abstract grammatical processing in broca's area: convergent

ABSTRACT GRAMMATICAL PROCESSING IN BROCA’S AREA: CONVERGENT EVIDENCE FROM FMRI AND INTRACRANIAL ELECTROPHYSIOLOGY

Ned T. Sahin 1,2,*, Eric Halgren 3,4, Istvan Ulbert 2,5, Don Schomer 6, Julian Wu 7, Anders Dale 2,3 and Steven Pinker 1

(1Department of Psychology, Harvard University, Cambridge, MA, USA; 2Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, USA; 3Department of Radiology, University of California at San Diego, La

Jolla, CA, USA; 4INSERM E9926, Marseilles, France; 5Institute for Psychology of the Hungarian Academy of Sciences, Budapest 1450, Hungary; Departments of 6Neurology and 7Neurosurgery, Beth Israel Deaconess Medical Center, Boston, MA, USA)

ABSTRACT

Though Broca’s Area has long been implicated in grammatical computation, neither aphasiological nor neuroimaging studies have confirmed such a connection, because syntactic processing is often confounded with working memory, articulation, or semantic selection. Inflectional morphology (grammatical combination inside the word) offers potentially simpler paradigms. We asked 18 subjects to inflect words silently in the past tense or plural, or to read them verbatim (tasks with minimal demands on working memory), and measured brain activation with fMRI and intracranial electroencephalography (iEEG). Subtraction of the fMRI signal for reading from that for inflection should index processes involved in inflecting a form, holding constant word recognition and articulatory planning. Such subtractions suggest that left BA 44/45 (Broca’s), 47 (pars orbitalis of the inferior frontal gyrus), and medial supplementary motor area (SMA) are engaged by inflection. The design also varied whether the forms were inflected by an overt suffix (the dogs; they walked) or no suffix (the dog; they walk), whether they were nouns or verbs, and whether they were regularly or irregularly inflected. Subtraction of activity for zero-inflection from that for overt inflection (which should index the manipulation of phonological content) implicated left anterior insula, BA 44/45, 47, and SMA. Subtraction of activity during reading from that during zero-inflection (which should index only the manipulation of inflectional features, since the two tasks require identical outputs) did not involve the insula or BA45 but activated distinct regions of BA44 and 47, in concert with a premotor region, implicating these regions in the manipulation of abstract grammatical features independently of articulatory demands. These patterns were largely similar in nouns and verbs and in regular and irregular forms, suggesting that these regions are involved in a central process of implementing inflectional features which cuts across word classes. Some regular-irregular differences were observed, with greater activity for irregular verbs in the anterior cingulate and SMA, possibly reflecting blocking of regular or competing irregular candidates. Major results were corroborated by single-subject fMRI and iEEG in an epilepsy patient with normal language skills whose treatment required surgical implantation of electrodes. Neuronal activity in regions overlapping with within-patient and healthy group fMRI revealed three neural generators with sensitivity to task conditions. Responses at 355 and 500 msec were highest for overt inflection and lowest for reading, and were localized superior to Broca’s area, the first more medial and the second more anterior. The third response varied with task in decay rate (rather than amplitude) in the 715-870 msec interval, and was in the IFG ventral to the core of Broca’s Area. This pattern, converging with the fMRI results, suggests at least three neural arrays sequentially involved in aspects of inflectional morphology. Results are interpreted in terms of a network of regions in left prefrontal cortex, including Broca’s Area, that are recruited for the processing of abstract morphosyntactic features and overt morphophonological content.

Key Words: morphology, iEEG, nouns and verbs, language, regular inflection, irregular inflection, grammar, neural basis of

BOLD, insula, epilepsy seizure

INTRODUCTION

Broca’s Area may be the most widely known region of the brain, and its discovery in 1861 as a major component of language ability marks the beginning of modern neuropsychology. Nonetheless, after more than a century, neither the function of Broca’s area nor the neural substrates of language are well understood. Using two methods, this paper measures the neural activity underlying a simple linguistic task, and presents evidence that Broca’s Area is (among other things) central to abstract grammatical computation.

Relation of Broca’s Area to Grammatical Processing and Other Functions

Early in the study of the aphasias, patients with

lesions to Broca’s Area were observed to be impaired in speech production, especially in the omission or misuse of inflections and other closed-

class morphemes, but seemingly intact in speech comprehension (Broca, 1861). This led to the view that that Broca’s Area handled expressive as opposed to receptive language (Wernicke, 1874), (Geschwind, 1970), and became a central assumption of the Wernicke-Geschwind model of language organization in the brain. It was subsequently challenged by the demonstration that Broca’s aphasics were unable to comprehend sentences whose meanings could not be accessed by simple word order but only by an analysis of grammatical structure (e.g. the boy that the girl is chasing is tall) (Zurif et al., 1972; Caramazza and Zurif, 1976). This led to the hypothesis that Broca’s area subserves the computation of grammar, both receptive or expressive (Caramazza and Zurif, 1976), (for review, see Dronkers, 2000). The hypothesis, if true, would play a major role of our understanding of language, because grammatical computation, by combining a finite set of memorized elements into novel sequences, is what gives language its infinite expressive power. It is the ability that most clearly differentiates human

2 Ned T Sahin and Others

Preprint: Confidential. Submitted (summer, 2004) to Cortex special issue: Broca’s Area and the Ventral Premotor Cortex.

language from animal communication (Nowak et al., 2000; Fitch and Hauser, 2004; Pinker and Jackendoff, in press), so identifying its neural mediator is central to the study of language and human cognition in general.

This equation of Broca’s area with grammar was challenged by (Linebarger et al., 1983a), who showed that classic Broca’s aphasics could make well-formedness judgments that hinged on subtle aspects of grammatical knowledge, such as the rules governing prepositions, particles, and other closed-class morphemes (* She went the stairs up in a hurry ). Broca's aphasics' ability to recognize that a sentence needs certain closed-class morphemes, combined with an inability to use those morphemes to understand the sentence, has been called the "syntax-there-but-not-there" paradox, (Linebarger et al., 1983b; Cornell et al., 1993). One possible resolution is that only a circumscribed subset of grammar is computed in Broca’s area and impaired by Broca’s aphasia, such as the building of tree structure or the linking up elements in different parts of the sentence that must refer to the same entity, as in anaphora and the binding of traces (Cornell et al., 1993). For example, Grodzinsky and colleagues argue the manipulation of traces is the only thing computed in Broca’s Area, and that Broca’s Aphasia results from deletion of the traces (Grodzinsky, 1986a, b, 2000). Another is to suggest that Broca’s area is involved in certain aspects of the on-line processing of grammar but not underlying grammatical knowledge (see (Linebarger et al., 1983a) Zurif and Grodzinsky (1983). Yet another is to underscore the heterogeneity of deficits labeled “Broca’s aphasia,” a consequence of the uniqueness of individual patients’ lesions and the complexity and variation of the language circuitry of the brain (Berndt and Caramazza, 1999).

The recent advent of functional neuroimaging to complement lesion studies has pinpointed neither the function of Broca’s area nor the substrate of grammatical computation. A set of studies by (Stromswold et al., 1996; Caplan and Waters, 1999) reinforced an association between the two. They presented subjects with sentences containing identical words and the same kind of meaning but varying in syntactic complexity, such as relatively easy right-branching sentences (e.g. The child spilled the juice that stained the rug) and more difficult center-embedded sentences (e.g. The juice that the child spilled stained the rug) Regional cerebral blood flow (rCBF), measured by PET, showed significant differences only in BA 44, the pars opercularis of Broca’s area. This finding does not, however, show that Broca’s area is responsible for grammatical knowledge and processing. The two kinds of sentences are, in many theories of grammar, structurally similar or identical, and differ only in the demands they make on working memory in sentence parsing, such as how long a person has juice in memory before encountering the predicates, such as enjoy or stain or both, that indicate its semantic role. In a recent review, Kaan and Swaab (2002) note that Broca’s area shows

increased activity not only to contrasts such as right-branching versus center-embedded sentences but to sentences with ambiguous words, low-frequency words, or the need to maintain words over extended distances. They conclude that Broca’s Area is sensitive to any increase of processing load, rather than being dedicated to linguistic computation. They argue that other findings tying Broca’s area to syntax can also be reinterpreted in terms of generic processing load, including comparisons of reading words lists versus sentences, studies of the reading of Jabberwocky sentences (consisting of meaningless words in grammatical structures), and studies on the detection of syntactic errors. Kaan and Swaab argue not only against the strong hypotheses that only Broca’s Area processes syntax, and that Broca’s Area only processes syntax, but against the weaker hypothesis that Broca’s Area is systematically involved in grammatical computation at all. They conclude that “Broca’s area is only systematically activated when processing demands increase due to working memory demands or task requirements.” Similar conclusions are found in (Just and Carpenter, 1992) and (Bates and Goodman, 1997), who note that because general working memory demands increase in comprehending complex sentences, the seeming grammatical difficulties of Broca’s aphasics could be attributable to their inability to store information temporarily.

Since grammar is a mechanism that relates sound to meaning, many grammatical differences will necessarily correlate with differences in meaning, so attempts to tie Broca’s area to grammar may be confounded by the cognitive demands of processing semantics. For example, (Thompson-Schill et al., 1997) argue that generalized “selection demands” increase in complex sentences, potentially confounding the signal from grammatical processing. In three tasks (generating a verb semantically associated with a presented noun, judging the consistency of a picture and a word, and judging the semantic similarity of a word to a target), Thompson-Schill et al. varied the degree to which the response competed amongst alternatives (e.g., “hand” requires selection from among a larger set of verb associates than does “gun”). Broca’s region was preferentially active for high selection conditions, and crucially was not activated by a task with low selection demands. They conclude that the Inferior Frontal Gyrus (which contains Broca’s area) is involved in selecting from among semantically specified items, though not in simply retrieving them or in grammatical processing per se.

The potential confound between syntactic complexity and semantic selection is difficult to eliminate even from studies that are carefully designed to focus on syntax. Using fMRI, (Embick et al., 2000) compared brain activity when subjects detected words that were misplaced in a sentence (e.g. John drove to store the in a very fast car two weeks ago), which presumably engages syntactic processing, and when they detected words that were merely misspelled (John drove to the store in a

Grammar in Broca’s Region: fMRI + iEEG 3


very fasvt car two weeks ago), which involve only orthographic and phonological processing. Classic language areas were active in both conditions, but the greatest difference was seen in Broca’s Area, leading the authors to conclude “that Broca's area is specifically involved in syntactic processing.” Yet it is still possible that only the sentences with syntactic anomalies trigger the listener to re-analyze the sentence, a process that may involve assuring that the revised sentence is consistent with a specific interpretation, thus activating the semantic system as well.

Yet another potential confound is articulation (Wise et al., 1999) and articulatory planning (Dronkers, 1996), long associated with Broca’s area on both anatomical grounds (proximity to the mouth and face region of the motor cortex) and aphasiological evidence (dysarthria and dyspraxia of speech are common symptoms in the family of syndromes known as Broca’s aphasia). It was specifically to avoid contamination of grammatically induced activity in Broca’s area by sub-vocal rehearsal (Smith et al., 1998) that Caplan (2000) had subjects repeat an unrelated word during their sentence comprehension task, with some danger of altering subjects’ normal mode of language processing.

Syntax versus Morphology as a Domain for Studying the Neural Bases of Grammatical Processing

In sum, while a long history of

neuropsychological and brain mapping has circumstantially implicated Broca’s Area in the processing of grammar, it has been seen as inconclusive in separating grammatical processing from more general demands such as working memory, semantic processing and selection, and articulatory demands. We suggest that many of these problems come from the focus on syntax in most of the neuroimaging experiments. Syntax is not the only component of combinatorial grammar. Traditionally grammar is divided into syntax – the combination of words into phrases and of phrases into sentence—and morphology—the combination of morphemes and simple words into complex words. Morphology in turn is often divided into derivation, which generates new words (learn + -able learnable; an elbow to elbow; mice + bait mice-bait), and inflection, which modifies a word according to its role in a sentence or discourse context (walk + -ed walked; hawk + -s hawks). These processes are, like syntax, highly productive; indeed, in many languages they show greater complexity than syntax. In Turkish, for example, each verb comes in millions of inflectional forms, and rules must be attributed to speakers to circumvent the combinatorial explosion of memory entries and learning episodes that would be required by sheer memorization. In languages with complex morphology, syntax often plays a subsidiary role, and speakers have considerable

freedom in ordering words, with thematic relations conveyed by inflections for case and agreement.

Though most studies of the neural bases of grammar have examined syntax, there may be advantages to examining morphology. Whereas syntax involves relationships across words, which are spread out in time, often by many seconds, morphology takes place within a single word, often a single syllable, and therefore places few of the demands on working memory that have confounded neuroimaging studies of syntax. The semantics of inflectional morphology can also be relatively simple, sometimes involving the addition of a single semantic feature such as “plural” or “past-tense.” The grammatical component of an act of morphological processing can be isolated relatively cleanly from the input-output components (such as recognizing and retrieving a word, preparing it for articulation, and articulating it) by comparing the task of inflecting a word (e.g., seeing walk and saying walked) with the task of repeating it verbatim (e.g., seeing walk and saying walk).

The inflectional process can be further subdivided into two component subprocesses, sometimes called morphosyntax and morphophonology. The first is the manipulation of features such as tense, person, number, and gender, generally in response to demands by syntax, as when a clause is obligatorily tensed (compare, e.g., I want him to leave/*that he left and I think that he left/*him to leave), or a subject must agree with a verb. The second is the encoding of such features into audible phonological signals. The difference between these subprocesses is made clear in cases of zero-morphology. For instance, an English verb stem (e.g., walk) is not modified by the addition of a suffix in the present tense for first and second persons and for third person plurals (I, you, we, they walk). Knowing that such an unmarked form is called for by these combinations of tense, number, and person is part of morphosyntax, and involves the pure manipulation of abstract features, with no phonological consequences. Knowing that suffixed forms are called for in the past tense (walked) and third person singular present tense (walks) involves both the manipulation of morphosyntactic features and, in addition, the execution of a process that appends the suffix –ed or –s to the stem.

A final advantage in using inflectional morphology to dissect grammatical processing is that the morphophonological process can in turn be dissected into two distinct kinds of cognitive operations. With regular verbs, such as walk-walked and hawk-hawks, a suffix is predictably applied to the stem, even to novel stems, as in the neologisms spammed and moshed, that have never been heard in the company of that suffix before and hence could not have been memorized in their past-tense forms. With irregular verbs, such as bring-brought, ring-rang, and fling-flung, no consistent phonological change is applied, and the inflected form must be retrieved from lexical memory. Under the assumption that regular forms generally require the concatenation of morphemes in real time, whereas irregular forms require lookup from



memory (the “words and rules” theory; (Pinker, 1991, 1999; Pinker and Ullman, 2002), a comparison of the two can reveal the respective neural substrates of grammatical combination and lexical lookup. Alternatively, there are theories that attribute both regular and irregular inflection to a single process, either computation by a battery of rules (including minor rules that generate irregular patterns such as –ing -ung; (Halle and Mohanan, 1985; Chomsky and Halle, 1968/1991; Albright and Hayes, 2003)) or lookup from a connectionist associative memory (Rumelhart and McClelland, 1986; Joanisse and Seidenberg, 1999; McClelland and Patterson, 2002). A failure to find any difference in the neural substrates of regular and irregular inflection would support such single-mechanism alternatives.

As mentioned, inflectional errors are some of the longest-documented and most apparent deficits in Broca’s aphasics (Goodglass, 1973; Dronkers et al., 1999; Friedmann and Grodzinsky, 1997), but there have been virtually no neuroimaging studies focusing on inflectional morphology, especially in production (other than a few, reviewed below, that compare regular to irregular inflection). In this study we use the more tractable but still combinatorial system of inflectional morphology to investigate the neural substrates of abstract grammatical processing, and the possible role of Broca’s area in such processing. Subjects read words on a screen and either repeated them verbatim or inflected them for tense or number while brain activity was recorded by one of two methods (functional Magnetic Resonance Imaging or Intracranial Electroencephalography). The simple task spares subjects from having to hold words of different lengths in working memory, and since the item being manipulated is a single word, one can control for low-level features such as length, syllables, frequency, pronounceability, and concreteness, in a way that would be prohibitive for an entire sentence.

Different conditions potentially can isolate the neuropsychological components involved in an act of grammatical processing. When people read a word and repeat it verbatim, the minimum processes include reading and recognizing the word, looking up its phonological representation, preparing it for articulation, and articulating it. When people inflect a word in the third person plural or another context calling for a zero-marked form (e.g., they see walk in the context “Everyday they ….” and say “walk”), they must do all these things and in addition determine that the linguistic context calls for leaving the form unchanged, a simple instance of morphosyntactic processing. When people inflect a word in the past tense (e.g., they see walk in the context “Yesterday they …” and say “walked”), they must do all the components of both tasks previously described and, in addition, execute some operation that results in a phonologically different output: either looking up the past-tense suffix and concatenating it to the verb stem (for regular verbs) or retrieving a distinct form (for irregular verbs).

Under the simplest assumption of how psycholinguistic processes, characterized in information-processing terms, map onto patterns of neural activity, we might expect the pattern of neural activity recorded for repeating a word to be a proper subset of the activity for producing a zero-inflected form, the difference indicating the neural substrates of the computation of morphosyntactic features. Similarly, we might expect the neural activity for uttering a zero-marked form to be a proper subset of the activity recorded for uttering an overtly inflected form, the difference indicating the neural substrates of morphophonological manipulation. We note that these assumptions correspond to the “pure insertion” model of how information processing components are combined, viz., that a given component operates in the same way, and has the same distribution in the brain, regardless of which other components accompany it in a given task. That assumption may or may not be true in any given case, but it can be tested in part by seeing whether the patterns of activity recorded in different tasks really do exhibit a subset-superset relationship, as opposed to being disjoint or overlapping.

