Discussion 2: Language and Cognition

Introduction

Towards a new functional anatomy of language

Abstract

The classical brain-language model derived from the work of Broca, Wernicke, Lichtheim,

Geschwind, and others has been useful as a heuristic model that stimulates research and as a clinical

model that guides diagnosis. However, it is now uncontroversial that the classical model is

(i) empirically wrong in that it cannot account for the range of aphasic syndromes, (ii) linguistically

underspecified to an extent that prohibits contact with the language sciences, and (iii) anatomically

underspecified. We briefly summarize some of the central issues that motivate why a new functional

anatomy of language is necessary, in the context of introducing a collection of articles that describe

systematic new attempts at specifying the new functional anatomy. The major convergent

observations are highlighted and the emergent conceptual and empirical trends are identified.

q 2004 Elsevier B.V. All rights reserved.

Keywords: Functional anatomy; Language

1. The old functional anatomy of language

Every student of the language sciences has come across the illustration of a left cerebral

hemisphere in which there is labeled an inferior frontal area, Broca’s area, a posterior

superior temporal region, Wernicke’s area, and a connecting fiber tract, the arcuate

fasciculus. This anatomic diagram, an image with iconic status in neuroscience, forms the

basis of a neurolinguistic model that has informed research for almost 150 years and

constitutes the canonical model of brain and language taught across disciplines. The model

originated in the 19th century with the influential work of Broca (Broca, 1861, 1865),

Wernicke (Wernicke, 1874/1977), and Lichtheim (Lichtheim, 1885). It lost favor, and

was, indeed, virtually forgotten over the course of the first half of the 20th century, but was

revived and popularized again beginning in the 1960s by Norman Geschwind

(e.g. Geschwind, 1965).

The classical conceptualization of the neural basis of language has been immensely

useful both as a heuristic model to stimulate research and as a clinical model to guide

diagnosis (Damasio, 1992). However, many of the neuropsychological, anatomical, and

linguistic assumptions implicit or explicit in the model are known to be problematic, and

therefore many researchers have abandoned classical ideas and assumed new theoretical

0022-2860/$ - see front matter q 2004 Elsevier B.V. All rights reserved.

doi:10.1016/j.cognition.2003.11.001

Cognition 92 (2004) 1–12

www.elsevier.com/locate/COGNIT

positions, almost invariably focused on much more circumscribed pieces of the puzzle

than was targeted by the classical theorists. This situation has placed us on the verge of

throwing out the classical-model baby with the bathwater. The bathwater, of course, is

comprised of several details of the model that have failed to stand up empirically

(discussed below). The baby is represented by many of the core ideas (again see below),

and importantly, the goal of the model, which, to paraphrase Wernicke’s (1874/1977)

subtitle, is to develop a psychological model of language on an anatomical basis. This is a

broad goal, and therefore one that is not easily achieved. It is much easier, and arguably

more rational, to restrict one’s effort to a smaller piece of the problem. But such a strategy

carries with it the danger of losing sight of the primary goal, ultimately leading to a

fractionated field exhibiting little theoretical convergence. Perhaps we can learn from the

classical models that if we must work locally, we should, as the political saying goes, at

least think globally. This line of argument applies with equal force to methodological

issues. We now have a range of investigative methods at our disposal – hemodynamic

imaging, electromagnetic recording, transcranial magnetic stimulation, neuropsycho-

logical, and behavioral methods – and it is critical that findings from these various vantage

points be integrated.

The goal of this special issue is to aim for theoretical and methodological integration

in language-brain research, and hence to refocus some attention on the broader

organization. Predictably, the collection of papers falls far short of this goal. However,

progress has been made. In the remainder of this paper, we highlight some of this

progress and point out some of the shortcomings. We consider the background issues that

form the basis for the new and critical work presented in the articles in this volume,

starting with a cursory review of the classical brain-language model and some of its

implicit assumptions and problems. We end by offering a perspective on what some of

the current trends are and where the research program on the neurobiological basis of

language might be headed.

