records are being placed in a computer data base
called ALDES (Aphasic Language Data Exchange
System), a system that will be open to all interested
investigators (Bates & Wulfeck, 1989b). Eventually,
we hope that other investigators will contribute to the
ALDES database, providing a base of free-speech data
that is large enough for the development of cross-
linguistic norms.
With that note of warning, and accompanying
message of hope, let us now turn to the nine
substantive contributions by our crosslinguistic
collaborators.
Overview
13
REFERENCES
Bates, E., Appelbaum, M., & Allard, L. (1991).
Statistical constraints on the use of single cases in
neuropsychological research. Brain and Language,
40, 295-329.
Bates, E., Chen, S., Tzeng, O., Li, P., & Opie, M.
(1991). The noun-verb problem in Chinese
aphasia. To appear in a special issue on
crosslinguistic aphasia in Brain and Language.
Bates, E., Friederici, A., & Wulfeck, B. (1987a).
Comprehension in aphasia: A crosslinguistic
study. Brain and Language, 32, 19-67.
Bates, E., Friederici, A., & Wulfeck, B. (1987b).
Grammatical morphology in aphasia: Evidence
from three languages. Cortex, 23, 545-574.
Bates, E., Friederici, A., & Wulfeck, B. (1988).
Grammatical morphology in aphasia: A reply to
Niemi et al. Cortex, 24, 583-588.
Bates, E., Friederici, A., Wulfeck, B., & Juarez, L.
(1988). On the preservation of word order in
aphasia: Cross-linguistic evidence. Brain &
Language, 33, 323-364.
Bates, E., Hamby, S., & Zurif, E. (1983). The effects
of focal brain damage on pragmatic expression.
Special issue on brain and language (Doreen
Kimura, Ed.). Canadian Journal of Psychology,
37, 59-63.
Bates, E., & MacWhinney, B. (1987). Competition,
variation and language learning. In B. MacWhinney
(Ed.), Mechanisms of language acquisition.
Hillsdale, NJ: Erlbaum.
Bates, E., McDonald, J., MacWhinney, B., &
Appelbaum, M. (1991). A maximum likelihood
procedure for the analysis of group and individual
data in aphasia research. Brain and Language, 40,
231-265.
Bates, E. & Wulfeck, B. (1989a). Crosslinguistic
studies of aphasia. In B. MacWhinney & E. Bates
(Eds.), The crosslinguistic study of sentence
processing. New York: Cambridge University
Press.
Bates, E. & Wulfeck, B. (1989b). Comparative
aphasiology: A cross-linguistic approach to
language breakdown. Aphasiology, 3, 111-142 and
161-168.
Blumstein, S. & Milberg, W. (1983, October).
Automatic and controlled processing in speech-
language deficits in aphasia. Symposium on
Automatic Speech. Abstracts of the 21st Annual
Meeting of the Academy of Aphasia, Minneapolis.
Bock, K. (1986). Syntactic persistence in language
production. Cognitive Psychology, 18, 355-387.
Bradley, D., Garrett, M., & Zurif, E. (1980). Syntactic
deficits in Broca's aphasia. In D. Caplan (Ed.),
Biological studies of mental processes.
Cambridge, MA: MIT Press.
Caramazza, A. (1986). On drawing inferences about the
structure of normal cognitive systems from the
analysis of patterns of impaired performance: The
case for single-patient studies. Brain and
Cognition, 5, 41-66.
Chen, S. (1989, April). On grammatical deficits in
Chinese aphasia. Abstracts of the 18th Annual
Linguistics Symposium. University of Wisconsin,
Milwaukee.
Chen, S., Bates, E., & Tzeng, O. (1990, March).
Sentence interpretation of Chinese aphasics.
Abstracts of the Second Language Research
Forum. University of Oregon, Eugene.
Friederici, A. (1986). Autonomy and automaticity:
Accessing function words during sentence
comprehension. In G. Denes, C. Semenza, P.