Regular and Irregular Inflectional Morphology What are the predictions about the effects of the

regular/irregular contrast? According to the words-and-rules theory, irregular forms (and any regular forms that are dependent on memory storage) should be tied to the neural substrate for lexical memory, which is often thought to be concentrated in temporal and temporoparietal regions (Damasio, 2000; Goodglass, 1993; Martin et al., 1996). Regular forms (especially those for low-frequency and novel words) should be tied to the substrate for grammatical combination, traditionally associated with circuits which include Broca’s area, other regions in the prefrontal cortex, and the basal ganglia (Ullman et al., 1997; Dronkers et al., 1999; Damasio, 1992). Many neuropsychological studies are consistent with this assignment. Patients with anomia following damage to left temporal/parietal regions are (compared to control patients) worse at producing irregular than regular verbs, produce regularization errors like swimmed (which occur when no memorized form comes to mind and the rule applies as the default), and are relatively unimpaired at generating novel regular forms like plammed (Ullman et al., 1997; Ullman et al., in press; Tyler et al., 2002b; Miozzo, 2003). Patients with agrammatism following damage to left frontal perisylvian regions show the opposite pattern: more trouble inflecting regular than irregular verbs, a lack of errors like swimmed, and difficulty suffixing novel words (Ullman et al., 1997; Ullman et al., in press). Other evidence linking anterior cortex with regular inflection and posterior cortex with irregular inflection comes from studies of inflectional priming in patients with brain damage (Tyler et al., 2002a; Marslen-Wilson and Tyler, 1997, 1998) and of event-related potentials in healthy speakers (Munte et al., 1999; Gross et al.,



1998; Penke et al., 1997; Weyerts et al., 1997). Involvement of the basal ganglia in regular inflection is suggested by the finding that Parkinson’s Disease patients have selective difficulty in inflecting regular and novel verbs, and seldom make overregularization errors (Ullman et al., 1997; Ullman et al., in press).

Neuroimaging studies on the regular-irregular distinction present a more complicated picture (Jaeger et al., 1996; Ullman et al., 1997; Indefrey et al., 1997; Rhee, 2001; Rhee et al., 2003; Beretta et al., 2003). All such studies show different patterns of activity when subjects inflect irregular and regular forms, consistent with the prediction of the words-and-rules theory that the two processes have different sets of neural substrates. In particular, all show greater overall activation for irregular than regular forms, and all show regular inflection to be more left-lateralized and irregular inflection to be more bilateral (consistent with much neuropsychological evidence that the lexicon is less lateralized than grammatical combination). Unfortunately, the respective areas associated with regular and irregular inflection differ from study to study, possibly because of methodological differences: some used PET, others fMRI; some used English, others German; some compared regular and irregular inflection directly, others first subtracted out activity during verbatim repetition of the stem (to control for differences between regular and irregular items in the stimulus sets that might be confounded with regularity per se). Some (Indefrey et al., 1997; Jaeger et al., 1996) used blocked designs in which subjects inflected regular and irregular forms in different blocks of trials, which may induce subjects to use different conscious strategies for the two kinds of verbs (Seidenberg and Hoeffner, 1998).

Moreover, there is little to no evidence that the regular-irregular distinction correlates with differences in functional neuroimaging activity between frontal and temporal-parietal regions. If anything, each of the five studies shows increased activity in left frontal regions for the irregulars. There are numerous possible explanations for the discrepancy between the neuroimaging data on the one hand and the neuropsychological and electroencephalographic data on the other. Neuroimaging studies identify the set of regions recruited in normal function, whereas lesion studies index single regions that are so necessary for a given function that the function is grossly compromised by the lesion.

Moreover, there are many reasons to expect that in normal functioning, the regular-irregular distinction does not map perfectly onto a neural distinction between grammatical computation and lexical lookup. First, both regular and irregular forms require the processing of morphosyntactic features such as “past tense” and “plural,” which originate in the syntactic representation of the sentence or in the speaker’s intentions and trigger a call for a specific inflected form; the difference is only in which of the two kinds of processes succeeds in supplying the form. Second, if, as

seems likely, regular and irregular processes are activated in a parallel race fashion (Baayen and Schreuder, in press; Pinker, 1999; Caramazza et al., 1988), both processes may operate for both kinds of forms, the difference lying only in which one terminates and which one runs to completion. Third, a strict dichotomy between whole regular and whole irregular forms may not always be appropriate. Some complex words may consist of an irregular stem with a regular suffix; this is common in languages other than English (Berent et al., 2002) and may be found in some English plurals such as leaf-leaves and house-houses; see (Senghas et al., 2004). Fourth, certain regular forms may be stored in memory -- diluting any difference from irregulars in average neural activity -- if they are high in frequency, higher in frequency than their stems, phonologically similar to irregulars, inflected with an affix which is homophonous with some other affix, or in alternation with an irregular variant (Pinker, 1999; Baayen and Schreuder, in press; Hay, 2001; Alegre and Gordon, 1999; Ullman, 1999). Fifth, even when they are computed in real time, regular forms may require at least two cycles of memory lookup, one for the phonology of the stem, another for the phonology of the past tense suffix –ed; irregular forms differ only in requiring secondary lookup of a form that is more phonologically and semantically substantial and less overlearned than the regular suffix. Sixth, irregular verbs, for their part, may require not just activation of the lexicon but the control processes that guide access to the lexicon (often linked to frontal regions such as Brodmann’s area 47 and other regions of lateral prefrontal cortex (PFC) (Kerns et al., 2004b)). These control processes must send out a search query for the form with the intersecting specification of the lexical item and the inflectional feature (e.g., to bring ∩ past-tense) and inhibiting partial or false matches from overlapping memory items (e.g., for brought, interference from drank and sprung). Seventh, irregular inflection requires not just retrieval of the irregular form but suppression or “blocking” of the regular rule, to prevent overregularizations such as bringed (Marcus et al., 1992; Pinker, 1999; Ullman, 1999). Though the neural substrates of blocking are unknown, they may overlap with cortical circuits that effect cognitive inhibition and control. These may include the anterior cingulate cortex (ACC), which has been implicated in the signaling of conflict situations, various regions of prefrontal cortex, which resolve the conflict (Miller and Cohen, 2001; Kerns et al., 2004a), and regions dorsal to classic ACC such as medial supplementary motor area (SMA), which has been implicated in error and conflict signals in trials with fixed stimulus-response mappings (Holroyd et al., 2004).

All these considerations suggest that while there are may be differences in the processing of regular and irregular forms for neuroimaging to reveal, they may not be restricted to a simple distinction between anterior and posterior regions, and that considerable design complexity may be needed to



tease apart the component processes for each kind of inflection. The present study is a first step in this project: it uses an event-related rather than a blocked design (to minimize the use of ad hoc strategies for regular and irregular forms), examines the inflection of both nouns and verbs, and examines the regular-irregular difference in the context of a larger set of variables designed to identify the processing components that regulars and irregulars share in addition to the ones on which they differ.

Nouns versus Verbs Another variable explored in the present study

is the distinction between nouns and verbs, which bears on the extent to which grammatical processing is spatially localized or distributed in the brain. The failure to find any region of the brain that is consistently associated with grammatical processing had led to the hypothesis that such processing is widely distributed across the brain, perhaps taking place in the same regions in which the words being modified are stored, and thereby obliterating any principled distinction between lexicon and grammar in the brain (e.g., (Bates and Goodman, 1997)). This alternative, loosely associated with connectionist approaches, would contrast with a more traditional box-and-arrow view in which words, regardless of where they are stored, are shunted to a central grammatical processor for inflection or combination with other words. This can be examined by comparing the inflection of verbs and nouns.

It is controversial whether nouns and verbs have differing neural substrates and if so whether the differences come from grammatical category per se or from other features confounded with the categories. Caramazza and colleagues have found patients selectively impaired on verbs or on nouns, including non-words (Caramazza and Hillis, 1991; Shapiro and Caramazza, 2003), as well as selective disruption of verbs during transcranial magnetic stimulation (TMS) disruption of left inferior prefrontal cortex (Shapiro et al., 2001; See also Cappa et al., 2002). They conclude that verbs are more concentrated in frontal neural regions, and nouns more concentrated temporal-lobe regions (Caramazza and Shapiro, in press). In contrast, Pulvermuller and colleagues have measured ERPs during reading (1996) and lexical decision (1999a) of nouns and verbs, and while they found category differences in similar locations (nouns near visual areas and verbs near motor areas) they attribute the difference to statistical associations of verb semantics with motor actions and noun semantics with visualizable objects, based on the finding that when they presented action-related nouns or visualizable verbs, the differences went away (Pulvermuller et al., 1999b; See also Luzzatti et al., 2002; and Bird et al., 2000; Bird et al., 2001). Neuroimaging studies have not resolved the debate. (Perani et al., 1999) found noun-verb category differences with PET, which did not interact with concreteness, yet only found voxels

more active for verbs, none more active for nouns, leaving it unclear whether the verbs are qualitatively different from nouns or just more demanding. Tyler et al. (2001) found extensive activation they interpret as a semantic network in two noun-verb PET experiments (lexical decision and semantic categorization) but found no differences as a function of word class.

In nearly all the studies of grammatical category, subjects process single words outside of a grammatical context, such as single word repetition, picture naming, or lexical decision. Only (Shapiro et al., 2001; Shapiro and Caramazza, 2003; Tyler et al., 2004) varied morphology, and only (Shapiro and Caramazza, 2003) used a sentence context (rather than a metalinguistic task) to cue the inflection. This makes it unsurprising that the measurable difference between categories is often dominated by differences in meaning rather than abstract grammatical properties. Any difference in grammatical properties would be more likely to emerge in tasks that require the use of nouns and verbs in their differing grammatical contexts. A task that compares the process of inflecting nouns and verbs according to their linguistic context with the process of repeating a word may help to specify whether nouns and verbs differ in storage, grammatical processing, or both. If inflectional processing simply emerges from the network of associations stored with words, then the inflection of nouns and verbs should be co-localized with any separate storage areas for nouns and verbs. If there is a central grammatical processor that interfaces with the lexicon but is distinct from it, one should see a common set of loci for inflection, whether it is nouns being pluralized or verbs being inflected for tense, person, and number. Indeed, such differences might be found even if nouns and verbs are stored in the same locations: after being retrieved, they may be processed in different areas to prepare them for their different grammatical roles in the sentence.

The present study, then, seeks to identify the neural substrates of grammar in the abstract sense in which linguists characterize it, rather than aspects of linguistic processing that are reducible to working memory, semantics, phonology, or lexical knowledge. Specifically, the current design tests whether there are brain regions that are active in inflectional morphology regardless of whether the inflectional modification is phonologically overt or silent (They walked versus They walk), whether it requires a predictable suffix or an unpredictable vowel change (walked versus came), whether it involves a verb or a noun (walked versus hawks), and with minimal demands on working memory and semantic selection.

fMRI and Intracranial Electroencephalography The study adds a complementary technique to

fMRI, namely in-vivo recording. fMRI visualizes localized change in percentage of deoxygenated blood (since it is paramagnetic and alters local



magnetic resonance), which is a stand-in for regional flow of oxygenated blood (since this overshoots need and thus changes local saturation), which is a stand-in for metabolism in neurons (since they have no energy stores and need perfusion in real time), which is finally a stand-in for the neuronal electrical activity assumed to subserve transient information processing. The nature and fidelity of each of these couplings is largely unknown and is currently in debate, especially in humans in response to changes in cognitive load (Logothetis, 2002; Devor et al., 2003). In order to license, then, an interpretation of hemodynamic fMRI results in terms of cognitive information processing, it is desirable to demonstrate a closer correspondence between fMRI and synaptic activity.

In pursuit of this goal, we applied the experimental paradigm in a companion study which employed direct in vivo recording from electrodes implanted in the brains of epileptic patients undergoing clinical electrophysiology. Such patients are eligible for surgical resection of the brain regions implicated in their epilepsy when non-invasive techniques fail, and before such surgery direct electrical recordings may be taken during seizures over several days to localize the onset of their seizures. The recommendation for iEEG monitoring, the targets of the iEEG electrodes, and the duration of monitoring are all decided upon entirely for clinical purposes and are unrelated to experimental protocols. A common clinical question is to distinguish anterior ventral temporal from posterior ventral prefrontal contributions to epileptogenesis, and this may be probed with electrodes passing through or near an intact Broca’s Area. We studied a patient with this implantation pattern, allowing us to seek converging results from single-subject fMRI, in-vivo human electrophysiology (from the same person), and group fMRI from healthy subjects. In addition to convergence with fMRI, the in-vivo technique (intracranial encephalography or iEEG) offers two advantages. Whereas fMRI smears the response to all the events in a multi-part trial, iEEG has the precise temporal resolution of traditional scalp-mount electroencephalography (such as that used in event-related potentials or ERP), and thus can reveal the neural response to each event within a trial. More generally, iEEG offers the millisecond precision required to separate the individual components of an episode of neuronal computation and relate them to the information-processing stages that underlie actual word production. The second advantage is that iEEG can localize neuronal generators unambiguously, and is limited in resolution mostly by the size of the electrode and in coverage by the location of the implant. Linked to both electrophysiology in primates and scalp electroencephalography in humans, iEEG provides a useful bridge to the spatially richer but more indirectly coupled noninvasive imaging of fMRI.

EXPERIMENT 1: FMRI

Method

Subjects

Eighteen healthy, right-handed native English speakers (7 female, 11 male) gave written consent and were paid to participate. Their mean age was 20.6 years, with a range of 18 to 25. Subjects were excluded if they had participated in more than five previous fMRI studies or an earlier version of this study. Participation was covered by Institutional Review Board approval, and data were treated according to the guidelines of the USA Health Insurance Portability and Accountability Act.

Task

The experiment employed a cued covert production task, schematized in Figure 1. The cue was a short context frame calling for a particular inflectional category: Yesterday they ___ (past tense), Every day they ____ (third person plural present tense), That is the _____ (singular), or Those are the ____ (plural). These allowed us to cue a different inflection on each trial without forcing subjects to think about metalinguistic categories such as “past tense” or to memorize arbitrary visual cues. In other trials, the subject was cued to read back the target word verbatim by the context frame repeat word: ___ . In all cases the context frame was followed by a target word, which appeared in a small phrase with the marker to (for verbs) or a/an (for nouns). The task was to produce silently the form of the target word that would fit into the blank (e.g., in response to Yesterday they ___ …… to walk, they would silently think “walked”), and then press a button. The button press was intended to keep the subject alert, warn the experimenter of waning attention or sleep, and to provide an estimate of reaction time (though these estimates are not presented in this paper). Subjects used only the left hand to press the button, to assure that the resulting motor activity would not be confounded with language-related activity in the left hemisphere. The marker (to or a/an) was included to inhibit a strategy of simply concatenating the target word to the context frame, which would work for two thirds of the trials (Zero-Inflect and Read) while on the other trials causing the subjects to experience an anomaly response (which strongly affects fMRI signals). The requirement to extract the target word from the phrase also helped to equalize the task steps and difficulty across the conditions. Since the markers are presented on all trials, their effects should disappear in subtractions of one condition from another.



Design The experiment had a 2 x 2 x 3 factorial design:

Grammatical Category (Noun/Verb), Regularity (Regular/Irregular) and Task (Overt-Inflect/Zero-Inflect/Read). For verbs, the Overt-Inflect condition corresponded to the frame Yesterday they ___, which calls for a past-tense form, either one with the regular suffix –ed or an irregular form. The Zero-Inflect condition corresponded to the frame Every day they ___, which calls for the third person plural present tense, which in English happens to have no phonologically overt marking. (Some linguistic theories posit a silent “zero morpheme” to preserve the idea that all inflected forms are suffixed.) For nouns, the Overt-Inflect condition corresponded to the frame Those are the ___, which calls for a plural form, either one with the regular suffix –s or an irregular form. The Zero-Inflect condition corresponded to the frame That is the ___, calling for a singular noun, which in English has no phonologically overt marking. For examples of each condition, see Figure 1.

Note that though the inflectional categories calling for zero-marked forms are phonologically identical to the word’s citation form in English (the one used when a word is produced in isolation), this is not true in many languages, and there are many reasons even in English to treat them as grammatically distinct. This provides us with two minimal contrasts. The responses in the Read and Zero-Inflect conditions are phonologically identical, but the latter requires the speaker to process inflectional features in order to determine that the zero-inflected form is called for in that slot in the inflectional paradigm. The responses in the Zero-Inflect and Overt-Inflect conditions both require the processing of inflectional features, but the former requires no manipulation of

phonological material (regular suffixation or irregular substitution).

Subjects saw each word only in one of the three tasks (Overt-Inflect, Zero-Inflect, or Read). The assignment of words to tasks was random but consistent across subjects. The sequence of trials was broken into three runs, each lasting 6 minutes and 25 seconds. To increase the number of trials and hence signal quality, the entire paradigm (i.e., the three runs) was repeated three times. The order of the runs was varied across the repetitions for a given subject, and differed for the different subjects. However, the order of trials within a given run was constant across subjects.

Materials

The materials are presented in the Appendix. One hundred twenty English nouns and 120 English verbs were used as targets, 60 each with regular and irregular forms. Stimuli were selected according to a semi-automated procedure to implement several criteria simultaneously (Sahin, 2003).

