Research report How the brain processes complex words: an...

Ž .Cognitive Brain Research 6 1997 37–52

Research report

How the brain processes complex words: an event-related potential studyof German verb inflections

Martina Penke a, Helga Weyerts b, Matthias Gross c, Elke Zander c, Thomas F. Munte c,d,¨Harald Clahsen e,)

a Department of Linguistics, UniÕersity of Dusseldorf, 40225 Dusseldorf, Germany¨ ¨b Department of Clinical Neurophysiology, UniÕersity of Magdeburg, 39120 Magdeburg, Germany

c Department of Neurology, Medical School of HannoÕer, 30625 HannoÕer, Germanyd Department of CognitiÕe Science, UniÕersity of California, San Diego, La Jolla, CA 92093-0515, USA

e Department of Language and Linguistics, UniÕersity of Essex, Colchester C04 3SQ, UK

Accepted 4 March 1997

Abstract

Ž .Event-related brain potentials ERPs were recorded as German-speaking subjects read verbs in correct and incorrect participle forms.The critical words were presented in three different versions to three different groups of subjects, as part of a simple sentence, in a wordlist, and embedded in a story; for each version separate ERPs were recorded. Three types of verbs were investigated, regulars, irregulars

( )and nonce verbs. We compared correct regular and irregular participles with incorrect ones; the latter had - e n on verbs that actually takeŽ) . ( ) ( ) Ž) .-t participles getanz-en , or - e t on verbs that require - e n gelad-et . For the nonce verbs, we compared participles with the

( )unexpected - e n ending with the expected -t participle forms. The ERP responses were very consistent across the three versions of theŽ . Ž) . Ž . Ž) .experiment: i incorrect irregular participles gelad-et elicited a left frontotemporal negativity; ii incorrect regulars getanz-en

Ž .produced no differences to the correct ones; iii nonce verbs were associated with an N400 component but did not show a differencebetween expected and unexpected endings. We will interpret these findings with respect to psycholinguistic models of morphologicalprocessing and argue that the brain processes regularly inflected words differently from irregularly inflected ones, the latter by accessingfull-form entries stored in memory and the former by a computational process that decomposes complex words into stems and affixes.q 1997 Elsevier Science B.V.

Keywords: Event-related potential; Language processing; Left anterior negativity; Regular and irregular inflection; Psycholinguistics; Morphology oflanguage

1. Introduction

One central issue in psycholinguistic research concernsthe question of whether regularities of language involveinternally represented symbolic rules or whether they canbetter be explained in terms of assemblies of stored units.Consider, for example, past tense formation in English.

Ž .Under the latter view, both regular fax-ed, walk-ed, etc.Ž .and irregular past tense forms went, came, saw are said

to be represented and processed in the same way, by thestorage of items and the creation of connections amongthem. On the symbol-manipulation view, however, regular

) Ž .Corresponding author. Fax: q44 1206 87-2085; e-mail: [email protected]

past tense forms are explained differently from irregulars,Žby a symbolic rule that instantiates a variable the stem of

.a verb with an instance, e.g. fax, and concatenates thatvariable with the -ed affix, yielding faxed. More generally,the controversy revolves around the question to whatextent the recognition and production of inflected wordsinvolve parsing processes, i.e., the segmentation into stemsand affixes, and to what extent it involves accessingwhole-word based representations stored in memory.

To date, the processing of morphologically complexwords has mainly been investigated by taking reaction-time

Žmeasures in various kinds of behavioral tasks e.g. lexical.decision, priming, naming . The results of these experi-

ments are discussed controversially with respect to threeprocessing models, cf. Section 2. Another set of evidencethat might bear on the theoretical debate on morphological

0926-6410r97r$17.00 q 1997 Elsevier Science B.V. All rights reserved.Ž .PII S0926-6410 97 00012-8

( )M. Penke et al.rCognitiÕe Brain Research 6 1997 37–5238

processing comes from modern brain imaging techniques.These can be roughly divided in those techniques thatprovide high spatial resolution but very low temporalresolution, and those that provide a comparatively lowspatial resolution but excellent temporal resolution. Theformer methods include functional MRI with a temporalresolution in the order of 10 s and stimulation PET usingradioactively labelled water or butanol with a temporalresolution of about 1 min, while the latter consist mostly ofelectrophysiological techniques, in particular the event-re-

Ž .lated brain potentials ERPs . The first published func-tional imaging study of processes underlying regular and

w xirregular morphology was done by Jaeger et al. 18 . In thisstudy, the PET technique was applied to a production taskin which subjects had to generate past tense forms of

Žregular, irregular and nonce verbs in English seew x8,45,46,48 for earlier PET studies on non-inflected

. w xwords . Jaeger et al. 18 found differences in the activa-tion patterns between regular and irregular verbs which,they argue, support the view that regular past tense formsare represented differently from irregular ones and involvethe processing of symbolic rules. These conclusions are,however, not completely borne out, as we will show inSection 5.3.

For the present study, we adapted the technique ofERPs to investigate German verb inflection processes.Following a number of unsystematic attempts to apply theERP method to language processing, Kutas and Hillyardw x23 in their seminal paper described an ERP component,

Žthe N400, that was elicited by semantic a semantically.inappropriate terminal word but not by physical violations

Ž .a sentence-final word written in another font . SubsequentŽ w x.research summarized, e.g. in 7,17,20,21,40,51,61,62 has

demonstrated that a semantic violation is not a necessaryprerequisite for the elicitation of the N400 response andthat the N400 amplitude varies as a function of degree ofsemantic association, contextual constraint, position of theword within a sentence, priming to name, etc. However,the violation paradigm has proven to be extremely reliablein evoking N400 responses. It is not surprising, therefore,that investigators have relied on the presentation of viola-tions in other language domains. Most research in themorpho-syntactic domain has been done on violations ofcase, number and tense in different languages, Englishw x w x w x1,6,24,33,39,42 , Dutch 14–16 , Spanish 22 , Germanw x w x11,31,34,35 , and Turkish 32 . Two ERP effects havebeen observed to these kinds of morpho-syntactic viola-tions: a positivity with a latency of about 600 ms variably

w x Ž w x.called P600 39,41 or SPS syntactic positive shift 16and an earlier negativity with a left anterior frontal distri-

Ž w x.bution called LAN left anterior negativity 11,19,22 .While the status of the two effects is still hotly debated, itis safe to say that morpho-syntactic violations reliablyelicit similar brain responses across different languages,see Section 5.2 for a discussion of LAN effects.

The violation paradigm has turned out to be one of the

most promising avenues for ERP research on language. Inour research, we have therefore adopted this paradigm toinvestigate ERP responses to morphological violations.

2. Models of morphological processing

Psycholinguists have suggested three broad approachesto capture the processing of morphologically complex

w xwords. First, proponents of full-listing models 2,27 donot attribute any significant role to the morphologicalstructure of words in the way they are produced or per-

w xceived. Recent connectionist networks 13,26,50,53 aresimilar in spirit. In these models, all words are claimed tobe stored in the mental lexicon irrespective of whether ornot the linguist’s analysis yields a morphologically struc-tured representation, e.g. a stemqaffix combination.

w xSecond, the full parsing model 57–60 claims that onlystems have entries in the mental lexicon and that morpho-logical variants need to be decomposed in processingbefore their stems can be accessed. This model assumesglobal affix-stripping mechanisms for processing purposeswhich do not necessarily correspond to the morphologicalstructure of a complex word.

w xThird, dual-route models 49,25,9,56 claim that ourmindrbrain provides us with two qualitatively differentclusters for treating morphologically complex words, eachwith its own representational basis and processing mecha-

w xnism. In the Pinker and Prince model 49 , these twoclusters are linked to the linguistic distinction betweenregular and irregular inflection. In the so-called irregularcluster, inflected words are represented in terms of lexicalentries, and they are processed by accessing full-formrepresentations stored in memory, whereas in the regularcluster inflected words are represented through affixationrules, and in processing they are decomposed into stemqaffix combinations.