2. The classical model

2.1. What’s right with the classical Wernicke – Lichtheim – Geschwind model?

The classical model contains many viable and relevant ideas, but the model is often

misunderstood by modern researchers (for discussion see de Bleser, Cubelli, & Luzzatti,

1993). We review some remarks by Wernicke (1874/1977) that outline the core

components of the model, and highlight the extent to which his theorizing was closely

related to many contemporary considerations. These remarks are based on Eggert’s 1977

translation of Wernicke’s original monograph of 1874 (Wernicke, 1874/1977).

Wernicke, drawing on the work of his mentor Theodore Meynert (Whitaker & Etlinger,

1993), conceived of two language centers, based on the hypothesis that cortex anterior to

the Rolandic sulcus served motor functions, whereas cortex posterior to the Rolandic

sulcus served sensory functions. According to Wernicke, sensory and motor cortices

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–122

served not only in the direct registry of sensory experience or in the direct execution of

movement, but also in the memory storage of sensory and motor imagery.

It followed from this general functional-anatomic framework that the anterior motor-

speech area identified by Broca a decade earlier was a site for the deposition of motor-

images for speech (cf. Indefrey & Levelt, this volume), and a posterior area should be the

site for the deposition of acoustic images for words (cf. Boatman, Scott, & Wise, this

volume; Hickok & Poeppel, this volume). And, indeed, Wernicke described cases that he

argued confirmed the latter site to be the superior temporal gyrus (STG).

Wernicke hypothesized that these two language areas were connected originally as a

consequence of a (subcortical) reflex arc active in the process of language acquisition. The

child acquiring language, upon hearing a word or syllable, reflexively mimics that

word/syllable in speech, which in turn causes the temporally coincident activation of

sensory and motor images in the cortex. The temporally coincident activation of cortical

sensory and motor memory images of the word, in turn, leads them to become associated

directly, via a cortico-cortical pathway that he believed coursed behind the insula.

These acoustic and motor images were distinct from the concepts with which they were

associated. According to Wernicke, the acoustic image of a word is a purely auditory

entity. The concept, on the other hand, is formed by the sum total of the memory images

associated with, say, a particular object. This meant that in order to comprehend the

meaning, an association had to be made between the acoustic image and the various

sensory memory images representing the concept itself. So not only was there an

anatomical connection between the two language centers, but also connections between

the conceptual representation systems distributed throughout cortex and the two language

centers, both sensory and motor (cf. Damasio et al., this volume). The frame of the famous

‘house diagram’ had been laid out.

Within this model, speech comprehension was seen to involve the activation of an

acoustic word image, which in turn activated a distributed set of sensory and motor

memory images, which comprised the concept associated with the word. Spontaneous

production involved the arousal of a conceptual representation, which in turn activated in

parallel both the motor and sensory word image associated with that concept (cf. Indefrey

& Levelt, this volume). Although Wernicke argued that the conceptual representation

made a direct connection with the motor image center and that it was sufficient to

activate such an image, he believed that the mapping between concept and motor word

image was less precise than that between the concept and the sensory word image (because

of its earlier and stronger association laid down in development). Thus, the activation of

the sensory word image during production served to facilitate and constrain the selection

of the appropriate motor image via the concept-to-sensory-to-motor-word-image pathway

(cf. the idea of analysis-by-synthesis, or forward models discussed in much contemporary

work). This idea, of course, entails the idea that there is some degree of overlap in systems

supporting speech perception and speech production, an idea that has gained recent

empirical support (cf. Hickok & Poeppel, this volume).

Wernicke could account for the symptom complex of aphasia as it was understood at

the time. Damage to the center for motor word images produced a deficit in the production

of speech while leaving comprehension intact. Damage to the center for acoustic word

images produced a deficit in the comprehension of speech since the acoustic memory

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–12 3

images that linked sounds with their meanings had been lost. Production in such ‘sensory’

(Wernicke’s) aphasics was fluent because the connections between conceptual

representations and the motor word center were intact, but speech was disordered

because the concept-to-sensory-to-motor-word-image pathway was no longer available to

facilitate motor-word-image selection. Wernicke also considered the consequences of

disconnection between the two language centers, concluding that it should produce a

disorder, which he termed conduction aphasia, similar to that of sensory aphasia except

with spared comprehension. That is, production should be disordered because words are

selected on the basis of the concept-to-motor-word-image alone, just as in sensory aphasia.