First, a master database was created using Microsoft Access, incorporating raw frequency numbers from the AP newswire corpus (See (Church and Hanks, 1991)), frequency and word length values from the Brown corpus (Francis, 1982), syllable counts and subjective ratings from MRC-2 linguistic database (Coltheart, 1981), including norms of imageability and familiarity (Paivio et al., 1968). Frequency data from the Brown corpus for the stem-cluster (the stem plus all the inflected forms of the word), and for the inflected forms of a given word, were juxtaposed by an algorithm for the regulars and by hand for the irregulars, and then supplemented by the data from the other databases. To exclude nouns that were easy to misread as verbs and vice-versa, an index of noun-verb category ambiguity (ranging from –1 to

Figure 1 - Summary of experimental conditions.



1) was calculated as the difference between the frequency of the word in the assigned category (noun or verb) and the frequency of the word in the other category (verb or noun, respectively), divided by the total frequency in both categories. Values near -1 indicated words clearly in the opposite category, values near 0 indicated category-ambiguous words, and values approaching 1 indicated words clearly situated in the target category. For example, light has an F&K frequency of 306 for all noun forms, and a frequency of 72 for all verb forms. Therefore, the F&K ambiguity index for light as a noun is (72-306)/378 = -0.619, and its category ambiguity index as a verb is 0.619.

Selection and matching were accomplished in multiple passes. In English there are far fewer irregular words than regular ones, and far fewer irregular nouns than irregular verbs. Therefore the limiting factor was the availability of irregular nouns, so they were used as the starting point. The English language makes this especially problematic because only a handful of common irregular plurals undergo some stem change (men, women, children, feet, teeth, mice, and geese). These are too few to yield interpretable fMRI data, so they were supplemented by somewhat more problematic kinds of irregular plural, including compounds (e.g., grandchildren), no-change (e.g. sheep-sheep), Latin (nucleus-nuclei), Greek (phenomenon-phenomena), and regressive-voicing fricatives (wolf-wolves). (Senghas et al., 2004) present evidence that English speakers treat borrowed Latin and Greek plurals as irregular, at least in how they treat them with regard to other grammatical processes such as compounding. But it is possible that at least some speakers apply special suffix-changing rules to generate them. Worse, Senghas et al. show that English speakers treat regressive-voicing plurals as hybrids consisting of an irregular stem (e.g., wolv-) subjected to regular suffixation. These unavoidable problems decrease the likelihood of finding a regular-irregular difference in the fMRI data for the nouns.

In assembling the matched list of irregular verbs, first verbs with regular variants and verbs with extreme frequencies were eliminated. An algorithm then selected, for each irregular noun, the

irregular verbs that best matched it on a number of weighted criteria. It attempted to achieve matches on each of the variables in the database of 90% or greater, while giving greater weight to Brown-corpus frequencies and stem-cluster frequencies than to the AP frequencies, greater weight to number of syllables than to raw length, and high weight to verbs with low category ambiguity values. The process was iterated, first for those irregulars that had values in the database for the Paivio norms, then the rest, until both Irregular lists were set. Next, the algorithm iteratively selected regular forms for each irregular, aiming for phonological similarity when possible (e.g., crept/cropped, wolves/valves, bound/downed, parentheses/democracies).

The result of this process was a set of item lists whose mean log frequencies for the major variables were mostly matched (no statistically significant differences), except for the average Francis-Kucera inflected-form frequency of the Irregular compared to Regular Noun plurals, and a consistent lower average frequency for nouns than verbs (a consequence of including Greek and Latin plurals and their matched regulars). All factors used to balance the stimulus lists are shown for the example in Table I, and a subset are shown for all items in the Appendix.

Procedure Presentation of the experimental materials was

controlled by Presentation ® software (Neuro-Behavioral Systems), version 0.50. Context frames were presented on a screen as image files, adjusted to be identical in horizontal length and to subtend a visual extent on screen small enough to allow subjects to view them without scanning away from the center.

The experiment used a rapid event-related paradigm (Buckner, 1998; Burock et al., 1998), and included all trial types in all runs in a pseudo-random order. Stimulus presentation was jittered in time, with an inter-trial interval (ITI) varying in units of one-half the trial duration, or 1.75 seconds, from 0 to 10.5 seconds. During the inter-trial

Table I

Two examples of stimulus pairs, with the variables used in stimulus balancing. Separate frequency counts for inflected form and stem-cluster (all forms); category ambiguity indexes how often a word is used in target versus opposite category (noun or verb); match percentages given as ratio of natural log transformed values.

Category Inflected Form

FK Freq (Stem)

FK Freq (Infl)

AP Freq (Stem)

AP Freq (Infl) Length Syllable Familiarity Image-

ability FK Category Ambiguity

AP Category Ambiguity

Irreg Noun feet 353 283 6149 5082 4 1 504 483 1.00 1.00 Irreg Verb felt 643 356 9645 3319 4 1 611 597 0.95 0.97

Match: 91% 96% 95% 95% 100% 100% 97% 97% 98% 99% Irreg Noun children 565 355 20326 14714 8 2 593 251 1.00 1.00 Irreg Verb began 689 312 27418 15394 5 2 608 597 1.00 1.00 Match: 97% 98% 97% 100% 82% 100% 100% 87% 100% 100%



interval the fixation cross remained on the screen and the signal was recorded; the average of this signal was analyzed as the “Fixation” baseline. The Fixation trials accounted for 27% of the total duration of the experiment, which was determined to be optimal (see (Sahin, 2003)). Jittering allowed deconvolution of the event-related fMRI signal without assuming a shape to the hemodynamic response, and was optimized by the “optseq” tool of the FS-FAST fMRI analysis toolkit, run iteratively 100,000 times to pick the best sequence (Dale, 1999).

Subjects received a schematic demonstration of the task on flash cards, and, after providing consent, a simulation of the task on a computer display similar to the one they would see in the scanner. Pilot testing had revealed that people can interpret the Every day they ___ context frame as consistent with the past tense (e.g., Every day they walked), so the experimenter emphasized that it was the present tense that was intended. Subjects reported no trouble complying with this instruction. Subjects were then trained by supervised performance of the task on a standalone computer workstation in the scanner control room immediately before the scan. They performed the equivalent of one full run of the task, first speaking the correct response out loud until the experimenter was satisfied they understood the task, then silently producing the rest while the experimenter observed the button presses. The training set contained all task conditions and categories of words, using different words from those included in the actual stimulus list.

fMRI Data Acquisition

MRI data were collected on Siemens Magnetom Trio 3-Tesla whole-body system. BOLD contrast was obtained with a gradient-echo echo-planar imaging (EPI) sequence (TR = 1750ms; TE = 30ms; flip angle = 90; FOV = 200mm; base matrix = 64 x 64 (3.125 x 3.125mm)). Twenty-five axial 5.0mm slices (skip 0.5mm) were collected to cover the brain, except, in some cases the cerebellum. High-resolution structural images, for functional underlay and group co-registration and averaging, were collected with a three-dimensional (3-D) MPRAGE protocol, at 1.0 x 1.0 x 1.33mm resolution.

Stimulus rear-projection was synchronized with millisecond precision to a TTL pulse from the scanner, preventing the experimental presentation from drifting in time relative to the scanner.

fMRI Data Analysis

fMRI data processing was carried out using FreeSurfer and FS-FAST (FreeSurfer – Functional Analysis Stream) software from the Massachusetts General Hospital Athinoula A. Martinos Center for

Biomedical Imaging and Cortechs Labs, LLC (Charlestown, MA).

The T1-weighed structural images were processed through FreeSurfer to reconstruct the cortical surfaces (Dale et al., 1999; Fischl et al., 1999a; Fischl et al., 2001). These surfaces were then registered with a surface-based atlas (Fischl et al., 1999b). This registration is nonlinear and high-dimensional based on matching the cortical folding patterns of an individual to that of a group atlas. This atlas, referred to as the “sphere”, provides a high-resolution space for performing inter-subject comparisons.

All functional (EPI) data sets were motion-corrected using AFNI (Cox, 1996), spatially smoothed with a 6 mm FWHM gaussian kernel, and intensity normalized (over time and space) to a grand mean value of 1000. The functional volume of each subject was registered to the structural (T1) volume for that subject in order to align the functional with the cortical surface. The hemodynamic response function (HRF) was modeled using a gamma-variate function (Dale and Buckner, 1997) with a delay of 2.25s and a dispersion of 1.25s. The HRF amplitude for each event type was estimated at each voxel using a general linear model (GLM). Autocorrelation in the fMRI noise was accounted for by pre-whitening with a filter estimated from the residual autocorrelation function averaged over all brain voxels (Burock and Dale, 2000). Low-frequency drift was removed by including a 5th order polynomial in the GLM. Contrasts were computed as linear combinations of the HRF amplitudes (i.e., regression coefficients). These contrasts were then resampled to the common surface space (ie, “spherical space”), and group random-effects analysis was performed in this space.

Results were then back-propagated through the spherical-normalization transformation matrix and visualized on the reconstructed surface anatomy of one representative study subject, in order to associate the BOLD activations with recognizable anatomical landmarks. The significance values for each surface-intersecting voxel were displayed as false-color overlay on the anatomy, in red-yellow scale for the positive tail of the contrast, and blue-light-blue for the negative tail.

Results and Discussion

Overall Pattern of Activation in the Linguistic Tasks

Given the complex, inconsistent, and often-puzzling patterns of activation seen in previous neuroimaging studies of inflection, we begin by examining the distribution of neural activity in each of the three tasks compared to the Fixation condition (used as a low baseline), to see if the overall pattern is intelligible in the light of existing knowledge of language and the brain.

Figure 2 contrasts fMRI activation for the Overt-Inflect, Zero-Inflect, and Read tasks with the



Fixation condition. The patterns from these three independent data sets are globally similar and fit well with classical models of the organization of language functions in the brain (Geschwind, 1979; Dronkers et al., 1999; Damasio, 1992). We observe activation in bilateral primary visual cortex (low-level perception of the visual stimuli), left-lateralized posterior inferior temporal regions (recognition of visual word forms (Dehaene et al., 2002; McCandliss et al., 2003; Cohen and Dehaene, 2004)), left middle temporal cortex near Wernicke’s area (retrieval of words’ phonological representations) (see also Figure 4a), left Broca’s area and surrounding inferior prefrontal cortex (planning of articulation, grammatical computation, or both), left premotor cortex near the areas for the articulators (planning of articulation and possibly other functions (Wise et al., 1999; Toni et al., 2002)), and right motor cortex (hand area for the left-handed button press). These patterns do not isolate grammatical computation or other components of linguistic processing (especially since each contrast includes the process of mentally extracting the noun or verb from the phrase containing to or a/an), but they confirm that each of the tasks yields an intelligible signature which makes contact with the literature and adds confidence to the interpretations of finer-grained contrasts among the conditions. Two strongly activated regions are less expected from the classic aphasiological literature but have a strong precedent in language neuroimaging. The first is a medial region (more pronounced on the left side) including the medial supplemental motor area (SMA) and anterior cingulate cortex (ACC). This is a frequently observed language task region (Turkeltaub et al., 2002) which may be involved in the initiation and suppression of articulation, especially in the context of selecting an appropriate response (Kerns et al., 2004b; Kerns et al., 2004a). Also observed is activation in the left intraparietal sulcus, possibly involved in visual attention to the linguistic stimuli (Jovicich et al., 2001; Wojciulik and Kanwisher, 1999).

Grammatical Inflection as a Sufficient Activator of Broca’s Area

To home in on the neural systems metabolically

active during the processing of inflectional morphology, we contrasted fMRI activation during Overt-Inflect and Read trials (Figure 3 and Figure 4b). This contrast, which averages over nouns and verbs and regular and irregular forms, should index most of the processes involved in grammatically inflecting English words, eliminating the more peripheral components of this task such as reading, recognizing, and preparing to articulate the word. Broca’s Area (BA44/45) was strongly activated in this contrast, within a network including much of the inferior frontal gyrus and the anterior insula. The medial views indicate involvement of the Supplementary Motor Area/Cingulate region in inflection; its role will be discussed clarified in subsequent comparisons, as will the relative

deactivation, compared to the Read task, in occipito-temporal cortex and other regions (the blue areas in Figure 4b and d).

The anatomical location of Broca’s Area is not

uniformly agreed upon but here we will take it to mean Brodmann Areas 44 and 45, or the pars opercularis and triangularis of the inferior frontal gyrus. All statistical activation maps are shown at p < 0.005, uncorrected, though the activation cluster with center of gravity in Broca’s Area in Figure 3 is significant to a threshold of p < 0.00001. This is likely to survive all but the most stringent corrections for multiple comparisons.

One of the primary questions posed in the Introduction can therefore be answered, namely that grammatical inflection was indeed sufficient to activate Broca’s Area. As noted, the task did not involve syntactic movement or long-distance dependencies, and the two conditions contrasted did not vary in working memory demands, especially sentential working memory. The result

Figure 2 - Brain activity for each condition, showing strong

concordance with known language areas. Thresholded at p < 0.005. The Fixation condition is used as a low baseline to reveal all regions involved in performing the (a) Overt Inflect, (b) Zero-Inflect, and (c) Read tasks.



challenges both the strong hypothesis that Broca’s area and surrounding regions are responsible only for a specific component of syntactic processing, such as maintaining traces, and the opposite but equally strong hypothesis that it has no specifically grammatical role at all but only mediates domain-general processing resources. At least one role of Broca’s Area appears to be the combinatorial grammatical processing required by inflectional morphology.

Broca’s Area and Overt Morphophonology

To assess whether Broca’s Area and associated regions are was involved in morphophonological processing (the manipulation of the overt phonological content of inflected forms), we contrasted the metabolic activity for the Overt-Inflect trials with the Zero-Inflect trials (Figure 4c). These conditions were cued with similar context frames, and both specified a particular inflectional category; the difference was that the form elicited by the Overt-Inflect task (plural or past-tense) displays an overt morphological change in English (suffixed or replaced), whereas the form elicited by Zero-Inflect trials is phonologically identical to the base form presented on the screen. This contrast should index the retrieval of the phonological content of the regular suffix or irregular form, the concatenation of the suffix with the stem and consequent phonological and phonetic adjustments of the juncture (for regular forms), the generation of the novel phonetic material, and the readying for final articulatory output. These processes were shown to activate large parts of the entire circuit discussed in the preceding section (note that this contrast includes half the number of trials used in the preceding one, and therefore is lower in statistical power). The results verify the involvement of Broca’s area in inflectional morphology, particularly in the manipulation of the overt phonological material.

Broca’s Area and Abstract Morphosyntactic Features

To determine if Broca’s Area is sensitive to manipulation of abstract grammatical features encoded by morphology in the absence of phonological changes, we contrasted the Zero-Inflect task with the Read task. Subjects covertly produced phonologically identical responses in the two conditions, but in the Zero-Inflect condition they arrived at the response as a solution to the problem of finding the form that satisfied the set of grammatical features (tense, number, and person) specified by the linguistic context. Since the contrast does not vary articulatory output, the hypothesis that the only role for Broca’s Area in linguistic processing is the manipulation of phonological content (either as preparation for articulation or in working memory) would hypothesize no Broca’s Area activity in this contrast. In fact, Broca’s Area was strongly activated (Figure 4d).

One possible confound in this comparison is that the Zero-Inflect context frame is longer than the Read frame and, when completed, yields a short sentence. It could be argued that the contrast between the two conditions includes some grammatical processing of this short sentence. However, subjects saw each context frame 120 times (not including practice), so reading them became automatic, and subjects reported in debriefing that they had ceased reading the frames all the way through and relied on the first word, which uniquely identified the condition. Even if they had processed the full sentences in most of the Zero-Inflect trials, the syntax required no movement, little if any working memory, and little semantic content other than in the inflected word itself. The main difference between the conditions thus appears to be the manipulation of the inflectional features.

Figure 3 - Pial surface representation, based on a single subject’s brain, of the contrast in fMRI blood-oxygen level dependent (BOLD) signal between the Overt-Inflect and Read tasks. The false coloring shows the results of a random-effects analysis of 18-subjects, thresholded at p < 0.005, with the brighter colors (yellow for an Overt-Inflect advantage, light blue for a Read advantage) indicating p < 0.0001 and lower. (a) View of lateral cortex, showing significant increased activation for Overt-Inflect in Broca’s Area (BA 44 and 45), BA47, ventral premotor and a middle premotor/M1 region, and increased activation in a middle occipital region for the Read task. (b) View of medial cortex showing increased activation for the Overt-Inflect task in the medial supplementary motor area (SMA).



Figure 4 - Inflated-surface representation of fMRI contrasts aimed at partitioning the inflection task into its components. The fMRI

parameters are identical to Figure 3 but depicted on the inflated left cortex of the same subject. The anatomical labels in (a) as a guide for localizing activation clusters. Brodmann areas 44,45 and 47 as marked; precentral sulcus (PrCS) and gyrus (PrCG) are mostly premotor Area 6 while primary motor Area 4 is the most posterior portion of PreCG; supramarginal gyrus (SMG); angular gyrus (AG); subparietal sulcus (SPS). Wernicke’s area has no consensual anatomical definition, and the Visual Word Form Area (VWFA) is a recently posited functional area; their locations are shown approximately. (b) fMRI activity for the contrast Overt-Inflect > Read, reveals activations in the anterior insula and medial SMA (bordering Anterior Cingulate) which were not visible on the pial surface representation in Figure 3. (c) fMRI activity for the contrast Overt-Inflect > Zero-Inflect, intended to identify the manipulation of morphophonological content, reveals a frontal network including BA 44, 45, 47, anterior insula, AG, and medially the anterior and posterior cingulate. (d) fMRI activity for the contrast Zero-Inflect > Read, which holds phonological output constant and hence varies only abstract grammatical features. The Zero-Inflect task shows activity in distinct regions of BA 44 and 47, as well as a middle precentral gyrus premotor region; and no activity increase in insula or BA45. The Read task elicits activity in the supramarginal gyrus cluster, middle lateral occipital, and medial precuneal and subparietal regions. (e) Depiction of the complementary nature of activation patterns in (c) and (d). Positive activations from the contrast Overt-Inflect > Read (as in b) are outlined in black and shown to be nearly a superset of non-overlapping positive activation subsets for Overt-Inflect > Zero-Inflect (purple) and Zero-Inflect > Read (green), with overlap in yellow.