To test these models of morphological processing, wehave investigated two inflectional systems of German,

w xnoun plurals and past participles 4,5,44,65,66 . These twosystems, we think, provide more appropriate grounds forexamining the cognitive status of inflectional rules and theregularrirregular distinction than the English past tense.Notice, for example, that 95% of the verbs in English areregular. Hence, regular past tense forms are much moreŽ . Ž w x.type -frequent than irregular ones see 28 . Moreover,only regular past tense forms contain a segmentable affix.This means that potential differences between forms suchas walk-ed versus came could be effects of frequencydifferences andror effects of the presence or absence of anovert affix, rather than effects of morphological regularityversus irregularity. In German plurals and participles, how-ever, regularity is not confounded with the presence of anovert affix and with type frequency. Regulars and irregu-lars have segmentable endings, and unlike in English, theregular forms do not represent the majority pattern in any

( )M. Penke et al.rCognitiÕe Brain Research 6 1997 37–52 39

Žof the two systems. Regular -t participles getanz-t ‘‘danc-. Ž .ed’’ and irregular -n participles gelad-en ‘‘load-ed’’

have similar type frequencies 1, and the regular -s pluralŽ .Auto-s ‘‘car-s’’ is less frequent than any of the irregularplurals.

The three morphological processing models mentionedabove make different predictions for our ERP experiments.If morphologically complex words are stored in the mentallexicon, as argued by the full-listing model, then we wouldnot expect to find any regularrirregular differences, i.e.,violations of regular and of irregular inflection shouldelicit similar ERP responses. According to the full parsingmodel, all morphologically complex words are decom-posed into stemqaffix if possible. In German, regular aswell as irregular participle and plural forms have seg-mentable endings. Thus, incorrect participle and pluralforms of both regulars and irregulars might be expected toproduce violation effects similar to those observed formorpho-syntactic anomalies in previous ERP studies, i.e.,

Ž .a P600 or a left anterior negativity LAN . In contrast tow xthe other two models, the dual-mechanism approach 49

would predict different ERP responses for regular andirregular inflection. Parsing violation effects should onlyoccur for incorrect regular affixation, i.e., when the regularparticiple ending -t and the plural -s are applied incor-rectly, and not for misapplications of irregular inflectionalpatterns.

w xIn a previous study 66 , we tested these predictions forGerman noun plurals. ERPs were recorded as 18 German-speaking subjects read sentences that contained Germannouns in correct and incorrect plural forms as criticalwords. We found different ERP responses to regular andirregular plurals: incorrect applications of -s pluralizationto nouns that normally take irregular plural forms, e.g.)Bauer-s instead of the correct form Bauer-n ‘‘farmer-s’’,were associated with a negativity that had an onset latencyof about 200 ms and a maximum at left anterior temporalsites. In contrast to that, incorrect regulars, e.g. )Auto-ninstead of the correct form Auto-s ‘‘car-s’’, did not pro-duce a LAN, but rather a low amplitude negativity with a

maximum at central sites. This dissociation, we argued,confirms the regularrirregular distinction of the dual-mechanism model in that violations of the -s plural ruleelicited clearly different brain responses from misapplica-tions of irregular plurals.

To determine whether these findings generalize to otherinflectional systems, we will in the present study reportresults from three experiments on German participle for-mation. We will show that, parallel to the plurals, thedistinction between regular and irregular participles isassociated with different brain responses. Most impor-tantly, violations of the regular -t participle rule elicited aLAN, just like incorrect -s pluralization did. The generalimplications of this finding for the LAN as a marker ofrulerparsing violations will be discussed in Section 5.

3. Methods and materials

3.1. Subjects

Healthy young right-handed native speakers of Germanwere recruited at the Medical School of Hannover, Ger-

Žmany. There were 20 subjects 11 women and 9 men, age.range 21–30 years, mean 24.8 in the sentence experiment,

Ž14 subjects 4 women and 10 men, age range 22–33 years,.mean 25.6 in the story experiment and 14 subjects in the

Žword-list experiment 4 women and 10 men, age range.22–37 years, mean 27.2 . None of the subjects participated

in more than one experiment. The data of several addi-tional subjects had to be discarded because of a highartifact rate leading to the rejection of more than 30% ofthe trials.

3.2. Stimuli

The critical stimuli differed with respect to verb typeŽ .regular, irregular, nonce verb and in the participle end-

Ž Ž ..ings, -t vs. -n. This resulted in six conditions see 1 ; 50items per condition were tested.

Ž . Ž .1 a. durchgetanzt ‘‘dance-d through’’ regular verb, correct -t) Ž .b. durchgetanzen regular verb, incorrect -n

Ž .c. aufgeladen ‘‘load-ed on’’ irregular verb, correct -n) Ž .d. aufgeladet irregular verb, incorrect -t

Ž .e. angeriltet ‘‘rilt-ed at’’ nonce verb, expected -tŽ .f. angerilten nonce verb, unexpected -n

1 w xBybee 3 argued that -t participle suffixation applies to the largest number of verb stems and therefore has a higher type frequency than -n participleŽ .forms. Bybee’s frequency counts, however, are only based on a rather small set of verb types, the 1258 so-called basic verbs ‘‘Grundverben’’ . Analyses

w xof larger sets of verbs showed that -t and -n participles have similar type frequencies 4,28,63 .


In order to minimize differences between participles ofregular and irregular verbs, we did not include any irregu-lar verbs with vowel changes in their participle forms.Instead, we only included irregular verbs, such as aufladen– aufgeladen ‘‘to load on – loaded on’’ in which the

Ž .participle stem s lad- ‘‘load-’’ is identical to the presenttense stem. Hence, the only difference between the irregu-lar and the regular participle forms used in the experimentwere the two endings -t vs. -n. To generate 50 irregularverbs without vowel changes in the participle, we madeuse of prefix-verb formation, a highly productive processin German. This process combines a prefix such as durch‘‘through’’, auf ‘‘on’’, an ‘‘at’’, with a base verb, e.g.tanzen ‘‘to dance’’, and laden ‘‘to load’’ to form verbs

Ž .such as those in 1 . Note that all the 300 critical stimuliconsist of such prefixrbase verb combinations. The itemsused in the experiments were also controlled with respectto frequency and similarity. All existing participles and the

Žcorresponding base verbs have low token frequencies par-w xticiple forms -30 30 ; participle forms of base verbs

w x w x.-500 30 ; base form of the verbs used -300 54 . Tominimize possible association effects, we made sure thatthe participles of nonce verbs used in the experiment didnot rhyme with any existing regular or irregular verb. Twosets of critical stimuli with 300 participles each wereprepared; verbs that had the -t ending in the first set hadthe -n ending in the second set, and vice versa.

The 300 critical words were presented in three differentversions, as elements of simple sentences, embedded in astory and in a word list. For each version separate ERPswere recorded.

3.2.1. Sentence experimentThe critical words were presented as the terminal words

within simple declarative sentences of the following form:

Ž .2 a. Sie haben die ganze Nacht durchgetanztŽ .they have danced the whole night through .

b. Sie haben ihre Mobel schon aufgeladen¨Ž .they have already loaded the furniture .

c. Er hat seine Schroge nicht angeriltet¨Ž Ž .he has not angeriltet snonce verb his Schroge¨Ž ..snonce noun .