But, comprehension is spared because there is no damage to the sensory-word-image-to-

concept pathway. Wernicke also presented the first case of conduction aphasia in his 1974

monograph.

Striking about this brief review of Wernicke’s work is the extent to which the issues

match many modern considerations on cortical mechanisms involved in speech

processing. If one considers Wernicke’s remarks in the context of the papers in the

present volume, one sees foreshadowed many of the leading ideas of contemporary

research on the neural basis of speech perception and language comprehension.

2.2. What’s wrong with the classical Wernicke – Lichtheim – Geschwind model?

Given that the classical model was so accurate (or prescient) regarding work on

speech, where do substantive criticisms of the model arise? There are (at least) three

issues. First, the model fails to account for a number of facts regarding the symptom

complex of aphasia. Second, the linguistic foundations of the model are

impoverished and conceptually underspecified. Third, the anatomical assertions of

the model have not held up in light of subsequent observations. We consider each

issue briefly.

2.2.1. Symptom complex of aphasia

A number of observations about aphasic symptomatology are not easily

accommodated by the model. These include (i) the existence of anomic aphasia,

(ii) the fact that Broca’s (and conduction) aphasics typically have mild sentence-level

comprehension deficits, (iii) that one finds different distributions of paraphasic errors in

different aphasic subtypes (for example Wernicke’s vs. conduction aphasia), and (iv) the

existence of agrammatism in the speech output of some aphasics, to name a few. More

generally, the various clinical syndromes were assumed to be homogeneous entities

with a largely fixed set of symptoms that could be explained straightforwardly by

damage to a single computational/representational system, or the connection between

such systems. We now appreciate that clinical aphasic syndromes are comprised of

variable clusters of symptoms. The fact that properties characteristic of a given clinical

type can dissociate suggests different computational underpinnings, and therefore a

much more complex architecture.

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–124

2.2.2. Impoverished linguistic model

Some of the shortcomings of the classical model can be traced to its dramatically

underspecified model of language. In the oldest conceptions, language was simply

fractionated into expression (production) and reception (comprehension), a conceptuali-

zation that is obviously too coarse (although it is important to appreciate that the brain

systems supporting these functions were not fractionated simplistically into expressive and

receptive regions – a common misunderstanding today). Later attempts assumed that

‘large scale’ linguistic subsystems could be assigned to the relevant areas, such as syntax

vs. semantics vs. phonology. What was not seriously considered is that such linguistic

domains are themselves not monolithic, but have rich internal structure with numerous

subcomponents and computational requirements. Therefore, it is very problematic to

assign a label such as semantics or syntax to a brain area without being explicit that these

are subdivided domains with specific computational needs.

2.2.3. Anatomical problems

It has become clear that the classical anatomical assertions are not true. Broca’s aphasia

is not caused by damage to Broca’s area (Mohr et al., 1978). Wernicke’s aphasia is not

caused by damage to Wernicke’s area (defined classically as the posterior third of the left

STG) (Bogen & Bogen, 1976; Dronkers, Redfern, & Knight, 2000). Conduction aphasia is

not caused by damage to a white matter pathway, arcuate fasciculus or otherwise, and

appears not to be a disconnection syndrome at all (Anderson et al., 1999; Damasio &

Damasio, 1980; Hickok, 2000; Hickok et al., 2000; Tanabe et al., 1987). Some would

argue that aspects of linguistic function, i.e. speech perception, are organized bilaterally

rather than unilaterally in the left hemisphere (Hickok & Poeppel, 2000). And there is now

good evidence that the classical speech-related regions are not anatomically or

functionally homogeneous (Amunts et al., 1999; Galaburda & Sanides, 1980; Wise

et al., 2001). Furthermore, modern work has identified areas outside of the classical

regions that are implicated in language processing. As is discussed in the contributions in

this volume (for example, Dronkers et al.; Ullman, Indefrey, & Levelt; Damasio et al.),

there are cortical and subcortical regions that clearly contribute to normal language

processing, including the anterior superior temporal lobe, the middle temporal gyrus

(MTG), the temporo-parietal junction, the basal ganglia, and many right-hemisphere

homologues. Since many areas are relevant in addition to Broca’s and Wernicke’s areas, it

is obvious that the model must be rethought.