In decomposing the Overt-Inflect and Zero-Inflect tasks, we have characterized them as being in a perfect superset-subset relation, in which overt inflection requires all the morphosyntactic feature manipulation required by zero inflection and, in addition, the operations involved in manipulating morphophonological content. Under an assumption of pure insertion, this would suggest that the areas active in the Overt-Inflect > Zero-Inflect contrast (feature processing) and the Zero-Inflect > Read contrast (phonological manipulation) would be disjoint, and that the union of these areas would be similar to the Overt-Inflection > Read contrast. Figure 4 shows that these assumptions are very nearly correct. Though the areas identified by the Overt > Zero and Zero > Read contrasts (Figure 4c and d respectively) are both concentrated in the left inferior frontal cortex, they do not overlap: Overt > Zero yields significant activation differences in BA44, 45, 47, the anterior insula, and medial SMA, whereas Zero > Read yields significant differences in a more anterior portion of BA44, more superior portions of BA47, and a BA 6 premotor region at the junction of the precentral and middle frontal gyri. Figure 4e illustrates how the positive activation clusters of the two contrasts are indeed non-overlapping and how they combine neatly as subsets of the Overt Inflect > Read activation. This pattern is not mandated by the statistics nor the putative neurophysiology. It would seem plausible, for instance, that the two contrasts would recruit the same neural resources to increasing degrees (resulting in summing but overlapping clusters). The simplest interpretation of the observed patterns is that the former set of areas is responsible for the manipulation of morphophonological content and that the latter set of areas is responsible for the manipulation of abstract morphosyntactic features (Figure 4e).

No doubt this is too simple, for many reasons, but at least one aspect of the contrast between contrasts may be interpretable in these terms. The activation of the insula in the Overt > Zero contrast implicates the insula in phonological manipulation, and this is consistent with the analysis of a large sample of aphasic patients by (Dronkers, 1996), who discovered a perfect correlation between damage to the insula and a diagnosis of Apraxia of Speech (AOS), a difficulty in articulatory programming resulting in distortions of the target word such as yawyer for lawyer or tornyadiyudder for tornado. These findings are also consistent with those of (Dogil, 2003), who showed increasing anterior insula activity for repeating syllables of increasing phonological complexity. The convergence of these three findings is imperfect. First, Dronkers’ lesion-overlap region (the precentral gyrus of the insula) is posterior to the region shown to be metabolically active by the current data and the Dogil study. One possibility is that the discrepancy is an artifact of the difference between neuroimaging and lesion methods, perhaps because of constraints on the latter imposed by cerebral vasculature (Sahin et al., 1998; Caviness et al., 2002). Another is that the phonological

programming functions of the insula are themselves subdivided according to whether they implement the pronunciation of isolated words or products of morphophonological manipulation. The second discrepancy is that Dogil, et al. found insula activation only for overt, not covert, speech, leading them to suggest that the anterior insula is involved in the interface between symbolic phonological representations and motor implementation, rather than the manipulating of phonological representations. It is possible that our task, designed to approach natural speech, evokes lower-level speech planning processes to a greater degree than Dogil’s repeated syllables and words. The precise role of the insula in morphophonology, phonological programming, and phonetic implementation deserves attention in future research.

To isolate the process of suffixation of morphemes, we repeated the main task contrasts only using trials where regularly inflected nouns and verbs were presented (Figure 5). These contrasts focus on the addition of inflectional morphemes (-ed or –s) according to a predictable pattern rather than the replacement of all or part of the form using an idiosyncratic pattern retrieved from memory. More generally they remove any interaction of word regularity from the signal related to the morphological demands of each task condition. Note that the same threshold of p < 0.005 was used even though only half the number of trials were averaged into these contrasts (regular only). The pattern above was largely preserved, except for the differences in the anterior insula: Overt-Inflect > Read did not activate the anterior insula significantly when only regularly-inflected words were considered, though there was a trend toward significance.

Figure 5 - Primary task contrasts in group fMRI (p <

0.005), only for regularly-inflected (a) nouns and (b) verbs.



Reading Words versus Inflecting Words

Though the relationship between overt and zero inflection is consistent with simple assumptions about pure insertion, the relationship between inflecting words and reading them aloud is not. A straightforward information-processing analysis of the task conditions suggests that all the processes involved in reading a printed word aloud (a common baseline task in studies of morphology as well as semantics) would be included as a subset of the processes for inflecting the word. The fMRI contrasts between inflecting and reading words do not sit well with this analysis. There are several blue regions in Figure 4 (b and d), and since these contrasts subtract two similar conditions, the deactivations are best interpreted as regions more activated by the Read condition. In Figure 4b, which compares the Read task with the Overt-Inflect task, the Read trials are shown to recruit the middle and superior lateral occipital gyri and the supramarginal gyrus. In Figure 4d, which compares the Read task with the Zero-Inflect task, the Read trials are shown to recruit these same regions (SMG more than occipital) together with the medial precuneus, subparietal sulcus, and anterior cingulate. This suggests that the processes involved in reading a printed word aloud are not a subset of the processes involved in inflecting it. Further support comes from data on the right hemisphere, where the Read task appears to activate extensive regions of the middle temporal gyrus and sulcus, superior temporal sulcus, and the homologue of Wernicke’s area. These differences are not shown in Figure 4 but can be seen as activations in the Read > Fixation contrast in Figure 2c. Furthermore the absence of these regions from the comparisons of Overt Inflect or Zero inflect to Fixation (Figure 2a and b) shows that the right-hemisphere regions are recruited specifically by the Read task rather than simply being recruited to a greater extent by this task than by other conditions.

Perhaps when people are cued that they merely have to pronounce a printed stimulus, they adopt a strategy of attending closely to the visual word and mapping it onto its pronunciation in as direct a path possible (See (Proverbio et al., 2004; Fiebach et al., 2002; Joubert et al., 2004)). This may explain why possible pathways between the secondary visual areas responsible for word recognition and the posterior perisylvian language areas involved in words’ phonological representations appear to be differentially activated by the Read task. Given our poor understanding of the role of the right hemisphere in language processing (especially as indexed by neuroimaging), one can only speculate as to why there was such extensive right-hemisphere activation in the Read task. Possibly words (but not grammar) have a diffuse and redundant representation in the right hemisphere, which is ordinarily suppressed when grammatical processing is engaged, but can play a role in word recognition when the task demands are shallower, such as reading aloud (Baynes et al., 1992; Baynes

et al., 1998) (Mellet et al., 1998; But see Fiebach and Friederici, 2004).

Recent work contrasting internally and externally focused brain activity may also bear on these results. Though baseline periods are designed to keep the brain uniformly ‘quiet’ or (more likely) uniformly noisy, many fMRI studies show puzzling “deactivations” for these baseline tasks in medial structures such as the precuneus and posterior cingulate. Raichle and colleagues specifically targeted the brain’s resting state by scanning with PET while subjects lay awake with their eyes closed (Raichle et al., 2001). They suggest that these midline areas are involved in actively maintaining a default mode (in which sensory information is monitored and evaluated for salience), and that they are actively inhibited when the brain switches to a goal-directed mode of operation. Convergently, the locus of the EEG alpha rhythm (stereotypical of calm or resting states) has been traced to similar brain regions (Martinez-Montes et al., 2004; Miwakeichi et al., 2004). It is possible that our Read task is best performed relatively automatically, with the brain closer to its default mode, and that this mode is actively inhibited in the more attention-demanding Inflection tasks.

Independent of the specific interpretation, the pattern of results clearly shows that Read trials involve cognitive processes and/or neural circuits not part of the ostensibly more grammatically involved tasks. Having multiple baselines and multiple conditions for comparisons allowed us to find this pattern, and yet to cross-validate our own results (e.g. of Broca region (e.g. BA44) and insula involvement in inflectional morphology, via the Overt-Inflect > Zero Inflect contrast). However, a vast preponderance of neuroimaging studies of language use single-word reading as the only baseline. Care should therefore be taken in using a simple reading baseline task, and interpretations of right-hemisphere involvement in normal language should be dissected in terms of the linguistic tasks and strategies involved in the paradigm.

Common Pathway for Inflection of Nouns and Verbs

As mentioned in the introduction, evidence from aphasiology, neuroimaging, and psycholinguistics suggests that nouns and verbs may differ in their patterns of cortical representation, though there are disagreements on whether the differences are based on the categories themselves or on their characteristic meanings (Caramazza and Shapiro, in press; Shapiro et al., 2001; Gentner, 1981; Luzzatti et al., 2002; Perani et al., 1999; Pulvermuller et al., 1996; Pulvermuller et al., 1999a). In the present investigation, the Read task allowsan approximate comparison to the tasks used in most of the previous noun-verb studies. The comparison of simple reading of verbs and nouns (Figure 6) yielded results generally convergent with (Perani et al., 1999), namely that the only



regions showing significant differences were those with a greater response to verbs; none showed a greater response to nouns. Moreover, the verb-selective regions were in lateral temporal and dorsolateral prefrontal cortices. Compared to Perani et al., the present frontal activations were more dorsal, in premotor areas, and activations were more bilateral. This contrast defines a starting point for verb-noun differences in inflection.

The overt inflection tasks used in this experiment could exaggerate such differences, as the verb inflection was past tense, which refers to the inherently verb-relevant semantic feature of time, whereas the noun inflection was plural, which refers to the inherently noun-relevant semantic feature of number. If the Read results and previous studies indicate the storage or access of verb and noun lexicons, and the process of grammatical inflection takes place where the words are stored, then the differences should be further accentuated in Inflection. Despite these reasons to find differences, the gross pattern of activation in the Overt-Inflect > Read condition (expected to index inflectional processing) showed substantial similarities (see Figure 7). The differences predicted by other studies in the literature are separate or greater activations in temporal lobe for nouns and in the inferior frontal gyrus for verbs. In fact, both contrasts significantly activate the set of areas seen in the global contrast, namely BA44/45, BA47, anterior insula, and (not shown in the figure) SMA. There are indeed magnitude differences in these circuits, with nouns showing a stronger response than verbs in BA44/45 and BA47. These differences may come from the lower frequencies of the nouns (Chee et al., 2002; Chee et al., 2003), and especially from the inflection of the unusual irregular nouns: when only regular forms are included in the contrast, the differences in the magnitude of the significantly activated areas is reduced. There appear to be no categorical differences in the four major frontal areas recruited by the different grammatical categories, which suggests that there may be a common circuit supporting inflectional morphology across different grammatical categories.

(Tyler et al., 2004) present the first neuro-imaging paper to look for possible differences in the inflection of verbs and nouns, using very different methods. Rather than using a production task triggered by a linguistic context, a they asked subjects to decide if one member of a triplet was semantically related to the others; all words were inflected, in the plural or the progressive present). They found common left inferior frontal activations for both words classes, much as in the present study (together with activations in the temporal lobe) and a greater magnitude of activation for verbs in the inferior frontal regions. These results are consistent with ours in showing more similarities than differences in which regions subserved inflection of verbs and nouns, despite differences between nouns and verbs in tasks that do not call on inflectional morphology.

The one region selectively active for the noun and not verb contrast is the intraparietal sulcus (IPS) (Figure 7b and c). Other contrasts in this experiment show that this region is not specifically activated by noun inflection; it is also more active when people overtly inflect or read verbs in comparison to the fixation condition. This suggests that the major function of the IPS in these tasks is to mediate task-related visual attention (Jovicich et al., 2001), and the greater activation seen here for inflecting nouns may be a consequence of their lower frequency and the unusualness of some of the irregular plurals. If the contrast really does come from some inherent difference between noun and verb inflection, one possibility is that the IPS is implicated in number cognition (Dehaene et al., 1996; Chochon et al., 1999) and our noun overt inflection task requires discriminating singular and plural number. (Verbs also mark number, specifically, the number of the subject, but that feature needs to be manipulated only in the Zero-Inflect condition in this study.)

Figure 6 - Silent reading of verbs and nouns (threshold p < 0.005).



Figure 7 - Verb versus Noun inflectional processing. Results (p < 0.005) of the Overt-Inflect > Read contrast as in b but performed separately for (a) verbs and (b) nouns. c.) Results of a direct contrast between verbs and nouns within only the Overt-Inflect task.

Regular and Irregular Inflection

Regular and irregular inflection, as mentioned, have been shown to be linguistically, psychologically, and neurologically distinct, leading to the expectation that neuroimaging studies might show increased involvement of left inferior frontal areas for regular inflection and of left temporal and parietal areas for irregular inflection. The current results fit with previous neuroimaging studies in failing to show such an association. Figure 8 reveals that there are indeed substantial differences in the brain areas engaged by inflecting regular and irregular forms. However, these differences were neither consistent across verbs and nouns (with the possible exception of regions of the anterior insula and anterior cingulate) nor were they in conformity with the expected anterior-posterior difference. The verbs showed, if anything, the opposite association, and both nouns and verbs showed increased engagement of inferior frontal areas by irregular inflection.

A few tentative interpretations are compatible with these data. In especially the verb analyses, there was bilateral activation of the medial portion of the supplemental motor area bordering on the anterior cingulate cortex (ACC). These areas have been implicated in the monitoring of conflict and the inhibition of habitual responses, especially in conjunction with lateral frontal regions (Kerns et al., 2004a; Miller and Cohen, 2001; Holroyd et al., 2004). Activation of cognitive control and inhibition regions may be required by irregular inflection for two reasons: blocking of the regular rule to prevent overregularizations (e.g., bringed), and suppression of conflicting responses among competing irregular patterns (e.g., brang or brung) that are generalized from families of similar irregular forms (e.g., spring-sprang, ring-rang, drink-drank, sing-sang). The latter explanation would be consistent with the weakness of such activation in the noun data, since irregular nouns do not fall into such phonological families. (Jaeger et al., 1996) also showed strong left ACC and middle/posterior cingulate activations preferentially for inflection of irregular past-tense forms, with weaker ACC for the inflection of nonce verbs.

A second possible systematicity is the involvement of left inferior prefrontal cortex and bilateral anteroventral prefrontal regions in irregular inflection, paralleling a finding by (Rhee, 2001; Rhee et al., 2003). These anterior and ventrolateral regions of cortex have been shown to be activated tasks requiring selection of words from memory on the basis of semantic and lexical cues, such as stem completion (Desmond et al., 1998) , generation of a verb related to a target noun (Petersen et al., 1998), and discrimination of abstract from concrete nouns (Demb et al., 1995; Poldrack et al., 1999). Possibly the selection of an irregular form from memory based on the intersection of the lexical item and a set of morphosyntactic features (and the suppression of inappropriate words that partly meet such criteria) engages the same circuit.

Figure 8 - Regular and irregular inflection of verbs (a) and nouns (b) in Overt Inflect trials. Since these analyses contrasted data from only one of the twelve conditions, a more liberal threshold was used (p < 0.05).



The medial views in Figure 8 show small regions preferentially activated by regular inflection (depicted in red) on the mesial surfaces of the hemispheres, in and below the posterior corpus callosum on the right, and more posterior on the left. These regions do not actually correspond to cortex, because their surface topography in these displays is partly an artifact of the cortical inflation algorithm. Rather, the immediately subjacent regions project through and are visualized on the virtual surface. The regular-specific activations are thus likely to be projections from subcortical structures, though their coordinates and identities are not uniquely identifiable by the methods of reconstruction used in this study. One possibility is that they arise from basal ganglia activity, which would be consistent with the claims of (Ullman et al., 1997; Ullman et al., in press) that regular inflection is computed in a circuit which includes the basal ganglia.

A final convergence of the present data with existing studies comes from overall patterns of activity for regulars and irregulars. Many more voxels were preferentially active for irregular inflection than regular inflection, a result found in all previous studies (Jaeger et al., 1996; Indefrey et al., 1997; Rhee, 2001; Beretta et al., 2003) and reviewed in (Beretta et al., 2003).

In the maps for noun inflection, there are regions of preferential activation for irregular inflection in the left inferior temporal cortex, sometimes associated with lexical knowledge of nouns, and left intraparietal sulcus, implicated in visuospatial attention as well as numerical processing (see above). It is possible that these areas are activated by the lexical retrieval of nouns, which is required by irregular but not regular inflection of nouns, nor by either kind of inflection of verbs.

These tentative speculations contrast with the systematicity of the activation maps presented above for inflection in general, and of the patterns of association found in studies of neurological patients. We suspect the discrepancy arises from the fact that neuroimaging, which reveals the full network of processes implicated in inflection rather than the indispensable ones, is especially sensitive to the ways (discussed in the introduction) in which the regular-irregular distinction maps only imperfectly onto the computation-memory distinction. Future studies may need more subtle manipulations, involving carefully selected subsets of words, rather than an across-the-board regular-irregular dichotomy, to systematically map the effect of irregularity on the interplay between memory and computation. It may also be necessary to focus directly on the regular-irregular contrast rather than embedding it in a matrix of twelve primary conditions.

EXPERIMENT 2: INTRACRANIAL ELECTROPHYSIOLOGY

To obtain converging evidence on the

localization of inflectional processing in the brain, and to obtain evidence on the brain’s fine temporal response to the experimental paradigm (difficult or impossible with fMRI), we administered the paradigm to a single patient and recorded her brain activity using both fMRI and intracranial electroencephalography (iEEG). Recall that iEEG is made possible by the fact that some epileptic patients who are eligible for surgery to remove the epileptogenic focus have electrodes implanted in their brains to help localize the focus. During the long interval in which the patient’s brain is being monitored in anticipation of a seizure, cognitive tasks may be administered.