Bisacchi & E. Andreewsky (Eds.), Perspectives in
cognitive neuropsychology. Hillsdale, NJ:
Erlbaum.
Friederici, A. & Kilborn, K. (1989). Temporal
constraints on language processing in Broca's
aphasia. Journal of Cognitive Neuroscience, 1,
262-272.
Friederici, A., Weissenborn, J., & Kail, M. (1991).
Pronoun comprehension in aphasia: A comparison
of three languages. To appear in a special issue on
crosslinguistic aphasia in Brain and Language.
Garrett, M. (1980). Levels of processing in sentence
production. In B. Butterworth (Ed.), Language
production: Vol. 1. Speech and talk. New
York/London: Academic.
Goldstein, K. (1948). Language and language
disturbances: Aphasic symptom complexes and
their significance for medicine and theory of
language. New York: Grune & Stratton.
Goodglass, H. & Kaplan, H. (1983). Boston
Diagnostic Aphasia Examination (2nd ed.).
Philadelphia: Lea and Febiger.
Grodzinsky, Y. (1986). Language deficits and the
theory of syntax. Brain and Language, 27, 135-
159.
Hinton, G.E., & Shallice, T.(1989). Lesioning a
connectionist network: Investigations of acquired
dyslexia. (Tech. rep. CRG-TR-89-30). University
of Toronto.
Kean, M.-L. (1979). Agrammatism: A phonological
deficit? Cognition, 7, 69-84.
Kean, M.-L. (Ed.). (1985). Agrammatism. New York:
Academic Press.
Kilborn, K. (1991). Selective impairment of
grammatical morphology due to induced stress in
normal listeners: Implications for aphasia. To
appear in a special issue on crosslinguistic aphasia
in Brain and Language.
Linebarger, M., Schwartz, M., & Saffran, E. (1983).
Sensitivity to grammatical structure in so-called
agrammatic aphasics. Cognition, 13, 361-392.
Overview
14
Lukatela, K., Crain, S., & Shankweiler, D. (1988).
Sensitivity to inflectional morphology in
agrammatism: Investigation of a highly inflected
language. Brain & Language, 33, 1-15.
Lukatela, G., Kostic, A., Feldman, L., & Turvey, M.
(1983). Grammatical priming of inflected nouns.
Memory and Cognition, 11, 59-63.
MacWhinney, B. & Bates, E. (1978). Sentential
devices for conveying givenness and newness: A
cross-cultural development study. Journal of
Verbal Learning and Verbal Behavior, 17, 539-
558.
MacWhinney, B. & Bates, E. (Eds.). (1989). The
crosslinguistic study of sentence processing. New
York: Cambridge University Press.
MacWhinney, B. & Osmn-Sgi, J. (1991).
Inflectional marking in Hungarian aphasics. To
appear in a special issue on crosslinguistic aphasia
in Brain and Language.
MacWhinney, B., Osmn-Sgi, J., & Slobin, D.
(1991). Sentence comprehension in aphasia in two
clear case-marking languages. To appear in a
special issue on crosslinguistic aphasia in Brain
and Language.
MacWhinney, B. & Snow, C. (1985). The child
language data exchange system. Journal of Child
Language, 12, 271-296.
Menn, L. & Obler, L.K. (Eds.). (1990). Agrammatic
aphasia: Cross-language narrative sourcebook.
Amsterdam: John Benjamins.
Miceli, G., Silveri, M.C., Romani, C., & Caramazza,
A. (1989). Variation in the pattern of omissions
and substitutions of grammatical morphemes in
the spontaneous speech of so-called agrammatic
patients. Brain and Language, 36, 447-492.
Milberg, W. & Blumstein, S. (1981). Lexical decision
and aphasia: Evidence for semantic processing.
Brain & Language, 14, 371-385.
Norman, D. & Bobrow, D. (1975). On data-limited and
resource-limited processes. Cognitive Psychology,
7, 44-64.
Osmn-Sgi, J. (1989). A note on action naming in
Hungarian aphasic patients. (Tech. rep. No.