Each sentence consisted of six words. To exclude po-tential semantic associations, the direct object in sentences

Ž .with nonce verbs such as 2c was a nonce word too. The300 sentences were randomized and presented one word ata time in yellow letters on a blue background. The wordssubtended 1.18 of visual angle in height and between 1.28

and 6.28 in width at the viewing distance of 140 cm.Throughout the whole experiment a yellow fixation dotwas present in the center of the screen.

The words were shown for 300 ms and the stimulusonset asynchrony between successive words of a sentencewas 700 ms. The interval between two sentences was 3500ms. After 10 critical sentences a ‘‘test sentence’’ wasshown in red letters, which either was an exact repetitionof one of the 10 sentences shown before or was slightlymodified, by exchanging one word. The task assigned tothe subjects was to judge whether the test sentence was orwas not an exact repetition of one of the previous 10sentences and to press one of two buttons accordingly. Theperformance on this task was virtually perfect and will notbe discussed further. Short breaks of 1–3 min were givenafter 40 sentences.

Subjects were tested in a single session while seated inan easy chair in a dimly lit experimental chamber. Thesubject was observed through an infra-red closed circuitvideo-system during the experiment. Instructions weregiven to minimize movements and eye-blinks during theexperiment. The total experiment took an average of 150min to complete including the placement of the electrodes.

3.2.2. Story experimentThe same 300 critical words as in the sentence experi-

ment were used. This time the words were embedded in

Table 1Brief summary of overall ANOVAs

Measure Electrodes Type Correctness T=C T=C=Antpost T=C=Hem T=C=A=H

Early 250–500 ms temporal 111, 222 11, 2, 333 111, 333 111, 222, 3 11, 3parasagittal 111, 222, 3 11, 3 11 1, 2midline 11, 222, 33 1, 33 1, 3 not applicable not applicable

Late 500–750 ms temporal 11 11 11, 33 11, 2 1parasagittal 1 1 11, 222midline 1 not applicable not applicable

Ž . Ž . Ž .Factors: TypesT irregular, regular, pseudo ; CorrectnesssC correct vs. incorrect ; AntpostsA, electrode site factor anterior to posterior direction ;HemsH, hemisphere factor; A=Bs interaction of factors A and B. Other main effects and interactions omitted.

Ž . Ž . Ž .1, 2, 3: P-0.05 for experiment 1 sentence , 2 story and 3 list , respectively.11, 22, 33: P-0.01.111, 222, 333: P-0.005.1111, 2222, 3333: P-0.001.


Tab

le2

Stat

istic

alre

sults

for

the

diff

eren

tw

ord

type

sa

Mea

sure

Exp

erim

ent

Tem

pora

lPa

rasa

gitta

lM

idlin

e

Cor

rect

ness

C=

Ant

post

C=

Hem

C=

A=

HC

orre

ctne

ssC

=A

ntpo

stC

=H

emC

=A

=H

Cor

rect

ness

C=

Ant

post

Fs

,P

-F

s,

P-

Fs

,P

-F

s,

P-

Fs

,P

-F

s,

P-

Fs

,P

-F

s,

P-

Fs

,P

-F

s,

P-

A.I

rreg

ular

Õer

bsE

arly

250

–50

0m

sSe

nten

ce24

.6,

0.00

018.

1,0.

008

15.1

,0.

001

6.9,

0.00

320

.0,

0.00

031.

46,

–4.

1,–

1.44

,–

15.7

,0.

0008

0.13

,–

Stor

y17

.9,

0.00

11.

0,–

17.1

,0.

002

0.02

,–

12.7

,0.

004

0.8,

–3.

9,–

0.4,

–8.

3,0.

015

0.25

,–

Lis

t22

.9,

0.00

040.

69,

–10

.1,

0.00

78.

6,0.

009

24.1

,0.

0003

1.7,

–2.

3,–

1.7,

–13

.9,

0.00

250.

48,

–L

ate

500–

750

ms

Sent

ence

8.9,

0.01

3.78

,0.0

43.

78,

–0.

94,

–3.

91,

–4.

0,0.

015

2.1,

–0.

9,–

1.93

,–

0.49

,–

Stor

y2.

7,–

2.2,

–10

.9,

0.00

60.

25,

–0.

18,

–0.

67,

–9.

0,0.

010.

29,

–0.

13,

–0.

16,

–L

ist

3.9,

–2.

4,–

0.7,

–1.

9,–

2.11

,–

0.8,

–1.

3,–

1.8,

–0.

92,

–0.

98,

–

B.R

egul

arÕer

bsE

arly

250–

500

ms

Sent

ence

0.69

,–

2.0,

–0.

47,

–0.

07,

–0.

32,

–3.

84,0

.02

0.06

,–

0.42

,–

0.18

,–

1.87

,–

Stor

y0.

57,

–1.

03,

–13

.8,

0.00

31.

22,

–0.

05,

–0.

7,–

1.2,

–0.

3,–

0.05

,–

0.8,

–L

ist

0.00

,–

2.25

,–

0.41

,–

0.53

,–

0.16

,–

0.26

,–

0.01

,–

0.18

,–

0.02

,–

0.7,

–L

ate

500–

750

ms

Sent

ence

0.85

,–

0.20

,–

2.3,

–2.

4,–

0.02

,–

0.17

,–

0.43

,–

0.55

,–

0.35

,–

3.8,

–St

ory

0.02

,–

0.6,

–2.

2,–

1.6,

–0.

03,

–0.

85,

–0.

56,

–0.

13,

–0.

06,

–1.

3,–

Lis

t0.

35,

–1.

88,

–0.

7,–

2.0,

–0.

12,

–0.

23,

–0.

13,

–1.

2,–

0.63

,–

0.43

,–

C.P

seud

o-Õer

bsE

arly

250

–50

0m

sSe

nten

ce1.

27,

–0.

20,

–2.

91,

–3.

1,–

1.85

,–

3.71

,0.0

23.

44,

–0.

84,

–0.

35,

–0.

48,

–St

ory

1.4,

–0.

34,

–0.

42,

–1.

9,–

0.04

,–

1.7,

–2.

4,–

0.76

,–

1.0,

–4.

0,0.

04L

ist

2.2,

–5.

5,0.

014.

87,0

.05

2.3,

–1.

0,–

2.1,

–2.

3,–

1.7,

–2.

0,–

2.5,

–L

ate

500–

750

ms

Sent

ence

3.84

,–

1.85

,–

0.66

,–

0.58

,–

0.68

,–

3.5,

0.04

0.26

,–

1.64

,–

1.19

,–

1.91

,–

Stor

y0.

2,–

0.18

,–

0.32

,–

1.7,

–0.

01,

–0.

67,

–0.

94,

–0.

68,

–2.

6,–

2.8,

–L

ist

1.6,

–3.

0,–

0.05

,–

0.5,

–0.

3,–

2.8,

–0.

04,

–1.

6,–

0.47

,–

5.2,

0.04

aŽ

.C

orre

ctne

sss

C;

Ant

post

sA

:el

ectr

ode

site

fact

oran

teri

orto

post

erio

rdi

rect

ion

;H

ems

H:

hem

isph

ere

fact

or;

A=

B:

inte

ract

ion

offa

ctor

san

dB

.O

ther

mai

nef

fect

san

din

tera

ctio

nsom

itted

.Ž

.Ž

.D

egre

esof

free

dom

,se

nten

ceex

peri

men

t:C

orre

ctne

ss,

Cor

rect

ness

=H

emis

pher

e:1,

19;

Cor

rect

ness

=A

ntpo

st,

C=

A=

Hte

mpo

ral,

mid

line

2,38

;pa

rasa

gitta

l4,

76.