Unlike EEG signals recorded from the scalp, intracranial EEG can encode unambiguous spatial information. It records task-evoked electrical potentials which come directly from nearby neural tissue, rather than being smeared through the scalp with signals originating everywhere else in the brain. Moreover, it represents the summed activity of several hundred to several thousand neurons (or fewer, with microelectrodes) rather than the millions of active neurons summarized by each fMRI voxel, and the activity is the actual electrophysiological response of the neurons, rather than the BOLD signal, which is only indirectly related to these responses. A final advantage of iEEG is that spatial inversions in nearby signals reveal the locations of the corresponding generator or generators: if a similar response peak is observed in two neighboring channels, but the sign is reversed (i.e., the peaks are reflections of each other across the zero line), then the neural generator is located more or less between the channels. (More complex inversion patterns can reveal slightly more complex spatial relationships among generators as well.) (Halgren et al., 1994)

Intracranial EEG is a relatively rare technique in the study of higher cognitive processes and has not been applied to investigate the grammatical role of Broca’s Area. An additional goal of this experiment is to explore how data obtained with this technique may be combined with data from fMRI to uncover the dynamics of Broca’s Area function.

Method

Subject

A patient scheduled to undergo surgery for treatment of epilepsy (J.J.) consented to participate in cognitive tests during an fMRI session prior to placement of the iEEG electrodes and during the iEEG monitoring period. J.J. is a right-handed woman who suffered seizures since the age of 14, and was 41 years old at testing. She held a regular job, had normal language skills, and was able to complete the task without difficulty. Clinical



analysis of intracranial recordings showed that seizure onset was localized to the left mesial temporal region. The recording leads examined here did not display interictal epileptiform spikes. A brief nonclinical electrical seizure episode localized to the left anterior hippocampus occurred 5 hours before the cognitive recording session, and no clinical seizure episodes occurred within 24 hours of the session.

Experimental Procedure Patient J.J. participated in the fMRI experiment

the night before the surgical opening of her cranium and implantation of the electrodes, and the experimental paradigm was administered again 10 days later while iEEG recordings were obtained. For iEEG, the experiment was displayed with a notebook computer, using an external mouse for the button-press, as the patient sat up in her hospital bed.

The task paradigm was identical to the one used with the healthy subjects, except that the timing was slowed to accommodate her slightly lower reaction time, as assessed during the training procedure. For the fMRI experiment, each half of the trial (locked to the TR) was slowed to 2.6 sec; for iEEG, each half was 2.5 sec. In both, the context frame was presented in the first half for 800 msec before being replaced by the fixation cross, and the target word was presented in the second half for 350 ms. The longer TR for the fMRI experiment was used to obtain 36 slices at 3 mm spacing, rather than 25 at 5 mm.

Surgical and Recording Procedure

In order to monitor J.J.’s seizures, surgeons exposed her left perisylvian cortex (Figure 9a) and implanted four multi-contact electrode arrays (also called macroelectrodes or depth electrodes) orthogonal to the cortical surface, each with 8 independent electrical contacts. An electrode array is like a long hollow plastic needle; the contacts consist of bands of silver along its length with lead wires embedded in the plastic shaft (see Figure 9e). Each contact was 1.5mm long and was separated by 3.5mm from the next contact. Two arrays yielded electrical activity significantly correlated with the task: the Orbital Frontal (OF) array, which traversed the ventral portion of the inferior frontal gyrus (IFG) and passed near the anterior insula, and the Middle Frontal (MF) array, which traversed the middle frontal gyrus (MFG) and basal ganglia on the way to the cingulate gyrus, just above the genu of the corpus callosum (Figure 9c and d). Other electrodes were implanted as well, but data will not be analyzed from them. These include amygdala- and hippocampus-targeting macroelectrode arrays (which showed no significant responses correlated with the tasks), a pial-surface grid (which malfunctioned), and microelectrodes threaded through the hollow shafts of the macroelectrode arrays to pick up multi-unit activity within

individual lamina of the cortex (which had become noisy by day 10, when J.J. was tested with the present task).

Recording was carried out using the standard clinical telemetry system in the Beth Israel Deaconess Medical Center clinical EEG laboratory (Ives et al., 1991; Schomer, in press). The “channels” consisted of the voltage difference (gradient potentials) across a neighboring pair of contacts along one of the multi-contact arrays (e.g., “channel 3-4” on a given array would be the voltage between contacts 3 and 4). Specifics of the clinical preamplification circuitry meant that gradient potentials were expressed in arbitrary units linearly related to standard voltage units (roughly 161nV per unit). Data processing was done with NeuroScan Scan 4.3 software on a Windows XP platform. Signals from each channel were sampled at 200Hz, high-pass filtered at 0.1Hz, and low-pass filtered at 100Hz. Electrical triggers sent at each trial by the stimulus presentation program were used to determine trial type, and the signal from each trial type was averaged and normalized to the pre-stimulus baseline. In one analysis, these Evoked Potentials were epoched from -1 second to +5 seconds relative to the onset of the context frame; in another, they were epoched from -1 to +3 seconds relative to the onset of the target word.

Average waveforms were inspected manually for peaks and peak clusters within a physiologically plausible timeframe (e.g. ~100ms to ~1000ms). Peak amplitude, latency, and time to 95% decay were compared across major condition types and across the various OF and MF channels. Special attention was given to inversions in the spatial dimension, where the waveforms for neighboring channels deflected with similar latency and amplitude but opposite sign. Since iEEG, unlike scalp EEG, reveals unambiguous localization of the electrical signal, a steep inversion across a short physical distance indicates a neural generator in the intervening space. The reasoning behind this inference is as follows. Field potentials are generated by summated neuronal activity. The frequency of this activity is low enough that it can be considered as quasi-static. When activation is localized, then at a distance of a few times the maximal charge separation, the net field will come to resemble a dipole (Nunez, 1981). The potential gradient recorded by a linear array of electrodes (as in the current recordings) in the presence of a dipole will be of constant polarity, regardless of the direction of the approach. Although potential gradients can be recorded at a distance from the generator intracranially, they will be of constant polarity, relatively small, and of nearly constant amplitude (Goff et al., 1978). Conversely, when the recordings are closer to the dipole, the amplitude of the gradient, and the rate of change of the gradient, will be larger. An inversion of the potential gradient (or double inversions as were sometimes observed here) implies that the recordings were obtained within a region with a complex charge distribution, and therefore that the generator lies



within a distance at the same scale as the inter-contact distance.

Where evoked potential components were visually observed to differ by condition, two-tailed

t-tests were carried out on the individual trials for each post-stimulus time point to estimate the statistical reliability of these differences.

Figure 9 - Surgical implantation of intra-cranial electrodes in patient J.J. (a) Patient J.J.’s left lateral cortex, reconstructed from pre-surgical MRI, with the outline of the cranial excision shown in the photograph in (b). Anatomy: middle frontal gyrus (MFG), pre-central gyrus (PrCG), post-central gyrus (PoCG), supra-marginal gyrus (SMG), superior temporal gyrus (STG), middle temporal gyrus (MTG). BA44 is partly obscured by a sub-pial vein following the pre-central sulcus. The Sylvian fissure extends laterally, also obscured by a vein, and branches of the middle cerebral artery (MCA) cover the visible gyri. The four insertion points of 8-contact macroelectrode arrays are colored, with labels for the Orbital Frontal (OF) and Middle Frontal (MF) arrays from which intracranial electroencephalography (iEEG) data are presented. (c) The trajectory of the OF array in a post-implantation MRI slice, with arrows indicating the contacts traversing ventral inferior frontal gyrus and anterior insula regions. (d) Trajectory of the MF array, traversing the MFG and basal ganglia and terminating in the cingulate gyrus, just superior to the genu of the corpus callosum. The dark nodes along each trajectory exaggerate its width, as they represent signal dropout artifacts. (e) Intra-operative view from a patient demonstrating clinical placement of the depth electrodes. Depth arrays are the termini of the hollow plastic leads seen piercing (and sutured to) the plastic surface grids.

Results

Comparison of fMRI Results to Group Data from Experiment 1

In the contrast between Overt-Inflect and

Fixation, the pre-implantation fMRI of J.J. showed an activation cluster near Broca’s Area (BA 44/45 and anterior insula), convergent with results from the healthy volunteers in Experiment 1 and within the range of the data from individual subjects in that study (see Figure 10a and b). The center of gravity for the corresponding fMRI activation lay slightly superior to the group average region. When the fMRI activation was displayed on a horizontal slice from the patient’s anatomical MRI, it

confirmed that there were voxels in the Broca’s-area region responsive to the principal contrast (Figure 10c). Some of the activated brain voxels lay where macro-electrode contacts 3 to 5 of the Orbital Frontal (OF) array were subsequently implanted (Figure 10d). fMRI activation was also seen in regions traversed by the Middle Frontal (MF) array. Electrical activity recorded from these arrays provided direct comparison of fMRI with intracranial electroencephalography. The region where fMRI and iEEG could be directly compared, was in the left inferior frontal gyrus, just anterior and subjacent to the core of Broca’s Area, and passing near the anterior insula.



Figure 10 - Broca’s Area activation in morphological processing, convergently observed in group fMRI, patient fMRI, and intracranial electroencephalography. (a) Group fMRI results from the Overt-Inflect > Fixation contrast in Experiment 1, depicted on the cortical surface of patient J.J. Regions of significant activity increases for Overt-Inflect (p < 0.0005) include inferior Broca’s Area and insula (circled). (b) The identical fMRI contrast in patient J.J. (pre-operative), depicted once again on the patient’s brain, also includes Broca region activity (p < 0.0005). (c) The same fMRI contrast, depicted on horizontal slice of patient’s brain. (d) A similar horizontal slice from clinical MRI taken after implantation surgery, showing the OF array within recording range of the fMRI voxels. (f) Electrical activity recorded from this array, averaged over all Overt-Inflect trials. Tracings represent channels, namely the gradient potentials or voltage differentials between pairs of contacts along the length of the array. Activity is calculated relative to a one-second baseline (here time zero is the onset of the target word), and are in arbitrary units dictated by clinical preamplification (each roughly 161 nV). Greatest responsiveness is exhibited in regions of fMRI activation, with steep inversions across channels 3-4, 4-5 and 5-6 indicating source localizations within this 1 cm region. Evoked potential components peak at roughly 255ms, 380ms, and ~500ms. (e) Lack of significant electrical activity at the same electrode during Fixation trials, the baseline of the corresponding fMRI contrasts.

It should be noted that some of J.J.’s fMRI

activation patterns in the primary contrasts were bilateral, with the right-hemisphere activations being more diffuse than the left-hemisphere ones, which included more focal activation in Broca’s region. This is a reminder that BA44/45 (Broca’s) is not the only language- or task-sensitive fMRI activation in this patient; we focus on it because it is a strong region of activation which converges with both the group fMRI data and the patient’s iEEG. Because the patient’s seizures started at age 14, with an apparent focus in the left temporal lobe, her right-hemisphere language activity may have developed as a complement for continuing left-hemisphere language abilities.

In order to combine data from fMRI and iEEG and relate them to traditional electroencephalo-graphic data from scalp electrodes, we set a number of criteria to identify any iEEG response components that might be specifically related to grammatical processing. Epoched, averaged, task-timelocked gradient potentials at all channels were inspected for statistically significant deflection patterns over a 5-second post-stimulus time window. Candidates for grammatical processing components needed to be:



1. Responsive to the task: There had to be significant deflections from the pre-stimulus baseline at the given channel during task epochs (see Figure 10f, Figure 11a and b).

2. Selective to the task: The deflections should not occur during the non-task (visual fixation) trials.

3. Selective to the triggering event within the trial: The deflections should occur in response to the target word that initiates the inflectional processing, rather than to the context frame that merely cues the task (Figure 11 c and d).

4. Sensitive to task complexity: There should be a larger response for more inflectionally demanding tasks. Specifically, Overt-inflect should elicit larger responses than Zero-inflect, which in turn should elicit a larger response than Read (see Figure 12, Figure 13).

5. Plausible cerebral location: The channels showing selective responses should be located in regions that have been implicated in prior fMRI results or that are thought to be relevant generator regions based on existing literature.

6. Plausible neuronal timescale: The temporal window of the response should be neurophysiologically credible and consistent with prior results from scalp-recorded event-related potentials.

Criteria 1 and 2: Responsivity and Selectivity to the Task

The electrical responses were screened first

according to the simplest criterion for correspondence to grammatical processing, that there be consistent evoked potential (EP) activity on a given channel while the subject performed the cognitive task. This criterion was met by several channels along the MF and OF macro-electrode arrays. Figure 10f shows the activity for all OF channels during the Overt-Inflect trials. Figure 11a narrows this down to the most responsive channels, while Figure 11b shows the activity for the responsive MF channels. The three most responsive channels in the OF array were from pairs of electrodes traversing the region of overlap with the within-subject and group fMRI data, namely inferior Broca’s area and the anterior insula. The responsive MF channels were anterior and superior of Broca’s Area, in the middle frontal gyrus. The other channels in the OF and MF showed little or no consistent activity (e.g. Figure 10f), and the other two arrays (inserted through the temporal lobes and targeting the hippocampus and amygdala, respectively) lacked channels with significant peak voltage deflections from the baseline in these comparisons. Thus we have found two sets of channels whose iEEG signals to the cognitive task converge with the fMRI signals.

The second criterion, task selectivity, ensures that the selected responses are not generic to all

mental events (i.e., as an index of mental effort elicited by external stimuli) by stipulating that the response should not exist during unrelated cognitive activity. The condition in this study that was not related to linguistic processing was the Fixation condition. Figure 10e shows that there were no significant response components in the OF channels during the fixation trials. The same was true for the MF channels (not shown).

The signals were then examined for temporal components, and for sign inversions in the spatial dimension. All Broca-region channels along the OF array showed a sharp component at 255 milliseconds post-onset of the target stimulus, a component with a maximal peak at approximately 435 milliseconds and a more gradual offset decaying to about 925 milliseconds, and a possible component between these two (Figure 11a). Responsive channels along the MF electrode array showed sharp peaks at 255, 355, and 500 msec post-onset of the target stimulus (Figure 11b).

In an effort to localize possible generators of the signal, signal inversions in pairs of nearby channels were examined. In the early component of the response in the OF array, channel 3-4 deflects in the negative-going direction, while channels 4-5 and 5-6 mirror the response but deflect in the positive-going direction. This indicates a nearby generator at the depth of contact 4, which is in the medial aspect of the inferior frontal gyrus at the level of the anterior insula. The later component, which peaks at 435 msec with a slower drop-off, shows a different inversion pattern: 3-4 and 4-5 still invert against each other, but in this case 5-6 follows 3-4 (negative-going) rather than 4-5 (see Figure 11a). This double-inversion pattern indicates a generator spatially segregated from the one previously described. This suggests that the two response components come from different neural processing events in different locations, perhaps representing communication between nearby and related nodes in a larger neural circuit.

Similar response components were observed in four channels along the MF array, coming from electrodes that traversed the MFG. The most medial channel, MF 2-3, showed only an early component, a steep peak at 250 msec. MF 3-4, 4-5, and 5-6 evidenced a homologous 255 msec component as well as steep peak deflections at 500 msec and less pronounced ones at 355 msec. The inversion patterns were seen in each of these components (see Figure 11b).

These results corroborate the fMRI data by showing task-related electrical responses in the same region as the same patient’s fMRI voxels, but in addition extend those data by breaking the response into at least three temporal components and suggesting that they may come from distinct neural generators, separated in space as well as time. The fMRI data blur these three generators into one.



Figure 11 - iEEG task responsivity and selectivity. (a) Electrical activity for the Overt-Inflect task from the consistently active channels OF 3-4, 4-5, and 5-6. Anatomical locations are indicated by color-coded lines superimposed on the OF array trajectory. (b) Electrical activity and anatomical locations of active channels along the MF array. Channels on both arrays show triphasic responses, with components at 255, 355, and 500 msec from presentation of the target word. Steep inversions of the signal across channels suggest the presence of a generator nearby and lengthwise between them. The 255 msec component in (a) has channel 4-5 and 5-6 deflecting together (positive-going) and against 3-4 (negative-going). The relationship between 3-4 and 4-5is also seen at 355 msec. The complex centered around 500 msec has channel 5-6 broadened and inverting against 4-5 and perhaps shifted. Such patterns suggest the presence of spatially and temporally distinct neural generators responsive to the task. (c) and (d) show the signals temporally aligned with the context frame, rather than the target word, as t = 0. For easy comparison of post-frame to post-target response, the pale vertical bars demarcate identical time windows after each stimulus epoch. The late responses (~500ms) are selective to the post-target delay, when word inflection must take place.

Criterion 3: Selectivity to the Relevant Event Within the Trial

According to our task analysis, it is the target

word, not the preceding context frame, that elicits the morphological processing we are examining. However, the fMRI data alone cannot show that the neural activity being measured is in fact triggered by the target word (the procedure included only inter-trial jitter, not intra-trial jitter). iEEG data, in contrast, can pin down the temporal trigger of a task-relevant signal. In this analysis, the electrophysiological responses to trials were averaged with the zero point set to the onset of the context frame rather than the onset of the target word. If the responses we have been examining are not specific to grammatical processing, but just, say, to seeing words, the iEEG response should be equal to each event in the trial (or possibly greater to the context frame because it contains more words). If the response is specific to grammatical processing, the iEEG signal should deflect primarily or solely to the target word that initiates the actual grammatical operation.