9012). San Diego: University of California,
Center for Research in Language.
Ostrin, R. & Schwartz, M. (1986). Reconstructing
from a degraded trace: A study of sentence
repetition in agrammatism. Brain & Language, 28,
328-345.
Paradis, M. (1987). The assessment of bilingual
aphasia. Hillsdale, NJ: Erlbaum.
Posner, M., Petersen, S., Fox, P., & Raichle, M.
(1988). Localization of cognitive operations in the
human brain. Science, 240, 1627-1631.
Rizzi, L. (1980). A restructuring rule in Italian syntax.
In S.J. Keyser (Ed.), Recent transformational
studies in European languages. Cambridge, MA:
MIT Press.
Rumelhart D. & McClelland J.L.(Eds.). (1986).
Parallel distributed processing: Explorations in
the microstructure of cognition. Cambridge, MA:
MIT Press.
Saffran, E., Berndt, R., & Schwartz, M. (1989). The
quantitative analysis of agrammatic production:
Procedure and data. Brain & Language, 37, 440 -
479.
Seidenberg, M., McClelland, J., & Patterson, K. (1987,
July). A distributed developmental model of visual
word recognition, naming and dyslexia.
Symposium on Connectionism. Abstracts of the
Annual Meeting of the Experimental
Psychological Society (U.K.). Oxford.
Shankweiler, D., Crain, S., Gorrell, P., & Tuller, N.
(1989). Reception of language in aphasia.
Language and Cognitive Processes, 4, 1-33.
Slobin, D. (1991). Aphasia in Turkish: Speech
production in Broca's and Wernicke's patients. To
appear in a special issue on crosslinguistic aphasia
in Brain and Language.
Smith, S. & Bates, E. (1987). Accessibility of case
and gender contrasts for agent-object assignment in
Broca's aphasics and fluent anomics. Brain and
Language, 30, 8-32.
Smith, S. & Mimica, I. (1984). Agrammatism in a
case-inflected language: Comprehension of agent-
object relations. Brain and Language, 21, 274-
290.
Swinney, D., Zurif, E., & Nicol, J. (1989). The effects
of focal brain damage on sentence processing: An
examination of the neurological organization of a
mental module. Journal of Cognitive
Neuroscience, 1, 25-37.
Tzeng, O., Bates, E., & Wong, W. (1990). Chinese
aphasia. Invited for submission to Scientific
American.
Tzeng, O., Chen, S., & Hung, D. (1991). Use of
nominal classifiers by Chinese aphasic patients.
To appear in a special issue on crosslinguistic
aphasia in Brain and Language.
Vaid, J. & Pandit, R. (1991). Sentence interpretation
in normal and aphasic Hindi speakers. To appear in
a special issue on crosslinguistic aphasia in Brain
and Language.
Volterra, V. & Bates, E. (1989). Selective impairment
of Italian grammatical morphology in the
congenitally deaf: A case study. Cognitive
Neuropsychology, 6, 273-308.
Wulfeck, B. (1987). Sensitivity to grammaticality in
agrammatic aphasia: Processing of word order
and agreement violations. Doctoral dissertation,
UCSD.
Wulfeck, B. (1988). Grammaticality judgments and
sentence comprehension in agrammatic aphasia.
Overview
15
Journal of Speech and Hearing Research, 31, 72-
81.
Wulfeck, B., Bates, E. (1990). Differential sensitivity
to errors of agreement and word order in Broca's
aphasia. Submitted for publication.
Wulfeck, B., Bates, E., & Capasso, R. (1991). A
crosslinguistic study of grammaticality judgments
in Broca's aphasia. To appear in a special issue on
crosslinguistic aphasia in Brain and Language.
Wulfeck, B., Bates, E., Juarez, L., Opie, M., Friederici,
A., MacWhinney, B., & Zurif, E. (1989).
Pragmatics in aphasia: Crosslinguistic evidence.
Language and Speech, 32, 315-336.