In summary, it is now rather uncontroversial that the classical model is (i) empirically

wrong in that it cannot reasonably account for the range of aphasic syndromes,

(ii) linguistically underspecified to an extent that prohibits contact with theoretical or

experimental research on language, and (iii) anatomically underspecified.

3. Sources of progress: linguistics and cognitive neuroscience

Unsurprisingly, progress on brain-language research has been driven by advances in

both linguistics and cognitive neuroscience. A major source of conceptual enrichment and

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–12 5

change has been research in theoretical linguistics and psycholinguistics. At least since the

1950s, we assume that there exist many different levels of representation that have

independent motivation and their own internal structure. Experimental research in

psycholinguistics and neurolinguistics has been (more or less) connected to theoretical

progress and has contributed much to our understanding of real-time language processing

(Altmann, 2002). Minimally, these lines of work have established that the linguistic

computational system – both in terms of its formal organization and real-time processing

components – is comprised of many distinct levels that have specialized computational

requirements. This imposes serious constraints on the neuroscience of language: it is clear

now that coarse categorizations of language functions into phonology, syntax, semantics,

or speech perception, lexical processing, speech articulation, syntactic parsing, and so on,

are hopelessly underspecified as linguistic concepts to guide brain mapping endeavors. In

fact, given our current understanding – where the major debates center on what regions

support such putative functions as ‘parsing’, ‘semantics’, or the ‘perception of intelligible

speech’ – it is quite clear that we are scarcely more sophisticated linguistically than

Wernicke was in the 1870s.

The increasingly articulated linguistic and psycholinguistic models provide one major

source (albeit underutilized) for progress in brain-language research. The other major

source derives from technical advances. First, the development of structural CT and MRI

allowed lesion mapping in living patients, and subsequently PET provided the first

glimpses into the normal functioning brain, followed by the development of functional

MRI, finally making non-invasive high-resolution brain imaging widely available.

Paralleling the development of the hemodynamic imaging techniques was the emergence

of multichannel EEG and the advent of MEG, both of which allowed researchers to chart

the timecourse of neural events underpinning language processing. The imaging

techniques have been used extensively – although we would argue not always effectively

(Poeppel, 1996) – and have generated progress in our understanding of the neuroscience

of language. The further development of these techniques and the development of new

ones (e.g. near infrared spectroscopy, transcranial magnetic stimulation, to name two) will

yield further empirical progress.

4. Towards a new functional anatomy of language: the present issue

What sets this group of papers apart and what unifies the selection of articles as a

group? What makes the contributions special is that each paper articulates a larger scale

model than is typically presented in experimental work. In other words, the authors are

putting their cards on the table about their larger perspective – and are therefore, of course,

willing to be wrong! But, right or wrong, the broad views developed here at the very least

provide hypotheses for future research, and in many cases lay out the agenda for research

programs.

Second, the aspect that unifies these papers is that they all build on the rich literature

deriving from the neuropsychological deficit-lesion tradition – but extend the empirical

base by considering many other sources of data that have not been assessed as deeply

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–126

before. In particular, neuroimaging data are evaluated. But beyond neuroimaging, the

range of empirical approaches discussed includes clinical approaches such as electrical

stimulation mapping and hemispheric anesthetization (Boatman), dichotic listening (Scott

& Wise), and electrophysiological recordings (Scott & Wise; Indefrey & Levelt; Hickok &

Poeppel; Ullman). In this context one should also consider other recent papers that present

large scale integrative models, including papers in the volume edited by Brown and

Haagort (1999), and the reviews by Friederici (2002), Kaan and Swaab (2002), and Stowe,

Haverkort, and Zwarts (in press). These papers, too, bring an empirically broad

perspective to the task of outlining brain-language models that are biologically sensible

and theoretically grounded.