Ž.

Ž.

Deg

rees

offr

eedo

m,

stor

yan

dlis

tex

peri

men

t:C

orre

ctne

ss,

Cor

rect

ness

=H

emis

pher

e:1,

14;

Cor

rect

ness

=A

ntpo

st,

C=

A=

Hte

mpo

ral,

mid

line

2,26

;pa

rasa

gitta

l4,

52.


ŽFig. 1. ERPs for the irregular participles. Left side: grand average ERPs for correct and incorrect participles. The layout of the electrodes see scheme in. Ž .right lower corner corresponds to the approximate positions on the scalp. On the right side of the figure the left anterior temporal site F7 is depicted on a

larger scale. In all three experiments incorrect irregular participles are associated with a more negative waveform. This negativity shows slight differencesin distribution across experiments but the left anterior temporal site displays the effect in all cases.


Žthree stories: an animal fable 1312 words total, 104r300. Žcritical words , an episode from the Odyssey 1486 words

.total, 102r300 critical words and the tale of the Cologne

Ž .dwarfs 1349 words total, 94r300 critical words . Unlikethe sentence study the critical words in this experiment didnot always appear clause-finally; they were rather some-

Fig. 2. ERPs for regular participles. Same layout as Fig. 1. For the incorrect participles no consistent difference across experiment is found.


times followed by a dislocated prepositional phrase, agrammatically well formed word-order variation in Ger-man. Two different versions of the stories were prepared

with correct and incorrect participle forms exchanged. Halfof the subjects were exposed to the first version, the other

Žhalf to the second one. The mode of presentation colours

Fig. 3. ERPs for participles of nonce verbs. Same layout as Fig. 1. No consistent differences between participles with the expected and the unexpectedending emerged across experiments.


.of words and background, viewing distance was identicalto the sentence experiment; each word was presented for300 ms, the stimulus onset asynchrony between words was800 ms. Subjects were told that they participated in amemory experiment and after each of the stories theyreceived a questionnaire with 10 simple questions pertain-ing to the contents of the story. Performance on thequestionnaires was flawless and will not be discussedfurther. The subjects were tested in a single session takingabout 180 min including electrode placement.

3.2.3. Word-list experimentThe critical stimuli were the same as in the sentence

and the story experiments. The 300 critical participle forms,Ž300 additional verb forms 100 infinitives, 100 present

.tense forms, 100 past tense forms used as filler items, and100 nouns and 20 nonce nouns were assembled in randomorder to yield a word list. The word list was presented oneword at a time in yellow letters against a blue backgroundŽword duration 300 ms, stimulus onset asynchrony 2000

.ms . The subjects’ task was to press a button held in theright hand whenever they encountered a noun. While inwritten German the first letter of nouns has to be uppercase, all stimuli of the list were presented in lower caseletters to prevent subjects from making the nounrverbdistinction simply on the basis of case. Performance under

Žthese conditions was very good )90% correct responses.to nouns in each case indicating that the subjects actually

read the stimuli in order to make the required distinction.Ž .Subjects were tested in a single short approx. 70 min

session.

3.3. Recording and analysis

EEG was recorded from all 19 standard scalp sites 2 ofthe 10r20 system using tin electrodes mounted in anelectrode cap with reference electrodes placed at the mas-toid processes. Additional electrodes were affixed at theright external canthus and at the right lower orbital ridge tomonitor eye movements for later off-line rejection. Thebiosignals were amplified with a bandpass from 0.01 to 70Hz, digitized at 250 pointsrs and stored on magnetic disk.After artifact rejection by an automated procedure using anindividualized amplitude criterion on the eye channel andthe frontopolar channels, ERPs were averaged for 1024epochs, including a 100 ms prestimulus interval. Thewaveforms were quantified by mean-amplitude measuresin time windows 250–500 ms and 500–750 ms. Thesedata were subjected to repeated measures analyses ofvariance. Since effects were differentially distributed overthe scalp, separate analyses were done for the midline

2 Ten additional electrodes were used in the sentence experiment.However, for the sake of consistency across experiments only the 10r20sites will be discussed in this paper.

Ž .Fig. 4. Comparison of participles of nonce verbs with the expected -tending and correct regular and irregular participles for the Cz site. Thenonce verb participles exhibit a negativity that corresponds to the N400component.

Ž . ŽML; Fz, Cz, Pz , parasagittal PS; Fp1r2, F3r4, C3r4,. Ž .P3r4, O1r2 , and temporal TE; F7r8, T3r4, T5r6

electrodes with the latter two sets split into an electrodeŽ .site anterior to posterior, PS: 5 levels, TE: 3 levels and a

Ž .hemisphere factor left vs. right hemisphere . The analysiswas conducted in two steps: first, an overall ANOVA was

Žcomputed with Word Type regular vs. irregular vs.. Ž .pseudo , Correctness correct vs. incorrect , Site and

Hemisphere as factors. As significant interaction effectsinvolving the Word Type and Correctness factors wereobtained in these analyses, additional ANOVAs were con-ducted separately for each word type to further delineatethe pattern of results. The Greenhouse–Geisser correctionfor inhomogeneities of covariance was applied wheneverapplicable. Reported P values are corrected.

4. Results

The pattern of results was very similar across the threeexperiments; the data from these experiments will there-fore be presented together organized according to thedifferent participle forms, regulars, irregulars and nonce

Fig. 5. Comparison of regular and irregular participles according toŽending for the F7 site: only the incorrect irregulars those with the -t

.ending are associated with the left anterior temporal negativity in allthree experiments.


forms. Table 1 shows a summary of the results of theoverall ANOVAs. For all experiments and for both timewindows, interaction effects of the factors Type and Cor-rectness were found, as well as interactions of these twofactors with the topography factors Site and Hemisphere.These interactions warranted the computation of separate

Ž .ANOVAs for the different word types see Table 2A–Cto elucidate their nature; these will be assessed in thefollowing.

4.1. ERPs to irregular words

The grand-average ERPs to the irregular words areshown in Fig. 1. The ERPs begin with an ensemble of

Žearly peaks and troughs negativity at about 100 ms,.positivity at about 200 ms that differed considerably

across the experiments. The differences in these earlyexogenous components are due to the physical conditionsin which the stimuli were presented, most notably to thedifferent inter-stimulus intervals. These were responsible,for example, for the higher amplitude occipital N1 in theword-list experiment.

More importantly, the ERPs in Fig. 1 show that in allthree experiments, starting at approximately 150–200 ms,irregular verbs with incorrect -t endings were associated

with a more negative waveform. This negativity differedsomewhat in its distribution across the experiments but

Ž .showed a preponderance at the left frontotemporal F7site, which is marked by boxes on the left of Fig. 1. At theright of Fig. 1, the F7 site is plotted on a larger scale.

Ž .Statistically see Table 2A , this negativity gave rise to amain effect of the factor Correctness for all three electrodesets across the three experiments for the mean amplitudemeasure in the 250–500 ms time window. For the tempo-ral electrode set, interactions of Correctness with the fac-

Ž .tors Site anterior to posterior and Hemisphere were ob-served indicating a circumscribed topographical distribu-tion of the negativity. While the effect appeared to extendinto later portions of the waveform, statistical effects forthe 500–750 ms time window were less pronounced.

4.2. ERPs to regular words

Ž .The ERPs to the regular verbs Fig. 2 showed a similargeneral waveshape as the irregular verbs. However, noconsistent differences between correct and incorrect verbswere seen in the three experiments. In the sentence experi-ment, the ERPs to incorrect verbs appeared to be slightlymore positive starting at approximately 200 ms, an effectthat was not present in the story and word-list experiments.