Figure 11c presents the results for the Overt-Inflect condition and the OF array. For the late components (400-900 msec range), there was no deviation from noise in the equivalent period following the context frame, but the strong

response continues to be seen in the period following the target word. The early component (255 msec) was minimally present in the epoch following the context frame and again strongly visible in the epoch following the target word. This confirms that the electrophysiological response we are focusing on is indeed tied to inflectional processing, and since the response came from a region that overlaps the one activated in fMRI, it ratifies our interpretation that the fMRI signal also indexes inflectional processing elicited by the word rather than a more generalized activity associated with the context frame or words indiscriminately. Figure 11d shows the same analysis carried out for the channels in the MF array. The late-component response showed an identical pattern, namely that the response at 500 msec was exclusive to the interval following the target word. However, the early component – at 255 msec – was elicited both by the context frame and the target word. This suggests that the 500 msec component of this generator is the one more closely related to grammatical processing. It is unlikely to be just a sentence wrap-up effect, simply indexing sentence closure after the final word, because if it were we would not expect its peak amplitude to be modulated by the morphological complexity of the task (see below). Another possible objection to the idea that this component indexed morphological processing is that it appears quite late (a criticism that, in the scalp EEG literature, has also been



levied against the P600 as an indicator of syntactic analysis). Yet the 500 msec latency does not imply that it takes 500 milliseconds to execute morphological processing; it may take a substantial portion of that interval for the necessary constituents to reach this processor, which then carries out its particular task more swiftly.

Criterion 4: Modulation by Complexity of Task

Focusing on the target-triggered components of the response, we then asked whether they were greater for tasks requiring greater engagement of the morphological system: Overt inflection should elicit a stronger response than zero inflection, which should in turn elicit a stronger response than simple reading. In Channels 3-4 and 5-6 of the MF array (Figure 12), the early component (255 msec) showed close overlap of the peaks of the signals for the three tasks, with a small advantage in amplitude for the simplest, Read. In contrast, the 500 msec component in channel MF 5-6 (the one nearest Broca’s Area) showed the lowest amplitude for the least morphologically demanding task, Read, with successively increasing amplitudes for Zero-Inflect ( t(161) = 2.40 (p = 0.018, 2-tailed), at 435 msec) and Overt-Inflect ( t(161) = 3.70 (p = 0.0003, 2-tailed) at 435 msec) (Figure 12a). For further statistics, see figure captions. This pattern paralleled the fMRI results, and further implicated the 500 msec response at channel MF 5-6 as one of the most likely candidates for a neurophysiological correlate of grammatical processing.

As seen in Figure 12b, the more medial channel MF 3-4 (recording from the medial middle frontal gyrus) showed a similarly orderly response in the 355 msec component, comparing Read to Zero-Inflect (p = 0.002 at 350 msec) and Overt-Inflect (p = 0.0006 at 350 msec), but no significant difference between the two inflection conditions. The peaks at 255 for the different conditions did not differ statistically from one another; nor did the peaks at 500 msec. This suggests that the main infection-sensitive response here occurred at 355 msec, perhaps as an antecedent to the processing occurring at 500 msec in the more lateral cortex. It should be noted that the 355 msec component is not confirmed as an inflection-related component by criterion 3, because it responded equally to the context frame and the target word. However, the statistically robust differences by task suggest that it must have some involvement in inflection. The response may be related to the preparation of morphosyntactic features specified by the context frame (past, plural, etc.) that are necessary for inflecting the word when it comes. This converges with the fact that the significant differences among tasks in this component are between the two inflection tasks and the Read condition, rather than between the Overt and Zero inflection conditions - suggesting that it is morphosyntactic features that the component is sensitive to.

Channel 5-6 of the OF electrode array, in the ventrolateral inferior frontal gyrus, also shows a

gradation of response according to the task, but in the duration rather than the amplitude of the response (see Figure 13). The Overt-Inflect trials show a significantly more sustained response than the Zero-Inflect trials (p = 0.0001, at 740 msec), which in turn show a more sustained response than the Read trials. (A later complex that may index a distinct generator showed the same pattern in the 1080-1500 msec range.) It is possible that these differences reflect the different amounts of time necessary to complete the inflectional task. Both increases in amplitude and increases in duration of a neuronal response can lead to increases in the BOLD signal measured by fMRI, so only electrophysiological data can distinguish the two.

One interpretation of these findings is that there was an early response at 255 msec, which may be an indicator of the low-level visual processing of words, as it is independent of task complexity and responds equally to the presence and closure of the stimuli. Next, at around 355 msec, a response began in medial areas, which was involved in the processing of grammatical features. Then, a response around 500 msec in more lateral regions of the middle frontal gyrus reflected the processing of features and possibly phonology. Finally, the post-500 msec responses in the ventral IFG were more clearly involved in phonological processing, since they showed significant differences between Overt inflection and Read, and between Zero inflection and Read, with the differences between Overt inflection and Zero inflection failing to reach statistical significance.

Criteria 5 and 6: A Plausible Place and Time for the Neuronal Response

The MF and OF macroelectrode arrays both traversed left lateral and ventrolateral prefrontal cortex, which are known to be associated with language processing, especially for syntax (as reviewed in the Introduction). It was in channels along these arrays, and not those implanted in the temporal lobe, that the task-selective responses were discovered. Therefore the location of the identified electrical responses are consistent with the prior literature.

The temporal parameters of the response must be compared with the literature on scalp-EEG, because there are no published studies using intracranial EEG to study inflection. Though scalp EEG and intracranial EEG are quite different in the spatial information they record, they should agree in the time dimension, since conduction through the skull that smears the spatial fidelity of the scalp EEG signal should not smear it temporally (the relevant electrical signals dissipate immediately). This makes both signals different from the fMRI signal, which is smeared through time by the fluid mechanics of moving blood, the dynamics of oxygen extraction, and the underlying hemodynamic response.



Figure 12 - Modulation of the iEEG signal amplitude by the inflectional complexity of the task. Each tracing represents the same electrode channel but in a different task, and thus represents independent data. The tracings line up perfectly, and the amplitudes of their later components (the ones seen in Figure 11c and d to be most specific to the target word) depend on the task, with Overt-Inflect exceeding Zero-Inflect which in turn exceeds Read. (a) Channel MF 5-6 is localizable to the left lateral MFG, and corresponds to the black tracing in Figure 11b. Only the 500 msec component is modulated in amplitude depending on the task. The brown bar indicates the temporal range of significant differences at p < 0.01, and peak significance levels of the three possible contrasts are indicated with asterisks (Overt-Inflect > Read at 435 msec, t(161)=3.70 (p = 0.0003); Zero-Inflect > Read at 435 msec, t(161)=2.40 (p = 0.018); Overt-Inflect > Zero n/s ). (b) Activity in the more medial MF 3-4 channel (the red tracing in Figure 11b, localizable to medial MFG) is modulated by inflectional complexity earlier, at 350ms, suggesting that its grammatical computation precedes computation in the more lateral regions. (Overt-Inflect > Read at 350 msec, t(161)=3.52 (p = 0.0006) ; Zero-Inflect > Read at 350 msec, t(161)=3.21 (p = 0.002)).

Figure 13 - Modulation of the iEEG signal decay rate by the inflectional complexity of the task. The response of channel OF 5-6 (corresponding to the black tracing in Figure 11a, localizable to the left ventral IFG) shows alignment across tasks of a response peaking at roughly 435 msec, but with variation in its duration and decay. As before, components are found to be ordered according to inflectional complexity (Overt-Inflect > Zero-Inflect > Read). Significant differences are found in the 715-870 and 1080-1150 msec ranges. For the early range, Overt-Inflect > Read at 750 msec, t(161)=3.33 (p = 0.001); Overt-Inflect > Zero-Inflect at 740 msec, t(161)=3.93 (p = 0.0001). For the later range, Overt-Inflect > Read peaked at 1120 msec, t(161)=2.75 (p = 0.007); Overt-Inflect > Zero-Inflect at 1145 msec, t(161)=2.56 (p = 0.011). A very early component is also observed, significant in the 50-100 msec range (Overt-Inflect > Read at 90 msec, t(161)=2.79 (p = 0.006); Overt-Inflect > Zero-Inflect at 65 msec, t(161)=2.27 (p = 0.025).



The major language-related EEG components,

such as the N400 and P600 (at 400 and 600 milliseconds, respectively), are a point of departure. Earlier EEG components are traditionally associated with lower-level linguistic features, and after one second, responses are generally low in amplitude and fairly complex (peaks beyond 1 sec were generally ignored in our analyses). Therefore, the feasible neural time scale is 100-1000 msec post-stimulus, with a bias toward sharp components in the 300-700 msec range and more sustained around 1000 msec. Our plausible temporal range is also confirmed by magnetoencephalographic (MEG) signals, which for instance, show N400-like responses at increasing latencies in the 250 – 400 msec range along a temporo-frontal axis (Halgren et al., 2002), including in inflection tasks similar to the ones used here (Rhee et al., 1999; Rhee et al., 2003; Rhee, 2001). Thus no response component examined herein is eliminated by the criterion of temporal plausibility. It should be noted that we see significant differences by task (in the familiar pattern Overt-Inflect > Zero-Inflect > Read), in very early components of the OF 5-6 response (50-100 msec range). The meaning of these results is unclear, but early components in scalp EEG have shown word-specific responses in this range (Pulvermuller et al., 2004). Possibly it represents the first moments of interaction of the patient’s preparatory set (what she had prepared herself to do when the word appeared) with the appearance of the word itself, triggering the subsequent processing steps necessary to effect the necessary output form.

In sum, the fMRI results of Experiment 1 were corroborated by intracortical EEG signals that were responsive to the onset, temporal structure, and computational complexity of the task, and which occurred in implanted regions that overlapped with the fMRI patterns for that patient and for our larger sample. These analyses revealed in addition that the neuronal computation underlying inflectional morphology includes more than one generator, though the generators are similar in their spatial and temporal domains. An early component in all responsive regions, at 255 msec, was generically responsive to linguistic stimuli, whereas a component at 355 msec from the left medial middle frontal gyrus (channel MF 3-4) reflected the inflectional processing demanded by the task. This was also true of a component at 500 msec coming from a more lateral MFG region (channel MF 5-6). Finally, more ventral regions in the IFG and inferior Broca region (channel OF 5-6) showed later sustained responses, in the range of 700 to 1200 msec, that were sensitive to the task demands. From earlier to later components there was also a shift from greater significance for the Zero-Inflect > Read differences (indexing abstract grammatical features) to greater significance for the Overt-Inflect > Zero differences (adding phonological manipulations). Taken together, this pattern suggests several components of grammatical inflection processed by distinct neural arrays in

quick succession (though not necessarily strictly serially), with abstract morphosyntactic features computed before implementation of accompanying morphophonological content.

GENERAL DISCUSSION

Converging contrasts from fMRI and intracranial EEG recordings of people engaged in a task that is inherently grammatical yet minimally confounded with semantics, working memory, or articulation rehabilitates the hypothesis that Broca’s Area and adjacent cortical regions execute abstract grammatical computation. The computation executed there is abstract in the sense that it is engaged in response to demands for the instantiation of inflectional features issued by the syntax of the sentence rather than the pragmatic demands of the conversational context, and that it embraces both nouns and verbs, both regular and irregular forms, and both unaltered and overtly altered forms.

The current data are consistent with models of language organization in the brain that attribute a role to Broca’s area (and associated regions) in grammatical processing based on deficits in morphology following brain lesions and on more recent neuroimaging studies (e.g., (Newman et al., 2003; Embick et al., 2000) . At the same time, they do not show that grammar is the only function computed in Broca’s Area, that Broca’s Area is the only region implicated in grammatical processing, or that the language-related function of Broca’s Area is applicable only to language. What they show is that Broca’s Area appears to have a key role in the computation of grammar that cannot be attributed to generic cognitive processes.

The results, moreover, refine the functional mapping of grammatical computation in the left frontal regions of the brain. The anterior insula appears to be selective to manipulation of overt phonological material, along with the medial SMA and BA45. Parts of BA44 and BA47 were also involved in phonology, while distinct regions of BA44 (more ventral) and BA 47 (more dorsal) are tied to the computation of abstract grammatical features in the absence of overt phonology. The iEEG results converged to attribute grammatical computation to at least three generators of activity within the overall region of Broca’s area, which were maximally sensitive to the differing demands of the three inflectional tasks in the 350, 500, and 700-900 msec ranges after the onset of the word.

The findings show that temporal and spatial patterns of brain activity can in some ways, but not others, be transparently mapped onto decompositions of cognitive tasks arrived at by analyses of the computational requirements of the task. Successes in the present experiment include the fact that (1) overt inflection activated a superset of the areas activated by zero inflection (when both are compared to the Read task), (2) that the areas uniquely attributed to the manipulation of morphophonological material and to the



manipulation of abstract morphosyntactic factors summed to coincide with the areas activated by a task that requires both, (3) that regular and irregular inflection elicits different sets of brain areas, including some that are plausibly tied to the blocking of the regular rule by irregulars, and (4) that iEEG signals at times and places plausibly tied to grammatical computation reflect the morphological complexity of the task in their amplitude or duration. One of the failures of simple correspondence between task components and brain activation patterns is the fact that reading a word does not activate a subset of the areas activated by inflection but rather includes areas that might be engaged by special strategies ad hoc to the reading-aloud task. Another is the finding that most of the regions that differentiate regular and irregular inflection are not explained by otherwise plausible theories that attribute regular inflection to active grammatical computation in frontal regions and irregular inflection to passive lookup in temporal or parietal regions. This in turn suggests that more attention be given in future studies to the ways in which different regular and irregular forms enlist different combinations of lookup and computation.

Previously, the involvement of Broca’s Area in language has been attributed both to more general and to more specific factors. The generalized working memory models of (Just et al., 1996) would not appear to apply to our results, because the simple morphological task should not require heavy demands on working memory (compared to, say, a reading span task), and also because the different conditions should not vary the number, duration, or complexity of items held in mind. For similar reasons, the specifically parsing-related working memory of (Caplan and Waters, 1999) also would not seem to apply, because the experimental contrasts (especially Over-Inflect versus Zero-Inflect) do not vary the number of linguistic heads that must be maintained or the duration of the maintenance, nor is there ambiguity to resolve. Also unclear is how semantic selection demands, such as those invoked by (Thompson-Schill et al., 1997), would map to the current task. If anything, the Zero-inflect task required more features to be evaluated in order to determine the correct form (third person, plural number, present tense, and imperfective aspect) than did the Overt-inflect task (past tense). Yet it was the Overt-inflect task that evoked greater recruitment of inferior frontal regions. Moreover, though both inflection tasks involve morphosyntactic features, and such features do have minimal semantic content, the features do not have to have such content, such as in languages with inflections sensitive to grammatical gender or arbitrary declensions and conjugations. Even in English the semantic selection in inflection is secondary to the demands of grammatical computation (Does the clause have to be tensed? Does the noun appear with an article requiring number concord?) rather than in service of communication in the conversational context. Finally, general articulatory demands (Wise et al., 1999) are ruled out by the Zero-Inflect > Read

contrast, in which motor output would be identical and only the abstract grammatical features vary.

What about theories attributing greater specificity to Broca’s and associated areas? The hypothesis of Grodzinsky (1986b; Grodzinsky, 2000, 1989) that the role of Broca’s Area is purely the binding of traces and moved elements does not predict the present results, since the task involves only two or three word context frames without moved elements (at least in traditional theories of syntactic movement), nor do the contrasts among conditions involve moved elements. It is true that recent theories in the Chomskyan framework do attribute inflection to movement. Roughly, the inflectional morpheme moves from an underlying position as the head of an inflection phrase (such as a Tense Phrase) onto the head of the verb phrase in its complement position, in order to become bound as an enclitic tense-marking affix (see (Baker, 2001)). If so, the present task would in fact involve movement, consistent with Grodzinsky’s hypothesis. Note, however, that in these theories movement and traces are not restricted to the plausible relationships between antecedents and gaps, such as in passive constructions, relative clauses, and wh-questions, which have some degree of psycholinguistic support and which were the original motivation for Grodzinsky’s theory. Rather, movement is so ubiquitous in these theories (largely for reasons of theory-internal consistency) that it is virtually indistinguishable from grammar itself. Also note that theories of this type typically posit zero affixes, which occupy determinate positions in grammatical structures and are identical to overt affixes except in lacking a phonological specification. This removes any grammatical difference between the Overt-Inflect and Zero-Inflect conditions, and since, as noted, the latter carries more inflectional features than the former, it is hard to see how the greater activation elicited by the Overt-Inflection condition can be explained.

The question of whether Broca’s Area involvement in language is domain-general is separate from whether Broca’s Area itself has domain-general functionality. The first question pertains to whether the processes that activate Broca’s during language tasks (or which are destroyed when it is lesioned) are general cognitive abilities like working memory, or are specifically involved in grammatical computation. The current data speak to this question, suggesting that the function is not likely to be any of the domain-general explanations provided so far. The second question is whether the anatomical structures included under the label “Broca’s Area” function outside of language—a question about the roles of a large anatomical structure, rather than about one of the kinds of computation performed by one of its components. For example, Broca’s Area recently has been implicated in imitations of human movements (Iacoboni et al., 1999) and in musical syntax (Maess et al., 2001), among other tasks. But to interpret these findings, one must keep in mind that “Broca’s Area,” even in the most restrictive



definition, is a swath of tissue containing at least 100 million neurons, and it sits on prime real estate in terms of connectivity with known cognitive functional regions of the prefrontal cortex. While fMRI and lesion studies treat Broca’s area as a single module or a few units, it may in fact be composed of numerous functional subunits, which may not necessarily even be defined in terms of neatly defined macroscopic anatomical territories. This is one of the reasons human intra-cortical electroencephalography could become increasingly useful, by allowing tests of the range of functions that specifically engage a circumscribed generator.