The ultimate goal in this program of research is, presumably, to have theoretically

precise, computationally explicit, biologically grounded explanatory models of the human

brain’s ability to comprehend and produce speech and language. Can we succeed with this

goal? Certainly not, at least in the near future. The goal here is, as Scott and Wise (this

volume) state it, “unashamedly phrenological” – which is a necessary step. In some sense,

this should be seen as an intermediate problem that forms the basis for subsequent

research, in which, once the functional anatomy has been specified, the detailed

computational properties of specific brain areas are explored, much as is done in other

areas of inquiry, say vision research.

Three of the contributions focus primarily on speech perception, but it is noteworthy

that all three also consider the interface of the speech perception system with other aspects

of the linguistic computational system and consider the broad implications for the

functional anatomy of language. Scott and Wise’s refreshingly opinionated perspective on

speech perception reviews neuroimaging, deficit-lesion, and dichotic listening data to

develop a model of speech perception and auditory cognition that is also closely informed

by animal data on auditory cortex. They approach the problem from the perspective of the

acoustic complexity of speech. Their article plays an important role in that they raise a

number of important issues concerning speech theories, the relevance of the concept

‘phoneme’, the importance of the syllable as a processing unit, and the problems inherent

in the deficit-lesion method.

Boatman’s article presents a unique approach not otherwise discussed, namely data

from electro-cortical stimulation mapping and hemispheric anesthetization (Wada test).

Based on patient work, Boatman attempts to functionally fractionate the STG, and the

research converges nicely with Scott and Wise’s approach (although there are

disagreements in detail, for example concerning the speech specificity of the left STG

and superior temporal sulcus (STS)). Moreover, both papers are careful to connect to the

animal literature. Roughly, Boatman argues (i) that left middle and posterior STG are the

substrate for acoustic-phonetic processing, (ii) that STG, left inferior frontal gyrus, and

inferior parietal cortex support phonological decoding, and (iii) that the interface with

lexical semantics is provided by more broadly distributed areas. Her review of

hemispheric anesthetization data suggests bilateral involvement for speech, a conclusion

that is also supported by Scott and Wise as well as Hickok and Poeppel.

Hickok and Poeppel further develop a neuroanatomic model first discussed in Hickok

and Poeppel (2000) that takes its central idea from the literature on cortical visual

pathways. Based on imaging, deficit-lesion, and electrophysiological data they argue that

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–12 7

there are, broadly speaking, two pathways that are relevant to speech perception and

language processing more broadly conceived, a ventral and a dorsal pathway dealing with

auditory comprehension and auditory – motor interaction, respectively. The framework

attempts to bridge gaps between domains of inquiry including speech perception and

production, language development, sensory – motor integration, and verbal short-term

memory. The proposal is closely related to the conceptualization of pathways presented

both by Scott and Wise and by Boatman. Indeed, there are important convergences and

disagreements among these papers. As a collection, these three articles on speech

perception proper, together with Indefrey and Levelt’s article on production, highlight that

research on speech is an area of inquiry in which there is striking convergence across

recording methods, labs, and experimental approaches. The disagreements (the specificity

of left STG and STS; the precise contribution of the right hemisphere areas; the nature of

the parallel pathways; the locus of the auditory – motor interface) are of a level of detail

and specificity that imply considerable agreement about the fundamentals.

Returning to production, Indefrey and Levelt develop a detailed model of the functional

anatomy of production based on imaging and electrophysiological data. In an extensive

and inclusive meta-analysis of imaging studies they argue for a specific functional

anatomy (that converges surprisingly well with the three perception articles) as well as a

time-line for speech production. Their review reveals the important role for speech

production of the STG, and, crucially, clearly implicates the MTG.

The important role of the MTG identified by Indefrey and Levelt is also highlighted in

the two deficit-lesion papers by Dronkers et al. and Damasio et al. Dronkers and

colleagues report the results of a deficit-lesion study involving 64 patients with left- and

eight with right-hemisphere damage. Their analysis of the neuropsychological data

implicates MTG in word-level comprehension, the anterior STG in the construction of

basic phrase structures, posterior STG and inferior parietal cortex in auditory short-term

memory, and anterior (non-Broca) inferior frontal gyrus (Brodmann’s areas 46, 47) in

working memory functions. Importantly, based on the test instrument they have used

(CYCLE), canonical Broca’s area is not implicated. These data, again, converge in

important respects with the conclusions drawn by Indefrey and Levelt on production data

as well as the models articulated by the papers on speech. For example, in addition to the

critical role of MTG for word-level processes there is clear involvement of the anterior

STG (see, e.g. Scott & Wise) as well.