Fig. 6. Illustration of the scalp distribution of the negative effect: depicted are the mean differences between correct and incorrect participles in theŽ .250–500 ms time window aligned for the prestimulus baseline y100 to 0 ms . TEMP, temporal electrodes; PARA, parasagittal electrodes. A remarkable

similarity across experiments emerges with the irregulars being associated with a negativity, which has a left sided preponderance, particularly for thetemporal electrodes. The difference between correct and incorrect regular participles words is close to zero in all cases.


Ž .Statistically see Table 2B , the positive-going waveformin the sentence study gave rise to a Correctness by SiteŽ .anterior to posterior interaction for the parasagittal elec-trode set in the early time window. In the story experi-ment, however, a more negative-going ERP was found atT3 which led to a Correctness by Hemisphere interactionfor the temporal electrode set in the early time window.Hence, there were no consistent ERP effects for incorrectregulars across the three experiments.

4.3. ERPs to nonce words

The ERPs to the nonce verbs are shown in Fig. 3. WithŽregard to the two participle forms expected -t versus

.unexpected -n , no consistent differences were obtainedacross the three experiments. Slightly more positive ERPsto the -n participles in the word-list experiment at the T5

Žand T6 sites gave rise to a Correctness by Site anterior to.posterior interaction for the temporal electrode set in the

Ž . Ž . Ž .Fig. 7. A: difference waves incorrect minus correct for the three verb types regular, irregular, nonce averaged across all three experiments ns48 .Only the difference wave for the irregulars shows a monophasic negativity with a left anterior maximum. B: isovoltage maps using spherical splineinterpolation. Depicted is the mean amplitude in the 250–500 ms time window measured on the incorrect–correct difference waves for the irregulars. Theisovoltage maps show a circumscribed left anterior negativity. Scaling: min, y2.7 mV; max, 1.25 mV. C: current source density maps on the same data as

Ž .in B. These maps provide a reference-free estimate of current flow through the skull. They are characterized by a prominent negative sink at F7 and T3.Scaling: min, y17.75 mVrm2; max, 9.00 mVrm2.


early time window. In the sentence experiment, on theother hand, a slightly more positive waveform for -nparticiples was observed at Fp1 and Fp2 causing a Correct-ness by Site interaction for the parasagittal electrode set.

Recall that the nonce word items used in our experi-ment are pronounceable words for German native speakersand that previous ERP studies found an N400 component

w xespecially for pronounceable nonce words 52,37 . To de-termine whether this also holds for our nonce word items,ERPs to correct irregular and regular participles werecompared to nonce participle forms with the expected -t

Ž .suffix for the Cz electrode see Fig. 4 . For all threeexperiments a negativity was observed for the nonce wordsthat was more prominent in the sentence and story experi-ments than in the word-list study.

4.4. The distribution of the ERPs to regular and irregularwords

In order to assess the distribution and the extent of theeffects associated with incorrect participles, three analyseswere performed. First, we compared the ERP responses to

3 Žregular and irregular participle forms at the F7 site see.Fig. 5 . The waveforms show that only the incorrect

irregular participles are associated with an enhanced nega-tivity, whereas incorrect occurrences of the participle -n onregular verbs produced ERPs similar to those of the correctregular participle forms.

Second, we compared the difference in mean amplitudebetween correct and incorrect participles within the 250–

Ž .500 ms time window for the temporal upper row and theŽ . Ž .parasagittal lower row electrode sets see Fig. 6 . A

remarkable similarity between the three experimentsemerges: the negativity associated with incorrect irregularverbs shows a left preponderance in all three experimentsespecially for the temporal set. Moreover, the negativity ismost prominent for left frontotemporal and frontal sites.Note also, that the difference between correct and incorrectregular participles is close to zero in all cases.

Finally, to obtain a more precise localization on theŽ .scalp, the difference waves incorrect–correct were pooled

from all three experiments. This averaging across experi-ments was done simply for illustrative purposes with theidea that the signal-to-noise ratio would be greatly en-hanced, thus affording us a better idea of the generalizabil-ity of the effects. We felt justified to do this meta-analysisas the critical items were the same in all three experiments,the recording parameters were identical and the subjectswere from the same pool. Fig. 7A presents the differencewaves for regular, irregular and nonce verbs, Fig. 7B theisovoltage maps and Fig. 7C the current source densitymaps for the difference between correct and incorrect

3 ERPs to nonce words had to be excluded from Fig. 5, as these wereindependently shown to produce an N400 component.

irregular participles. The difference waves show that it isonly the incorrect occurrence of -t on irregulars that elicitsa clear ERP response, namely a monophasic negativitywith a left anterior maximum, whereas the incorrect regu-lars and the nonce verbs produce a flat line. The isovoltagemaps indicate that the ERP response we found for incor-rect irregular participles exhibits a clearly circumscribeddistribution. This is further corroborated by the computa-tion of the current source densities, which provides areference-free estimate of the currents passing through theskull, and suggests a left frontal source for the negativity.

5. Discussion

The most salient and consistent result of the presentstudy is that incorrect irregular participles were associatedwith a negative ERP with a left anterior temporal distribu-tion starting at about 200 ms, an effect that has been calledLAN. The LAN for incorrect irregulars was found in eachof the three participle experiments presented in this study.Even more strikingly, this brain response to incorrectirregulars generalizes to an entirely different inflectionalsystem, noun plurals. We found that incorrect irregular

w xnoun plurals of German elicited a LAN 66 just likeincorrect irregular participles. Note that noun plurals differfrom participle forms, both in terms of structural propertiesand in terms of vocabulary distribution. Whereas the regu-lar participle ending -t is added to verb stems, e.g. tanz-‘‘danc-’’, kauf- ‘‘buy-’’, leg- ‘‘put’’, the regular pluralending -s is added to word forms such as Auto ‘‘car’’,Pizza, Bar, etc., i.e., to elements which exist as indepen-dent words in the language. Moreover, the frequencydistribution and the similarity patterns in these two inflec-tional systems are quite different. Irregular verbs do notoutnumber regular ones, independent of the method of

w xfrequency calculation chosen 64 , but 92% of the Germannouns form their plurals according to one of the variousirregular patterns. Despite these linguistic differences, theERP response to incorrect irregulars is very similar fornoun plurals and participles, indicating the stability, repro-ducibility and generalizability of this negativity.

5.1. ERPs and morphological processing models

The current set of ERP results demonstrate clear elec-trophysiological differences between regular and irregularinflection for noun plurals and participles, a finding whichis directly relevant to the competing models of morpholog-ical processing introduced in Section 2. Recall that accord-ing to the full-listing model regular and irregular inflectionare processed in similar ways, i.e., by the storage of itemsand the creation of connections in the mental lexicon.From such a model, one would predict violations of regu-lar and of irregular inflections to elicit similar ERP re-sponses. Our results show that this prediction is not borne


out. The second variant of a single-mechanism approach isthe full parsing model according to which all morphologi-cally complex words are decomposed into stemqaffix ifpossible. As German participle and plural forms havesegmentable endings, we would again expect to find simi-lar brain responses to incorrect forms of both regulars andirregulars. This is not confirmed, however. Hence, ourERP results do not appear to support any of the single-mechanism models.