Acknowledgments. This work was supported by grants

NIH HD 18381 (Pinker), NS 18741 (Halgren), The National Center for Research Resources (P41 RR14075), the Mental Illness and Neuroscience Discovery (MIND) Institute, and the Sackler Scholar Program in Psychobiology. The authors thank Doug Greve, Andre van der Kouwe, Suresh Narayanan, Chungmao Wang, Paul Raines, David N. Caplan, Verne S. Caviness, Jr., Evelina Busa, Cedric Boeckx, Larry Wald, Chris I. Moore, and the editors of this special issue of Cortex.

REFERENCES ALBRIGHT A and HAYES B. Rules vs. analogy in English past tenses:

a computational/experimental study. Cognition, 90: 119-61, 2003.

ALEGRE M and GORDON P. Frequency effects and the representational status of regular inflections. Journal of Memory and Language, 40: 41-61, 1999.

BAAYEN RH and SCHREUDER R. The balance of storage and computation in the mental lexicon: The case of morphological processing in language comprehension. In S. Nootebohm F.W. and Wijnen F. (Eds.), Storage and computation in the language faculty. Dordrect: Kluwer, in press.

BAKER MC. The Atoms of Language: The Mind's Hidden Rules of Grammar. New York: Basic Books, 2001.

BATES E and GOODMAN J. On the inseparability of grammar and the lexicon: Evidence from acquisition, aphasia, and real-time processing. Language and Cognitive Processes, 12, 1997.

BAYNES K, ELIASSEN JC, LUTSEP HL, and GAZZANIGA MS. Modular organization of cognitive systems masked by interhemispheric integration. Science, 280: 902-5, 1998.

BAYNES K, TRAMO MJ, and GAZZANIGA MS. Reading with a limited lexicon in the right hemisphere of a callosotomy patient. Neuropsychologia, 30: 187-200, 1992.

BERENT I, PINKER S, and SHIMRON J. The nature of regularity and irregularity: evidence from Hebrew nominal inflection. J Psycholinguist Res, 31: 459-502, 2002.

BERETTA A, CAMPBELL C, CARR TH, HUANG J, SCHMITT LM, CHRISTIANSON K, and CAO Y. An ER-fMRI investigation of morphological inflection in German reveals that the brain makes a distinction between regular and irregular forms. Brain Lang, 85: 67-92, 2003.

BERNDT RS and CARAMAZZA A. How "regular" is sentence comprehension in Broca's aphasia? It depends on how you select the patients. Brain Lang, 67: 242-7, 1999.

BIRD H, HOWARD D, and FRANKLIN S. Why is a verb like an inanimate object? Grammatical category and semantic category deficits. Brain Lang, 72: 246-309, 2000.

BIRD H, HOWARD D, and FRANKLIN S. Noun-verb differences? a question of semantics: a response to shapiro and caramazza. Brain Lang, 76: 213-22, 2001.

BROCA P. Remarques sur le siége de la faculté du langage articulé, suivies d'une observation d'aphémie (perte de la parole). Bulletin de la Société Anatomique, 6: 330-357, 1861.

BUCKNER RL. Event-related fMRI and the hemodynamic response. Hum Brain Mapp, 6: 373-7, 1998.

BUROCK MA, BUCKNER RL, WOLDORFF MG, ROSEN BR, and DALE AM. Randomized event-related experimental designs allow for extremely rapid presentation rates using functional MRI. Neuroreport, 9: 3735-9, 1998.

BUROCK MA and DALE AM. Estimation and detection of event-related fMRI signals with temporally correlated noise: a statistically efficient and unbiased approach. Hum Brain Mapp, 11: 249-60, 2000.

CAPLAN D, ALPERT N, WATERS G, and OLIVIERI A. Activation of Broca's area by syntactic processing under conditions of concurrent articulation. Hum Brain Mapp, 9: 65-71, 2000.

CAPLAN D and WATERS GS. Verbal working memory and sentence comprehension. Behav Brain Sci, 22: 77-94; discussion 95-126, 1999.

CAPPA SF, SANDRINI M, ROSSINI PM, SOSTA K, and MINIUSSI C. The role of the left frontal lobe in action naming: rTMS evidence. Neurology, 59: 720-3, 2002.

CARAMAZZA A and HILLIS AE. Lexical organization of nouns and verbs in the brain. Nature, 349: 788-90, 1991.

CARAMAZZA A, LAUDANNA A, and ROMANI C. Lexical access and inflectional morphology. Cognition, 28: 297-332, 1988.

CARAMAZZA A and SHAPIRO KA. The representation of grammatical knowledge in the brain. In Jenkins L. (Ed.) Variation and universals in biolinguistics. Amsterdam: Elsevier, in press.

CARAMAZZA A and ZURIF EB. Dissociation of algorithmic and heuristic processes in language comprehension: evidence from aphasia. Brain Lang, 3: 572-82, 1976.

CAVINESS VS, MAKRIS N, MONTINARO E, SAHIN NT, BATES JF, SCHWAMM L, CAPLAN D, and KENNEDY DN. Anatomy of stroke, Part I: an MRI-based topographic and volumetric System of analysis. Stroke, 33: 2549-56, 2002.

CHEE MW, HON NH, CAPLAN D, LEE HL, and GOH J. Frequency of concrete words modulates prefrontal activation during semantic judgments. NeuroImage, 16: 259-68, 2002.

CHEE MW, WESTPHAL C, GOH J, GRAHAM S, and SONG AW. Word frequency and subsequent memory effects studied using event-related fMRI. NeuroImage, 20: 1042-51, 2003.

CHOCHON F, COHEN L, VAN DE MOORTELE PF, and DEHAENE S. Differential contributions of the left and right inferior parietal lobules to number processing. J Cogn Neurosci, 11: 617-30, 1999.

CHOMSKY N and HALLE M. The sound pattern of English. Cambridge, MA: MIT Press, 1968/1991.

CHURCH K and HANKS P. Word Association Norms, Mutual Information and Lexicography. Computational Linguistics, 16: 22-29, 1991.

COHEN L and DEHAENE S. Specialization within the ventral stream: the case for the visual word form area. NeuroImage, 22: 466-76, 2004.

COLTHEART M. The MRC Psycholinguistic Database. Quarterly Journal of Experimental Psychology, 33A: 497-505, 1981.

CORNELL TL, FROMKIN VA, and MAUNER G. The syntax-there-but-not-there paradox: A linguistic account. Current Directions in Psychological Science, 1993.

COX RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res, 29: 162-73, 1996.

DALE AM. Optimal experimental design for event-related fMRI. Hum Brain Mapp, 8: 109-14, 1999.

DALE AM and BUCKNER RL. Selective averaging of rapidly presented individual trials using fMRI. Hum Brain Mapp, 5: 329-340, 1997.

DALE AM, FISCHL B, and SERENO MI. Cortical surface-based analysis. I. Segmentation and surface reconstruction. NeuroImage, 9: 179-94, 1999.

DAMASIO AR. Aphasia. N Engl J Med, 326: 531-9, 1992. DAMASIO H. The lesion method in cognitive neuroscience. In Boller

F. and Grafman J. (Eds.), Handbook of neuropsychology, 2000. DEHAENE S, LE CLEC HG, POLINE JB, LE BIHAN D, and COHEN L.

The visual word form area: a prelexical representation of visual words in the fusiform gyrus. Neuroreport, 13: 321-5, 2002.

DEHAENE S, TZOURIO N, FRAK V, RAYNAUD L, COHEN L, MEHLER J, and MAZOYER B. Cerebral activations during number multiplication and comparison: a PET study. Neuropsychologia, 34: 1097-106, 1996.

DEMB JB, DESMOND JE, WAGNER AD, VAIDYA CJ, GLOVER GH, and GABRIELI JD. Semantic encoding and retrieval in the left inferior prefrontal cortex: a functional MRI study of task difficulty and process specificity. J Neurosci, 15: 5870-8, 1995.

DESMOND JE, GABRIELI JD, and GLOVER GH. Dissociation of frontal and cerebellar activity in a cognitive task: evidence for a distinction between selection and search. NeuroImage, 7: 368-76, 1998.

DEVOR A, DUNN AK, ANDERMANN ML, ULBERT I, BOAS DA, and DALE AM. Coupling of total hemoglobin concentration, oxygenation, and neural activity in rat somatosensory cortex. Neuron, 39: 353-9, 2003.



DOGIL G, MAYER, J., ACKERMANN, H., WILDGRUBER, D., RIECKER, A. Overt speech, insular cortex and apraxia of speech. Brain Lang, 86: 418, 2003.

DRONKERS N, PINKER S, and DAMASIO AR. Language and the aphasias. In E. R. Kandel J.H.S. and Jessell T.M. (Eds.), Principles of neural science. Norwalk, CT: Appleton;Lange, 1999.

DRONKERS N, PINKER, S., AND DAMASIO, A. Language and the Aphasias. In Kandel E.R., Schwartz, J.H., and Jessell, T.M. (Ed.) Principles of Neural Science. New York: McGraw-Hill, 2000, pp. 1169-1187.

DRONKERS NF. A new brain region for coordinating speech articulation. Nature, 384: 159-61, 1996.

EMBICK D, MARANTZ A, MIYASHITA Y, O'NEIL W, and SAKAI KL. A syntactic specialization for Broca's area. Proc Natl Acad Sci U S A, 97: 6150-4, 2000.

FIEBACH CJ and FRIEDERICI AD. Processing concrete words: fMRI evidence against a specific right-hemisphere involvement. Neuropsychologia, 42: 62-70, 2004.

FIEBACH CJ, FRIEDERICI AD, MULLER K, and VON CRAMON DY. fMRI evidence for dual routes to the mental lexicon in visual word recognition. J Cogn Neurosci, 14: 11-23, 2002.

FISCHL B, LIU A, and DALE AM. Automated manifold surgery: constructing geometrically accurate and topologically correct models of the human cerebral cortex. IEEE Trans Med Imaging, 20: 70-80, 2001.

FISCHL B, SERENO MI, and DALE AM. Cortical surface-based analysis. II: Inflation, flattening, and a surface-based coordinate system. NeuroImage, 9: 195-207, 1999a.

FISCHL B, SERENO MI, TOOTELL RB, and DALE AM. High-resolution intersubject averaging and a coordinate system for the cortical surface. Hum Brain Mapp, 8: 272-84, 1999b.

FITCH WT and HAUSER MD. Computational constraints on syntactic processing in a nonhuman primate. Science, 303: 377-80, 2004.

FRANCIS WN, & KUCERA, H. Frequency Analysis of English Usage. Boston: Houghton Mifflin, 1982.

FRIEDMANN N and GRODZINSKY Y. Tense and agreement in agrammatic production: pruning the syntactic tree. Brain Lang, 56: 397-425, 1997.

GENTNER D. Some interesting differences between verbs and nouns. Cognition and Brain Theory, 4, 1981.

GESCHWIND N. The organization of language and the brain. Science, 170: 940-4, 1970.

GESCHWIND N. Specializations of the human brain. Sci Am, 241: 180-99, 1979.

GOFF WR, ALLISON TA, and VAUGHAN HG, JR. The functional neuroanatomy of event related potentials. In Callaway E., Tueting P., and Koslow S.H. (Eds.), Event-related brain potentials in man. New York: Academic Press, 1978, pp. 1-79.

GOODGLASS H. Studies on the grammar of aphasics. In Goodglass H. and Blumstein S.E. (Eds.), Psycholinguistics and aphasia. Baltimore: Johns Hopkins University Press, 1973.

GOODGLASS H. Understanding aphasia. San Diego: Academic Press, 1993.

GRODZINSKY Y. Cognitive deficits, their proper description, and its theoretical relevance. Brain Lang, 27: 178-91, 1986a.

GRODZINSKY Y. Language deficits and the theory of syntax. Brain Lang, 27: 135-59, 1986b.

GRODZINSKY Y. Agrammatic comprehension of relative clauses. Brain Lang, 37: 480-99, 1989.

GRODZINSKY Y. The neurology of syntax: language use without Broca's area. Behav Brain Sci, 23: 1-21; discussion 21-71, 2000.

GROSS M, SAY T, KLEINGERS M, CLAHSEN H, and MUNTE TF. Human brain potentials to violations in morphologically complex Italian words. Neurosci Lett, 241: 83-6, 1998.

HALGREN E, BAUDENA P, HEIT G, CLARKE JM, MARINKOVIC K, and CLARKE M. Spatio-temporal stages in face and word processing. I. Depth-recorded potentials in the human occipital, temporal and parietal lobes [corrected]. J Physiol Paris, 88: 1-50, 1994.

HALGREN E, DHOND RP, CHRISTENSEN N, VAN PETTEN C, MARINKOVIC K, LEWINE JD, and DALE AM. N400-like magnetoencephalography responses modulated by semantic context, word frequency, and lexical class in sentences. NeuroImage, 17: 1101-16, 2002.

HALLE M and MOHANAN KP. Segmental phonology of modern English. Linguistic Inquiry, 16: 57-116, 1985.

HAY J. Lexical frequency in morphology: Is everything relative? Linguistics, 39: 1041-1070, 2001.

HOLROYD CB, NIEUWENHUIS S, YEUNG N, NYSTROM L, MARS RB, COLES MG, and COHEN JD. Dorsal anterior cingulate cortex shows fMRI response to internal and external error signals. Nat Neurosci, 7: 497-8, 2004.

IACOBONI M, WOODS RP, BRASS M, BEKKERING H, MAZZIOTTA JC, and RIZZOLATTI G. Cortical mechanisms of human imitation. Science, 286: 2526-8, 1999.

INDEFREY P, BROWN C, HAGOORT P, HERZOG H, SACH M, and SEITZ RJ. A PET study of cerebral activation patterns induced by verb inflection. NeuroImage, 5: S548, 1997.

IVES JR, MAINWARING NR, GRUBER LJ, COSGROVE GR, BLUME HW, and SCHOMER DL. 128-channel cable-telemetry EEG recording system for long-term invasive monitoring. Electroencephalogr Clin Neurophysiol, 79: 69-72, 1991.

JAEGER JJ, LOCKWOOD AH, KEMMERER DL, VAN VALIN RD, MURPHY BW, and KHALAK HG. A positron emission tomography study of regular and irregular verb morphology in English. Language, 72: 451-497, 1996.

JOANISSE MF and SEIDENBERG MS. Impairments in verb morphology after brain injury: a connectionist model. Proc Natl Acad Sci U S A, 96: 7592-7, 1999.

JOUBERT S, BEAUREGARD M, WALTER N, BOURGOUIN P, BEAUDOIN G, LEROUX JM, KARAMA S, and LECOURS AR. Neural correlates of lexical and sublexical processes in reading. Brain Lang, 89: 9-20, 2004.

JOVICICH J, PETERS RJ, KOCH C, BRAUN J, CHANG L, and ERNST T. Brain areas specific for attentional load in a motion-tracking task. J Cogn Neurosci, 13: 1048-58, 2001.

JUST MA and CARPENTER PA. A capacity theory of comprehension: individual differences in working memory. Psychol Rev, 99: 122-49, 1992.

JUST MA, CARPENTER PA, and KELLER TA. The capacity theory of comprehension: new frontiers of evidence and arguments. Psychol Rev, 103: 773-80, 1996.

KAAN E and SWAAB TY. The brain circuitry of syntactic comprehension. Trends Cogn Sci, 6: 350-356, 2002.

KERNS JG, COHEN JD, MACDONALD AW, 3RD, CHO RY, STENGER VA, and CARTER CS. Anterior cingulate conflict monitoring and adjustments in control. Science, 303: 1023-6, 2004a.

KERNS JG, COHEN JD, STENGER VA, and CARTER CS. Prefrontal Cortex Guides Context-Appropriate Responding during Language Production. Neuron, 43: 283-91, 2004b.

LINEBARGER MC, SCHWARTZ MF, and SAFFRAN EM. Sensitivity to grammatical structure in so-called agrammatic aphasics. Cognition, 13: 361-92, 1983a.

LINEBARGER MC, SCHWARTZ MF, and SAFFRAN EM. Syntactic processing in agrammatism: A reply to Zurif and Grodzinsky. Cognition, 15: 215-225, 1983b.

LOGOTHETIS NK. The neural basis of the blood-oxygen-level-dependent functional magnetic resonance imaging signal. Philos Trans R Soc Lond B Biol Sci, 357: 1003-37, 2002.

LUZZATTI C, RAGGI R, ZONCA G, PISTARINI C, CONTARDI A, and PINNA GD. Verb-noun double dissociation in aphasic lexical impairments: the role of word frequency and imageability. Brain Lang, 81: 432-44, 2002.

MAESS B, KOELSCH S, GUNTER TC, and FRIEDERICI AD. Musical syntax is processed in Broca's area: an MEG study. Nat Neurosci, 4: 540-5, 2001.

MARCUS GF, PINKER S, ULLMAN M, HOLLANDER M, ROSEN TJ, and XU F. Overregularization in language acquisition. Monographs of the Society for Research in Child Development, 57, 1992.

MARSLEN-WILSON WD and TYLER LK. Dissociating types of mental computation. Nature, 387: 592-4, 1997.

MARSLEN-WILSON WD and TYLER LK. Rules, representations, and the English past tense. Trends in Cognitive Science, 2: 428-435, 1998.

MARTIN A, WIGGS CL, UNGERLEIDER LG, and HAXBY JV. Neural correlates of category-specific knowledge. Nature, 379: 649-52, 1996.