Damasio et al.’s contribution is methodologically closely related to Dronkers et al.’s

work. Here, too, a large number of deficit-lesion cases are examined (84 left- and 55 right-

hemisphere damage cases are analyzed). Also, both articles are concerned with developing

the next generation of lesion reconstruction approaches. In contrast to Dronkers et al.,

Damasio et al. are interested in a particular cognitive operation, recognition and naming of

concepts. Therefore, in addition to lesion data, Damasio et al. present imaging data (PET)

investigating recognition and naming of categories such as persons, animals, and tools.

Their data suggest a segregation of function, primarily in inferior temporal cortex, such

that the substrate supporting these conceptual distinctions differs, ranging from anterior

temporal lobe (persons) to posterior temporo-occipital cortex (tools).

Interestingly, Damasio et al. are also concerned with the important role that

experimental tasks play in the generation of neurobiological results. In general,

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–128

the nature of tasks and their problematic role, i.e. the role of ecological validity, is a

theme that is also discussed by Scott and Wise and by Hickok and Poeppel, suggesting that

it is an issue that worries practitioners with regard to the interpretation of their

neurobiological data.

Ullman presents his Declarative-Procedural model, a neurocognitive approach that is

designed to make contact between aspects of language and non-linguistic neural functions

such as declarative and procedural memory. An extensive review of lesion and imaging

data (as well as numerous other sources of data, including data on sex differences as well

as normal and pathological language acquisition) suggests that there is a close relationship

between the brain systems that drive the different forms of memory and those linguistic

computations that have ‘declarative’ (e.g. retrieval of lexical items) or ‘procedural’

(e.g. online computation of an inflected form) requirements. Ullman’s paper makes a

notable effort to connect the model to numerous aspects of systems neuroscience.

A brief tour through the major figures of the papers provides a visually striking

overview of the research. To obtain a quick and intuitive but very useful impression of the

work that is reviewed, summarized, and integrated in the papers in this volume, it is

helpful to turn straight to the images. We recommend that the reader study Fig. 1 in Scott

and Wise, Fig. 4 in Boatman, Figs. 1 and 2 in Hickok and Poeppel, Figs. 4 and 5 in Indefrey

and Levelt, and Fig. 9 in Damasio et al. Cumulatively, these images summarize an

enormous range of data, approaches, and theoretical predilections. Strikingly, the figures,

as a collection, illustrate the emergence of consensus.

5. Trends, convergence, and the emergence of a new model

It is impossible – and perhaps even incoherent – to attempt to summarize the set of

findings and models discussed in these articles. Nevertheless, there are several convergent

observations and hypotheses.

1. Broca’s area and Wernicke’s area are no longer viewed as monolithic or homogeneous

pieces of tissue. Rather, there are attempts to define, subdivide, and functionally

interpret both of these cortical regions. It is particularly noteworthy that no paper in the

present collection focuses on or attributes any special role to Broca’s area or

Wernicke’s area.

2. The fractionation of STG and its functional role is a very active area of imaging

research, with a major proposal being that functionally and anatomically distinct

parallel dorsal and ventral pathways originate in the STG.

3. There is a dramatic increase in attention to cortical areas outside the traditional

perisylvian language zone. Some of these regions include the middle and interior

sectors of the temporal lobe for its role in word-level processes, the anterior STG for its

role in the construction of phrases as well as intelligibility, and subcortical structures

(basal ganglia, cerebellum) for their role in linguistic computation.

4. There is increasing interest in the relation between perception/comprehension and

production and the potential role of posterior temporal and inferior parietal cortex in

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–12 9

the auditory – motor interface. There is a hypothesis that a Sylvian parieto-temporal

area (Spt) drives an auditory – motor interface, as well as proposals that areas 7 and 40

perform subroutines of verbal working memory.