The dual-mechanism model, on the other hand, wouldbe consistent with different ERP responses for regular andirregular inflections. This model distinguishes between twodifferent clusters for morphologically complex words, theirregular cluster involving the retrieval of lexical entriesstored in memory and the regular cluster involving mor-phological parsing. Consider, from the perspective of thedual-mechanism model, the specific linguistic conditionstested in our experiments. Two kinds of morphological

Ž .anomalies were investigated: i regularizations, -t affixesŽ) . Ž .appearing on irregular verbs geladet ; ii irregulariza-

Ž) .tions, the participle -n on regular verbs getanzen orŽ) .nonce verbs gerilten . The structure of these morpholog-

ical anomalies is quite different. The first case is a parsingviolation, i.e., a misapplication of the -t affixation rule to averb that would normally block the rule, to produce illegalstemqaffix combinations such as ) gelad-et. The othertwo cases are incorrect or unexpected -n participle formsof verbs that would normally undergo the -t affixation rule.According to the model, there is no -n suffixation rule inGerman participle formation; irregular participles are rathersaid to be stored in memory together with the -n ending.Hence, it is only regularizations that involve rule viola-tions. In such cases, we found a characteristic ERP re-sponse in the three participle experiments and the plural

w xstudy 66 : the LAN occurred when the regular participleending -t and the plural -s were applied incorrectly, butnot in cases of irregularizations. Thus, the LAN foundunder these conditions can be interpreted as reflecting

Ž .processes involved in morphological structure building.

5.2. A comparison with left anterior temporal negatiÕitiesfrom other ERP studies

In previous studies, several kinds of morpho-syntacticanomalies have been claimed to elicit a LAN, e.g. viola-tions of syntactic agreement, phrase-structure violations,illegal filler-gap constructions and certain kinds of gram-matically well formed embedded clauses. Some re-searchers have interpreted the LAN as reflecting working

Žw x.memory operations 6,19,22 . From a linguistic perspec-tive, however, the LAN appears to vary as a function ofprocesses involved in morpho-syntactic structure building.

ŽThe LAN occurs when rules of affixation e.g. subject-verb.agreement and phrase-structure building are incorrectly

applied and in cases of complex syntactic dependenciesthat require extra parsing effort, such as filler-gap con-structions and re-analysis of theta role assignment. Our

findings on regular and irregular inflections are compatiblewith this interpretation; a LAN was found for violations ofregular inflectional rules, i.e., for cases that involve mor-phological parsing processes, and it was not associatedwith violations of irregular inflectional patterns, which donot involve affixation rules.

w x w xMunte et al. 33 and Adamson et al. 1 found that¨violations of subject-verb agreement in English, e.g. ) yougoes and ) your spend, were associated with a moreanterior and left preponderant negativity, while in semanticjudgements pairs of nouns such as sky–thief compared torobber–thief produced an N400 component. The negativ-ity in the agreement condition was shown to be distinctfrom the N400. Most probably, it is related to the fact thatagreement violations involve parsing processes and arestructurally more complex than simple noun pairs. Oster-

w xhout and Mobley 38 also found a LAN for violations ofsubject-verb agreement but not for incorrectly inflectedreflexive pronouns, such as himself instead of herself orthemselÕes instead of himself. They take subject-verbagreement to be more syntactic than number or gendererrors on pronouns and argue that the LAN reflects thisdifference.

Left anterior negativities were also found in cases ofphrase-structure violations, such as )Ted’s about films

w x )America 36 , and German sentences such as Der Freundw xwurde im besucht ‘‘The friend was in visited’’ 10,12 .

The latter authors argue that the LAN is associated with arelatively low-level early stage of structure building.

Some ERP studies have discovered a LAN in illegalfiller-gap constructions, but results are less consistentacross studies than for agreement and phrase-structure

w xviolations. Kutas and Kluender 22 found a LAN forungrammatical wh questions such as )What can’t youremember who he adÕised .... This negativity was inter-preted as reflecting working memory operations associatedwith the filler-gap construction. Such constructions wereclaimed to put a strain on the limited capacity of theworking memory system, because the system must storethe filler along with intermediate material. However, re-sults from other ERP studies on the same phenomenon

w xhave produced different results. Neville et al. 36 found a( )LAN for so-called subjacency violations What wasi

)sketch of — admired by the man , but not for otheri

violations of wh movement such as )What did the mani

admire Don’s sketch of — , and McKinnon and Osterhoutiw x29 elicited a P600, but no LAN in such cases.

Summarizing, we suggest a linguistic interpretation ofthe results mentioned above in terms of morpho-syntacticstructure building, with the LAN being elicited by viola-tions of morpho-syntactic rules or in cases of complexsyntactic dependencies. The question, however, to whatextent the LAN reflects processes at the level of theworking memory requires further study of linguistic andnon-linguistic stimuli; we will have to leave this questionopen.


5.3. Which brain structures are inÕolÕed in the processingof inflected words?

If we accept the negativity obtained for incorrect -tparticiples and incorrect -s plurals as a correlate of theprocessing differences between regular and irregular words,the question arises as to which brain structures might beinvolved in its generation. Recall that the difference wavesand the maps shown in Fig. 7 suggest a left frontal sourceof the negativity found for regularizations, i.e., for viola-tions of the -t affixation rule. This might in turn fuelspeculations that it is generated by the same brain areaswhich are traditionally said to cause agrammatism. Sincewe cannot be sure, however, that the activity measured atsome electrode site is generated by the brain tissue directlybelow the electrode, caution must be exerted in advancingsuch an interpretation. Clearly, experiments using moreelectrode sites, a larger number of stimuli and the employ-

w xment of the emerging new localization techniques 55,43andror magnetoencephalographic recordings are needed towarrant further reasoning along these lines.

The question as to whether similar or different brainstructures are responsible for irregular and regular inflec-

w xtion has also been the subject of a recent PET study 18on the English past tense. While lying in the PET camera,

Ž .nine male subjects were performing two different tasks: ireading aloud of visually presented stems of regular, irreg-

Ž .ular and nonce verbs; and ii producing past tense formsof these verbs. Using radioactively labelled water it waspossible to assess rCBF differences as a function of thetasks. For this purpose, subtractions were performed of therCBF differences between the past-tense production andthe stem-form reading tasks for the three kinds of verbs.

w xJaeger et al. 18 found different areas of cortical activationin the regular and the irregular tasks: the left dorsolateral

Ž .prefrontal cortex Brodmann’s area 46 and the left ante-Ž .rior congulate gyrus area 24 showed activation in regular

and nonce word past-tense production, while the mid-tem-Ž .poral gyrus area 21 and the left superior frontal gyrus

Ž .area 10 were active in the irregular past tense condition,the latter also being active in the production of nonce past

Ž .tense forms. Moreover, Broca’s area 44, 45 was active inall three production tasks. Jaeger et al. compared theiractivation patterns with other neuropsychological data and,on the basis of that, they made claims about the localiza-tion of past-tense formation processes in specific brainstructures. Broca’s area, for example, is said to process the

Ž .tense feature qpast , area 46 computes regularly inflectedforms, and area 21 is responsible for the retrieval ofirregular forms. They conclude that the different activationpatterns found for regulars and irregulars provide evidenceagainst single-mechanism models such as the connectionistapproach and support the dual-mechanism view that regu-lars are processed differently from irregulars.

Evaluating these far-reaching conclusions should be thesubject of a further study involving a detailed assessment

of the methods and materials used for their PET scanswhich we will not address here. One comment, however,seems to us to be pertinent with respect to the Jaeger et al.

w xstudy 18 . Whereas most of their claims are about process-ing aspects of inflection, particularly the computation ofregulars versus the retrieval of stored irregulars and thebrain structures involved in these processes, the empiricalevidence Jaeger et al. provide from their PET study doesnot bear on rapid processes, such as past tense formation,which take place within 200–400 ms. This is due to thepoor temporal resolution of PETs. What Jaeger et al.’srCBF patterns show instead are cortical activations whilesubjects produce a list of past tense forms. Leaving thiscriticism aside, Jaeger et al.’s data do nonetheless suggestthat regular and irregular inflections are represented incortically distinct areas, thus providing support for theview that regulars and irregulars are based on differentlinguistic representations. To address issues of morpholog-ical processing, however, ERP experiments such as thosepresented here appear to be more revealing than PETstudies.