MARTINEZ-MONTES E, VALDES-SOSA PA, MIWAKEICHI F, GOLDMAN RI, and COHEN MS. Concurrent EEG/fMRI analysis by multiway Partial Least Squares. NeuroImage, 22: 1023-34, 2004.

MCCANDLISS BD, COHEN L, and DEHAENE S. The visual word form area: expertise for reading in the fusiform gyrus. Trends Cogn Sci, 7: 293-299, 2003.

MCCLELLAND JL and PATTERSON K. Rules or connections in past-tense inflections: what does the evidence rule out? Trends Cogn Sci, 6: 465-472, 2002.

MELLET E, TZOURIO N, DENIS M, and MAZOYER B. Cortical anatomy of mental imagery of concrete nouns based on their dictionary definition. Neuroreport, 9: 803-8, 1998.

MILLER EK and COHEN JD. An integrative theory of prefrontal cortex function. Annu Rev Neurosci, 24: 167-202, 2001.

MIOZZO M. On the processing of regular and irregular forms of verbs and nouns: evidence from neuropsychology. Cognition, 87: 101-27, 2003.

MIWAKEICHI F, MARTINEZ-MONTES E, VALDES-SOSA PA, NISHIYAMA N, MIZUHARA H, and YAMAGUCHI Y. Decomposing EEG data into space-time-frequency components using Parallel Factor Analysis. NeuroImage, 22: 1035-45, 2004.



MUNTE TF, SAY T, CLAHSEN H, SCHILTZ K, and KUTAS M. Decomposition of morphologically complex words in English: evidence from event-related brain potentials. Brain Res Cogn Brain Res, 7: 241-53, 1999.

NEWMAN SD, JUST MA, KELLER TA, ROTH J, and CARPENTER PA. Differential effects of syntactic and semantic processing on the subregions of Broca's area. Brain Res Cogn Brain Res, 16: 297-307, 2003.

NOWAK MA, PLOTKIN JB, and JANSEN VA. The evolution of syntactic communication. Nature, 404: 495-8, 2000.

NUNEZ PL. Electric Fields of the Brain. New York: Oxford University, 1981.

PAIVIO A, YUILLE JC, and MADIGAN SA. Concreteness, imagery, and meaningfulness values for 925 nouns. J Exp Psychol, 76: Suppl:1-25, 1968.

PENKE M, WEYERTS H, GROSS M, ZANDER E, MUNTE TF, and CLAHSEN H. How the brain processes complex words: an event-related potential study of German verb inflections. Brain Res Cogn Brain Res, 6: 37-52, 1997.

PERANI D, CAPPA SF, SCHNUR T, TETTAMANTI M, COLLINA S, ROSA MM, and FAZIO F. The neural correlates of verb and noun processing. A PET study. Brain, 122 ( Pt 12): 2337-44, 1999.

PETERSEN SE, VAN MIER H, FIEZ JA, and RAICHLE ME. The effects of practice on the functional anatomy of task performance. Proc Natl Acad Sci U S A, 95: 853-60, 1998.

PINKER S. Rules of language. Science, 253: 530-5, 1991. PINKER S. Words and rules: The ingredients of language. New York:

HarperCollins, 1999. PINKER S and JACKENDOFF R. The faculty of language: What's

special about it? Cognition, in press. PINKER S and ULLMAN MT. The past and future of the past tense.

Trends Cogn Sci, 6: 456-463, 2002. POLDRACK RA, WAGNER AD, PRULL MW, DESMOND JE, GLOVER

GH, and GABRIELI JD. Functional specialization for semantic and phonological processing in the left inferior prefrontal cortex. NeuroImage, 10: 15-35, 1999.

PROVERBIO AM, VECCHI L, and ZANI A. From orthography to phonetics: ERP measures of grapheme-to-phoneme conversion mechanisms in reading. J Cogn Neurosci, 16: 301-17, 2004.

PULVERMULLER F, LUTZENBERGER W, and PREISSL H. Nouns and verbs in the intact brain: evidence from event-related potentials and high-frequency cortical responses. Cereb Cortex, 9: 497-506, 1999a.

PULVERMULLER F, MOHR B, and SCHLEICHERT H. Semantic or lexico-syntactic factors: what determines word-class specific activity in the human brain? Neurosci Lett, 275: 81-4, 1999b.

PULVERMULLER F, PREISSL H, LUTZENBERGER W, and BIRBAUMER N. Brain rhythms of language: nouns versus verbs. Eur J Neurosci, 8: 937-41, 1996.

PULVERMULLER F, SHTYROV Y, KUJALA T, and NAATANEN R. Word-specific cortical activity as revealed by the mismatch negativity. Psychophysiology, 41: 106-12, 2004.

RAICHLE ME, MACLEOD AM, SNYDER AZ, POWERS WJ, GUSNARD DA, and SHULMAN GL. A default mode of brain function. Proc Natl Acad Sci U S A, 98: 676-82, 2001.

RHEE J. Words and Rules in the Brain. Secondary Titl. Cambridge, MA: MIT: 147, 2001.

RHEE J, HALGREN E, and PINKER S. Neural time course of English past-tense generation: Combined fMRI and MEG analyses. Secondary Titl. New York, 2003.

RHEE J, PINKER S, and ULLMAN MT. A magnetoencephalographic study of English past tense production. Secondary Titl. Washington, 1999.

RUMELHART DE and MCCLELLAND JL. On learning the past tenses of English verbs. In McClelland J.L. and Rumelhart D.E. (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition. Cambridge, MA: MIT Press, 1986.

SAHIN NT. Seeking the Neural Basis of Grammar: English Noun and Verb Morphological Processing Investigated with Rapid Event-Related fMRI and Intracortical Electrophysiology. Secondary Titl. Cambridge, MA: MIT: 271, 2003.

SAHIN NT, MAKRIS N, BATES JF, PATTI MR, MEYER JW, KENNEDY DN, CAPLAN DN, and CAVINESS VS, JR. MRI-Based Topographic and Quantitative Mapping of Stroke. Secondary Titl. Montreal: NeuroImage. 7 (4): S692, 1998.

SCHOMER DL. Advances in EEG Telemetry. In Hallet M., Phillips L., Massey J., and Schomer D.L. (Eds.), Advances in Clinical Neurophysiology: Elsevier, in press.

SEIDENBERG MS and HOEFFNER J. Evaluating behavioral and neuroimaging data on past tense processing. Language, 74, 1998.

SENGHAS A, KIM JJ, and PINKER S. The plurals-in-compounds effect. Secondary Titl. Cambridge, MA: Department of Psychology, Harvard University, 2004.

SHAPIRO K and CARAMAZZA A. Grammatical processing of nouns and verbs in left frontal cortex? Neuropsychologia, 41: 1189-98, 2003.

SHAPIRO KA, PASCUAL-LEONE A, MOTTAGHY FM, GANGITANO M, and CARAMAZZA A. Grammatical distinctions in the left frontal cortex. J Cogn Neurosci, 13: 713-20, 2001.

SMITH EE, JONIDES J, MARSHUETZ C, and KOEPPE RA. Components of verbal working memory: evidence from neuroimaging. Proc Natl Acad Sci U S A, 95: 876-82, 1998.

STROMSWOLD K, CAPLAN D, ALPERT N, and RAUCH S. Localization of syntactic comprehension by positron emission tomography. Brain Lang, 52: 452-73, 1996.

THOMPSON-SCHILL SL, D'ESPOSITO M, AGUIRRE GK, and FARAH MJ. Role of left inferior prefrontal cortex in retrieval of semantic knowledge: a reevaluation. Proc Natl Acad Sci U S A, 94: 14792-7, 1997.

TONI I, SHAH NJ, FINK GR, THOENISSEN D, PASSINGHAM RE, and ZILLES K. Multiple movement representations in the human brain: an event-related fMRI study. J Cogn Neurosci, 14: 769-84, 2002.

TURKELTAUB PE, EDEN GF, JONES KM, and ZEFFIRO TA. Meta-analysis of the functional neuroanatomy of single-word reading: method and validation. NeuroImage, 16: 765-80, 2002.

TYLER LK, BRIGHT P, FLETCHER P, and STAMATAKIS EA. Neural processing of nouns and verbs: the role of inflectional morphology. Neuropsychologia, 42: 512-23, 2004.

TYLER LK, DEMORNAY-DAVIES P, ANOKHINA R, LONGWORTH C, RANDALL B, and MARSLEN-WILSON WD. Dissociations in processing past tense morphology: neuropathology and behavioral studies. J Cogn Neurosci, 14: 79-94, 2002a.

TYLER LK, RANDALL B, and MARSLEN-WILSON WD. Phonology and neuropsychology of the English past tense. Neuropsychologia, 40: 1154-66, 2002b.

TYLER LK, RUSSELL R, FADILI J, and MOSS HE. The neural representation of nouns and verbs: PET studies. Brain, 124: 1619-34, 2001.

ULLMAN M, CORKIN S, COPPOLA M, HICKOK G, GROWDON JH, KOROSHETZ WJ, and PINKER S. A neural dissociation within language: Evidence that the mental dictionary is part of declarative memory, and that grammatical rules are processed by the procedural system. Journal of Cognitive Neuroscience, 9: 289-299, 1997.

ULLMAN MT. Acceptability ratings of regular and irregular past-tense forms: Evidence for a dual-system model of language from word frequency and phonological neighborhood effects. Language and Cognitive Processes, 14: 47-67, 1999.

ULLMAN MT, IZVORSKI R, LOVE T, YEE E, SWINNEY D, and HICKOK G. Neural correlates of lexicon and grammar: Evidence from the production, reading, and judgment of inflection in aphasia. Brain and Language, in press.

WERNICKE C. Der aphasische Symptomenkompleks. Breslau: Franck und Weigert, 1874.

WEYERTS H, PENKE M, DOHRN U, CLAHSEN H, and MUNTE TF. Brain potentials indicate differences between regular and irregular German plurals. Neuroreport, 8: 957-62, 1997.

WISE RJ, GREENE J, BUCHEL C, and SCOTT SK. Brain regions involved in articulation. Lancet, 353: 1057-61, 1999.

WOJCIULIK E and KANWISHER N. The generality of parietal involvement in visual attention. Neuron, 23: 747-64, 1999.

ZURIF E and GRODZINSKY Y. Sensitivity to grammatical structure in agrammatic aphasics: a reply to Linebarger, Schwartz and Saffran. Cognition, 15: 207-13, 215-25, 1983.

ZURIF EB, CARAMAZZA A, and MYERSON R. Grammatical judgments of agrammatic aphasics. Neuropsychologia, 10: 405-17, 1972.

Ned T. Sahin, Department of Psychology, Harvard University, William James Hall 964, 33 Kirkland Street, Cambridge, MA 02138, USA. e-mail: [email protected]



Appendix

Infl Freq

Stem Freq

Class Ambig

Infl Freq

Stem Freq

Class Ambig

Infl Freq

Stem Freq

Class Ambig

Infl Freq

Stem Freq

Class Ambig

apples 6 15 1.00 analyses 13 121 1.00 begged 13 34 1.00 bore 28 242 0.83arches 5 16 0.03 bacteria 8 8 1.00 called 401 626 0.92 bent 31 50 0.59assets 13 18 1.00 bluefish 1 1 1.00 cared 15 108 0.10 bound 37 51 0.92axes 7 13 1.00 bookshelves 3 4 1.00 caused 90 186 0.26 broke 67 225 0.79bats 5 18 -0.12 cacti 1.00 changed 95 225 -0.12 bred 1 6 -0.43bellows 1 3 -0.50 calves 6 17 1.00 cried 30 64 0.29 brought 253 487 1.00braces 4 13 0.24 chairmen 9 76 1.00 dared 14 44 1.00 built 103 248 0.98breeches 1 1 1.00 children 355 565 1.00 dropped 101 159 0.65 bought 56 160 0.99buttons 10 20 0.91 crises 21 102 1.00 drowned 6 14 1.00 caught 98 145 0.93canals 1 4 1.00 criteria 11 22 1.00 dried 28 70 1.00 clung 14 30 1.00caps 5 22 0.57 curricula 3 20 1.00 dyed 4 36 1.00 crept 11 27 0.80claps 1 2 -0.64 data 96 97 1.00 failed 74 142 1.00 dealt 22 124 0.11clips 2 8 0.46 diagnoses 1 14 1.00 fanned 4 13 -0.44 dug 15 32 0.94clumps 4 8 1.00 elves 1.00 frowned 8 22 1.00 dove 0 11 -0.37democracies 1 25 1.00 emphases 2 60 1.00 glued 19 20 0.48 ate 16 121 1.00dives 4 24 0.37 fieldmice 1 1 1.00 helped 66 352 0.56 fed 41 132 0.34divers 2 3 1.00 firemen 5 6 1.00 hoped 48 164 0.09 felt 356 643 0.95dogs 69 145 0.99 feet 283 353 0.99 joined 56 138 1.00 fought 46 154 0.45dragons 2 3 1.00 freshmen 3 11 1.00 jumped 35 58 0.71 flung 14 16 1.00drawers 5 13 1.00 fungi 1.00 looked 367 906 0.79 flew 27 92 0.60dumps 1 2 -0.75 gentlemen 21 23 1.00 owed 15 34 1.00 froze 10 59 0.97dynamics 4 5 1.00 geese 3 7 1.00 passed 157 298 0.80 ground 16 26 1.00earnings 19 19 1.00 grandchildren 6 6 1.00 planned 75 200 -0.16 grew 65 300 1.00farms 16 137 0.79 halves 2 20 1.00 played 104 332 0.46 hid 12 68 0.79fives 2 2 1.00 hooves 2 11 1.00 poured 29 47 1.00 held 264 508 0.90galaxies 7 10 1.00 hypotheses 4 22 1.00 pulled 73 145 0.84 kept 186 523 0.98ganders 1 1 1.00 knives 7 83 1.00 raised 101 187 0.92 lent 5 29 1.00goings 1 4 1.00 laymen 6 9 1.00 roared 18 27 0.39 lost 173 274 1.00graves 6 20 1.00 leaves 2 14 0.87 rolled 48 88 0.59 meant 100 375 0.93gripes 1 1 1.00 loaves 3 8 0.78 sailed 10 33 0.65 rang 21 43 1.00impurities 9 11 1.00 men 763 1966 0.98 scored 15 35 -0.30 ran 134 429 0.64innings 4 16 1.00 moose 1.00 seemed 333 831 1.00 sought 55 178 1.00leads 2 49 -0.73 mice 10 20 1.00 shared 40 105 0.02 sold 47 99 0.98links 7 19 -0.14 neuroses 4 10 1.00 sighed 22 28 0.44 sent 145 253 1.00mains 1 11 1.00 nuclei 13 24 1.00 signed 37 62 -0.39 shot 52 114 0.98markings 2 3 1.00 oases 2 2 1.00 slipped 32 47 0.42 sang 56 125 1.00medics 1 1 1.00 oxen 10 16 1.00 snapped 19 38 0.95 sank 18 46 0.59moors 1 1 -0.50 parentheses 1 1 1.00 sprayed 6 14 -0.03 slept 27 97 0.48mops 1 2 -0.64 people 18 193 1.00 stared 60 95 0.88 slid 24 43 0.43pants 9 9 -0.05 phenomena 26 61 1.00 stayed 75 194 0.85 slung 2 3 0.20panties 1 1 1.00 prognoses 1 3 1.00 stepped 40 71 -0.53 spoke 86 310 1.00papers 51 207 0.99 radii 4 13 1.00 stirred 15 39 1.00 spent 104 194 1.00pens 2 18 0.57 salesmen 19 31 1.00 stopped 129 240 0.77 spun 16 31 0.82piles 1 23 -0.06 salmon 1.00 stored 36 47 -0.37 stole 10 38 1.00pleas 3 14 1.00 schoolchildren 1 1 1.00 strained 11 19 -0.33 stuck 23 50 0.09primates 1 1 1.00 selves 4 40 1.00 stripped 17 22 -0.28 stung 2 6 0.09riches 2 3 1.00 series 10 130 1.00 swayed 9 13 0.53 strode 20 25 0.14scopes 1 28 1.00 sheep 7 23 1.00 talked 58 274 0.67 struck 118 175 0.65shingles 5 5 1.00 shelves 8 20 1.00 tied 34 50 0.30 strung 4 7 -0.66sneakers 3 5 1.00 species 14 37 1.00 tried 170 470 0.97 swore 14 33 1.00stoves 2 17 1.00 stimuli 5 20 1.00 urged 35 64 0.73 swept 34 54 0.74taxis 3 19 0.81 subspecies 3 3 1.00 used 611 1016 0.45 swam 6 55 0.96teens 5 11 1.00 theses 1 11 1.00 vied 1 5 1.00 swung 48 75 0.70theories 20 150 1.00 thieves 9 17 1.00 viewed 25 55 -0.60 taught 50 153 1.00tongs 1 1 1.00 teeth 103 123 1.00 walked 159 286 0.76 told 413 758 1.00traps 7 27 0.46 vertebrae 1 1 1.00 watched 81 209 0.74 thought 414 870 1.00trousers 7 10 1.00 wharves 2 4 1.00 weighed 16 33 0.00 threw 46 146 0.91valves 4 7 1.00 wives 21 249 1.00 whipped 12 24 0.23 wept 9 28 1.00wonders 6 34 -0.55 wolves 4 9 1.00 wished 56 161 0.65 won 68 158 0.95wounds 8 24 -0.04 women 195 418 1.00 worked 128 492 -0.16 wrote 181 559 1.00

Verb Irregular

FKFKFK Inflected Form

Inflected Form

FK

Noun Regular Noun Irregular Verb Regular

Inflected Form

Inflected Form

Documents

abstract grammatical processing in broca's area: convergent