5. The right hemisphere, the ugly step-hemisphere in brain-language models, is being

rehabilitated. There is broad consensus that, at least in speech perception, the right

temporal lobe plays an important role, and, more generally, one of the main

consequences of imaging research has been to highlight the extensive activation of the

right hemisphere in language tasks. On balance, a modification of the virulent left-

hemisphere imperialism characteristic of the field is in order.

6. The future: where have we been and where are we going?

The history of neurolinguistics looks like any other domain of the natural sciences. That

is, initially, rather coarse distinctions sufficed to account for the basic phenomena of

interest, but as knowledge in the biological and the linguistic domains accumulated,

numerous necessary modifications were made as the concepts in both domains became

finer-grained. An (admittedly naive) optimistic perspective therefore suggests that all is

well and that progress in linguistics and psycholinguistics and parallel progress in

neuroscience will continue to lead to periodic improvements and small changes in our

models. But will real understanding and explanation emerge? A less charitable view is that

the progress of ‘normal science’ will not suffice and that paradigmatic changes are in the

air – or should be. Specifically, the question of ‘unification’ has been raised (e.g.

Chomsky, 2000). Will there be a unified account, motivated by disciplinary and

interdisciplinary considerations, which captures both the biological realities and the

generalizations accounted for by linguistic research?

A unified account of brain and language is likely to require two things. First, we

need linking hypotheses that bridge the conceptual and technical apparatus of

linguistics with the machinery of neuroscience, i.e. linking hypotheses that bridge

concepts such as ‘distinctive feature’ or ‘phrase’ or ‘lemma’ with the mechanisms of

neurobiology (e.g. concepts such as ‘receptive field’ or ‘oscillation’). Second, a deeper

requirement, going beyond the postulation of bridging or linking hypotheses, is the

possibility of conceptual change in both domains of inquiry. Will linguistics and

biology have to modify their conceptual apparatus in order to develop unified

explanatory accounts? This is a distinct possibility. Beyond a functional anatomy of

language, the modest beginnings of which are sketched out here, we will also require a

functional physiology of language, an area that we have not addressed. Future,

increasingly unified accounts will present a computationally explicit connection

between linguistics, psycholinguistics, and neurobiological mechanisms. The present

work simply represents a very first step in that direction by attempting to identify the

brain areas that must be included for further investigation.

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–1210

Acknowledgements

During the preparation of this manuscript, D.P. was a Fellow at the Wissen-

schaftskolleg zu Berlin and was supported by NIH R01DC05660. G.H. was supported by

NIH R01DC0361.

References

Altmann, G. T. M. (Ed.), (2002). Psycholinguistics. New York: Routledge.

Amunts, K., Schleicher, A., Burgel, U., Mohlberg, H., Uylings, H. B., & Zilles, K. (1999). Broca’s region

revisited: cytoarchitecture and intersubject variability. Journal of Comparative Neurology, 412(2), 319 – 341.

Anderson, J. M., Gilmore, R., Roper, S., Crosson, B., Bauer, R. M., Nadeau, S., Beversdorf, D. Q., Cibula, J.,

Rogish, M., III, Kortencamp, S., Hughes, J. D., Gonzalez Rothi, L. J., & Heilman, K. M. (1999). Conduction

aphasia and the arcuate fasciculus: a reexamination of the Wernicke-Geschwind model. Brain and Language,

70, 1 – 12.

Bogen, J. E., & Bogen, G. M. (1976). Wernicke’s region – where is it? Annals of the New York Academy of

Sciences, 280, 834 – 843.

Broca, P (1861). Remarques sur le siège de la faculté du langage articulé; suivies d’une observation d’aphémie

(perte de la parole). Bulletins de la Société Anatomique (Paris), 6, 330 – 357, 398 – 407.

Broca, P. (1865). Sur le siège de la faculté du langage articulé. Bulletins de la Société d’Anthropologie, 6,

337 – 393.

Brown, C. M., & Haagort, P. (Eds.), (1999). The neurocognition of language. Oxford: Oxford University Press.