6. Conclusion

We conclude that two complementary mechanisms co-exist in how our brain processes morphologically complex

Ž .words: i accessing full-form entries stored in memory;Ž .and ii a computational system that decomposes complex

words into stems and affixes. This distinction is reflectedŽ . Ž .in different ERP responses to i and ii .

Acknowledgements

Thanks are due to Berdieke Wieringa, Marta Kutas,Gary Marcus, Sonke Johannes, Mike Matzke, and Jonathan¨King for comments, Jobst Kilian for excellent technicalassistance and Sam Featherston for correcting our English.T.F.M. was supported by a grant from the Hermann and

Ž .Lilly Schilling Foundation TS 013r177r96 , and H.C. byŽa German Science Foundation grant SFB 282rProject

.B7 .

References

w x1 A.M. Adamson, T.F. Munte, G.R. Mangun, Differences in timing¨and distribution of semantic and syntactic incongruency effectsdemonstrated by multichannel ERP recording, in: T.F. Munte, H.-J.¨

Ž .Heinze, G.R. Mangun, H. Scheich Eds. , Mapping Cognition inTime and Space, IOS Press, Amsterdam, 1997, in press.

w x Ž .2 B.L. Butterworth, Lexical representation, in: B.L. Butterworth Ed. ,Language Production, Vol. 2, Academic Press, London, 1983, pp.257–294.

w x3 J.L. Bybee, Regular morphology and the lexicon, Lang. Cogn. Proc.Ž .10 1995 425–455.


w x4 H. Clahsen, The representation of German participles in the Germanmental lexicon: evidence for the dual-mechanism model, in: Year-book of Morphology, Kluwer, Dordrecht, The Netherlands, 1996,pp. 1–31.

w x5 H. Clahsen, S. Eisenbeiss, I. Sonnenstuhl, Morphological structureŽ .and the processing of inflected words, Theor. Ling., 22 1997 in

press.w x6 S. Coulson, J.W. King, S. Coulson, Expect the unexpected: event-re-

lated brain response to morphosyntactic violations, Lang. Cogn.Proc., submitted.

w x7 I. Fischler, G.E. Raney, Language by eye: behavioral and psy-Ž .chophysiological approaches to reading, in: M.G.H. Coles Ed. ,

Handbook of Cognitive Psychophysiology: Central and AutonomicNervous System Approaches, Wiley, Chichester, 1991, pp. 511–574.

w x8 R.S.J. Frackowiak, Functional mapping of verbal memory and lan-Ž .guage, Trends Neurosci. 17 1994 109–115.

w x9 U. Frauenfelder, R. Schreuder, Constraining psycholinguistic modelsof morphological processing and representation: the role of produc-tivity, in: Yearbook of Morphology, 1991, pp. 165–183.

w x10 A.D. Friederici, A. Mecklinger, Syntactic parsing as revealed bybrain responses. First-pass and second-pass parsing processes, J.

Ž .Psycholing. Res., 25 1996 157–176.w x11 A.D. Friederici, A. Hahne, A. Mecklinger, Temporal structure of

syntactic parsing: early and late event-related brain potential effects,Ž .J. Exp. Psychol.: Mem. Cogn. 22 1996 1219–1248.

w x12 A.D. Friederici, E. Pfeifer, A. Hahne, Event-related brain potentialsduring natural speech processing: effects of semantic, morphologi-

Ž .cal, and syntactic violations, Cogn. Brain Res. 1 1993 183–192.w x13 M. Gassar, Learning words in time: towards a modular connectionist

account of the acquisition of receptive morphology, Unpublishedmanuscript, Indiana University, Bloomington, IN, 1993.

w x14 T.C. Gunter, S.H. Vos, G. Mulder, Syntactic violations and ERPs:P600 or P3b?, Poster presented at the CUNY Sentence ProcessingConference, Tucson, AZ, 1995.

w x15 P. Hagoort, C. Brown, Brain responses to lexical ambiguity resolu-Ž .tion and parsing, in: C. Clifton, L. Frazier, K. Rayner Eds. ,

Perspectives on Sentence Processing, Erlbaum, Hillsdale, NJ, 1994,pp. 45–80.

w x16 P. Hagoort, J. Groothusen, C. Brown, The syntactic positive shiftŽ . w xSPS as an ERP 15 measure of syntactic processing, Lang. Cogn.

Ž .Proc. 9 1993 439–483.w x17 S.A. Hillyard, T.W. Picton, Electrophysiology of cognition, in: F.

Ž .Plum Ed. , Handbook of Physiology. Section 1: Neurophysiology,American Physiological Society, New York, 1987, pp. 187–211.

w x18 J.J. Jaeger, A.H. Lockwood, D.L. Kemmerer, R.D. Van Valin, B.W.Murphy, H.G. Khalak, A positron emission tomographic study ofregular and irregular verb morphology in English, Language 72Ž .1996 451–497.

w x19 J.W. King, M. Kutas, A brain potential whose latency indexes theŽ .length and frequency of words, CRL Newslett. 10 1995 1–9.

w x20 M. Kutas, C. Van Petten, Event-related brain potential studies ofŽ .language, in: P.K. Ackles, J.R. Jennings, M.G.H. Coles Eds. ,

Advances in Psychophysiology, Vol. 3, JAI Press, Greenwich, CT,1988, pp. 129–187.

w x21 M. Kutas, C. Van Petten, Psycholinguistics electrified, in: M.A.Ž .Gernsbacher Ed. , Handbook of Psycholinguistics, Academic Press,

San Diego, CA, 1994, pp. 83–143.w x22 M. Kutas, R. Kluender, What is who violating: a reconsideration of

linguistic violations in light of event-related brain potentials, in:Ž .H.-J., Heinze, T.F. Munte, G.R. Mangun Eds. , Cognitive Electro-¨

physiology, Birkhauser, Boston, MA, 1994, pp. 181–210.¨w x23 M. Kutas, S.A. Hillyard, Reading senseless sentences: brain poten-

Ž .tials reflect semantic incongruity, Science 207 1980 203–205.w x24 M. Kutas, S.A. Hillyard, Event-related brain potentials to grammati-

Ž .cal errors and semantic anomalies, Mem. Cogn. 11 1983 539–550.w x25 A. Laudanna, C. Burani, Distributional properties of derivational

Ž .affixes: implications for processing, in: L.B. Feldman Ed. , Mor-

phological Aspects of Language Processing, Erlbaum Press, Hills-dale, NJ, 1995, pp. 345–364.

w x26 B. MacWhinney, J. Leinbach, Implementations are not conceptual-Ž .izations: revising the verb learning model, Cognition 40 1991

121–157.w x27 L. Manelis, D. Tharp, The processing of affixed words, Mem. Cogn.

Ž .5 1977 690–695.w x28 G.F. Marcus, U. Brinkmann, H. Clahsen, R. Wiese, S. Pinker,

German inflection: the exception that proves the rule, Cogn. Psy-Ž .chol. 29 1995 189–256.

w x29 R. McKinnon, L. Osterhout, Constraints on movement phenomenain sentence processing: evidence from event related brain potentials,Unpublished manuscript, University of Washington, Seattle, WA,1994.

w x30 H. Meier, Deutsche Sprachstatistik, Buske, Hamburg, 1964.w x31 T.F. Munte, H.-J. Heinze, ERP negativities during syntactic process-¨

ing of written words, in: H.-J., Heinze, T.F. Munte, G.R. Mangun¨Ž .Eds. , Cognitive Electrophysiology, Birkhauser, Boston, MA, 1994,¨pp. 211–238.

w x32 T.F. Munte, E. Anvari, M. Matzke, S. Johannes, Brain potential¨correlates of umber errors in the Turkish language, Neurosci. Lett.