Chomsky, N. (2000). Linguistics and brain science. In Y. Miyashita, A. Marantz, & W. O’Neil (Eds.), Image,

language, brain. Cambridge, MA: MIT Press.

Damasio, A. R. (1992). Aphasia. New England Journal of Medicine, 326, 531 – 539.

Damasio, H., & Damasio, A. R. (1980). The anatomical basis of conduction aphasia. Brain, 103, 337 – 350.

de Bleser, R., Cubelli, R., & Luzzatti, C. (1993). Conduction aphasia, misrepresentations, and word

representations. Brain and Language, 45, 475 – 494.

Dronkers, N. F., Redfern, B. B., & Knight, R. T. (2000). The neural architecture of language disorders. In M. S.

Gazzaniga (Ed.), The new cognitive neurosciences (pp. 949 – 958). Cambridge, MA: MIT Press.

Friederici, A. (2002). Towards a neural basis of auditory sentence processing. Trends in Cognitive Sciences, 6,

78 – 84.

Galaburda, A., & Sanides, F. (1980). Cytoarchitectonic organization of the human auditory cortex. Journal of

Comparative Neurology, 190, 597 – 610.

Geschwind, N (1965). Disconnexion syndromes in animals and man. Brain, 88, 237 – 294, 585 – 644.

Hickok, G. (2000). Speech perception, conduction aphasia, and the functional neuroanatomy of language. In Y.

Grodzinsky, L. Shapiro, & D. Swinney (Eds.), Language and the brain (pp. 87 – 104). San Diego, CA:

Academic Press.

Hickok, G., Erhard, P., Kassubek, J., Helms-Tillery, A. K., Naeve-Velguth, S., Strupp, J. P., Strick, P. L., &

Ugurbil, K. (2000). A functional magnetic resonance imaging study of the role of left posterior superior

temporal gyrus in speech production: implications for the explanation of conduction aphasia. Neuroscience

Letters, 287, 156 – 160.

Hickok, G., & Poeppel, D. (2000). Towards a functional neuroanatomy of speech perception. Trends in Cognitive

Sciences, 4, 131 – 138.

Kaan, E., & Swaab, T. (2002). The brain circuitry of syntactic comprehension. Trends in Cognitive Sciences, 6,

350 – 356.

Lichtheim, L. (1885). On aphasia. Brain, 7, 433 – 484.

Mohr, J. P., Pessin, M. S., Finkelstein, S., Funkenstein, H. H., Duncan, G. W., & Davis, K. R. (1978). Broca’s

aphasia: pathological and clinical. Neurology, 28, 311 – 324.

Poeppel, D. (1996). A critical review of PET studies of language. Brain and Language, 55, 317 – 351.

Stowe, L., Haverkort, M., & Zwarts, F (in press). Rethinking the neurological basis of language. Lingua.

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–12 11

Tanabe, H., Sawada, T., Inoue, N., Ogawa, M., Kuriyama, Y., & Shiraishi, J. (1987). Conduction aphasia and

arcuate fasciculus. Acta Neurologica Scandinavica, 76(6), 422 – 427.

Wernicke, C. (18741977). Der aphasische symptomencomplex: eine psychologische studie auf anatomischer

basis. In G. H. Eggert (Ed.), Wernicke’s works on aphasia: a sourcebook and review (pp. 91 – 145). The

Hague: Mouton.

Whitaker, H. A., & Etlinger, S. C. (1993). Theodor Meynert’s contribution to classical 19th century aphasia

studies. Brain and Language, 45(4), 560 – 571.

Wise, R. J. S., Scott, S. K., Blank, S. C., Mummery, C. J., Murphy, K., & Warburton, E. A. (2001). Separate

neural sub-systems within “Wernicke’s area”. Brain, 124, 83 – 95.

David Poeppel a,*, Gregory Hickok

b

a Department of Linguistics, Department of Biology,

University of Maryland College Park, 1401 Marie Mount Hall,

College Park, MD 20742, USA

E-mail address: [email protected]

* Corresponding author. Tel.: þ1-301-405-1016; fax: þ1-301-314-5052.

D. Poeppel, G. Hickok / Cognition 92 (2004) 1–1212