Ž .199 1995 57–60.w x33 T.F. Munte, H.-J. Heinze, G.R. Mangun, Dissociation of brain¨

activity related to syntactic and semantic aspects of language, J.Ž .Cogn. Neurosci. 5 1993 335–344.

w x34 T.F. Munte, H.-J. Heinze, H. Prevedel, Erreigniskorrelierte Poten-¨tiale reflektieren semantische und syntaktische Fehler bei der

Ž .Sprachverarbeitung, Z. EEG-EMG 21 1990 75–81.w x35 T.F. Munte, M. Matzke, S. Johannes, Brain activity associated with¨

number errors in words and pseudo-words, J. Cogn. Neurosci.,Ž .1997 in press.

w x36 H.J. Neville, J.L. Nicol, A. Barss, K.I. Forster, M.F. Garrett, Syntac-tically based sentence processing classes: evidence from event-re-

Ž .lated brain potentials, J. Cogn. Neurosci. 3 1991 151–165.w x37 A.C. Nobre, G. McCarthy, Language related ERPs – scalp distribu-

tions and modulation by word type and semantic priming, J. Cogn.Ž .Neurosci. 6 1994 233–255.

w x38 L. Osterhout, L. Mobley, Event-related brain potentials elicite byŽ .failure to agree, J. Mem. Lang., 34 1995 354–375.

w x39 L. Osterhout, P.J. Holcomb, Event-related brain potentials elicitedŽ .by syntactic anomaly, J. Mem. Lang. 31 1992 785–806.

w x40 L. Osterhout, P.J. Holcomb, Event-related potentials and languageŽ .comprehension, in: M.D. Rugg, M.G.H. Coles Eds. , Electrophysi-

ology of Mind: Event-Related Brain Potentials and Cognition, Ox-ford University Press, Oxford, 1995, pp. 175–215.

w x41 L. Osterhout, P.J. Holcomb, Event-related potentials and syntacticanomaly: evidence of anomaly detection during the perception of

Ž .continuous speech, Lang. Cogn. Proc. 8 1993 413–437.w x42 L. Osterhout, R. McKinnon, M. Bersick, V. Corey, On the

language-specificity of the brain response to syntactic anomalies: isthe syntactic positive shift a member of the P300 family?, J. Cogn.

Ž .Neurosci., 1997 in press.w x43 R.D. Pascual-Marqui, C.M. Michel, D. Lehmann, Low resolution

electromagnetic tomography: a new method for localizing electricalŽ .activity in the brain, Int. J. Psychophysiol. 18 1994 49–65.

w x44 M. Penke, A. Trommer, H. Weyerts, T.F. Munte, H. Clahsen,¨Electrophysiological correlates of German participle formation, in:

Ž .T.F. Munte, H.-J. Heinze, G.R. Mangun, H. Scheich Eds. , Map-¨ping Cognition in Time and Space, IOS Press, Amsterdam, 1997, inpress.

w x45 S.E. Petersen, P. Fox, M. Posner, M. Mintun, M. Raichle, Positronemission tomographic studies of the processing of single words, J.

Ž .Cogn. Neurosci. 1 1989 153–170.w x46 S.E. Petersen, P. Fox, M. Posner, M. Mintun, M. Raichle, Positron

emission tomographic studies of he cortical anatomy of single-wordŽ .processing, Nature 331 1988 585–589.

w x48 S.E. Petersen, J.A. Fiez, The processing of single words studied with


Ž .positron emission tomography, Annu. Rev. Neurosci. 16 1993509–530.

w x49 S. Pinker, A. Prince, Regular and irregular morphology and thepsychological status of rules of grammar, in: L.A. Sutton, C. John-

Ž .son, R. Shields Eds. , Proc. 17th Annual Meeting of the BerkeleyLinguistics Society, Berkeley Linguistics Society, Berkeley, CA,1992.

w x50 K. Plunkett, V. Marchman, From rote learning to system building,Ž .Cognition 38 1993 43–102.

w x51 F. Rosler, P. Putz, A.D. Friederici, A. Hahne, Event-related brain¨ ¨potentials while encountering semantic and syntactic constraint vio-

Ž .lations, J. Cogn. Neurosci. 5 1993 345–362.w x52 M.D. Rugg, Dissociation of semantic priming, word and non-word

repetition by event-related potentials, Quart. J. Exp. Psychol. 39AŽ .1987 123–148.

w x53 D.E. Rumelhart, J.L. McClelland, On learning the past tenses ofEnglish verbs, in: J.L. McClelland, D.E. Rumelhart, PDP ResearchGroup, Parallel Distributed Processing: Explorations in the Mi-crostructure of Cognition. Vol. 2. Psychological and BiologicalModels, MIT Press, Cambridge, MA, 1986, pp. 216–271.

w x54 A. Ruoff, Haufigkeitsworterbuch gesprochener Sprache, Niemeyer,¨ ¨Tubingen, 1981.¨

w x55 M. Scherg, Functional imaging and localization of electromagneticŽ .brain activity, Brain Topogr. 5 1992 103–111.

w x56 R. Schreuder, H. Baayen, Modeling morphological processing, in:Ž .L.B. Feldman Ed. , Morphological Aspects of Language Process-

ing, Erlbaum Press, Hillsdale, NJ, 1995, pp. 131–154.w x57 M. Taft, K. Forster, Lexical storage and retrieval of prefixed words,

Ž .J. Verb. Learn. Verb. Behav. 14 1975 638–647.

w x58 M. Taft, A morphological decomposition model of lexical represen-Ž .tation, Linguistics 26 1988 657–667.

w x59 M. Taft, G. Hambly, S. Kinoshita, Visual and auditory recognitionŽ .of prefixed words, Quart. J. Exp. Psychol. 38 1986 357–366.

w x60 M. Taft, Recognition of affixed words and the word-frequencyŽ .effect, Mem. Cogn. 7 1979 263–272.

w x61 M. Ullman, S. Corkin, S. Pinker, M. Coppola, J.H. Growdon, J.Locascio, Which neural structures subserve language? Evidencefrom inflectional morphology, Paper presented at the 1st AnnualMeeting of the Cognitive Neuroscience Society, San Francisco, CA,March 1994.

w x62 C. Van Petten, M. Kutas, Influences of semantic and syntacticŽ .content on open and closed class words, Mem. Cogn. 19 1991

95–112.w x63 C. Van Petten, M. Kutas, Interaction between sentence context and

word frequency in event-related brain potentials, Mem. Cogn. 18Ž .1990 380–393.

w x64 H. Weyerts, H. Clahsen, Netzwerke und symbolische Regeln imSpracherwerb: experimentelle Ergebnisse zur Entwicklung der Flex-

Ž .ionsmorphologie, Ling. Ber. 154 1994 430–460.w x65 H. Weyerts, T.F. Munte, H.G.O.M. Smid, H.J. Heinze, Mental¨

representations of morphologically complex words: an event-relatedŽ .potential study with adult humans, Neurosci. Lett. 206 1996

125–128.w x66 H. Weyerts, M. Penke, U. Dohrn, H. Clahsen, T.F. Munte, Brain¨

potentials indicate differences between regular and irregular GermanŽ .noun plurals, NeuroReport, 1997 in press.

Research report How the brain processes complex words: an...

Documents

Transcript of Research report How the brain processes complex words: an...