Pre-attentive auditory processing of lexicality q Thomas Jacobsen, a, * Janos Horvath, b Erich Schr€ oger, a Sonja Lattner, c Andreas Widmann, a and Istvan Winkler b a Institut f € ur Allgemeine Psychologie, Universit € at Leipzig, Seeburgstraße 14-20, Leipzig 04103, Germany b Institute of Psychology, Hungarian Academy of Sciences, Budapest, Hungary c Max Planck Institute of Cognitive Neuroscience, Leipzig, Germany Accepted 22 May 2003 Abstract The effects of lexicality on auditory change detection based on auditory sensory memory representations were investigated by presenting oddball sequences of repeatedly presented stimuli, while participants ignored the auditory stimuli. In a cross-linguistic study of Hungarian and German participants, stimulus sequences were composed of words that were language-familiar, lexical, meaningful in Hungarian but language-unfamiliar, not lexical, meaningless in German, and words with the opposite characteristics. The roles of frequently presented stimuli (Standards) and infrequently presented one (Deviants) were fully crossed. Language-fa- miliar and language-unfamiliar Deviants elicited the Mismatch Negativity component of the event-related brain potential. We found differences in processes of change detection depending on whether the Standard was language-familiar, or not. Whereas, the lexi- cality of the Deviant had no effect on the processes of change detection. Also, language-familiar Standards processed differently than language-unfamiliar ones. We suggest that pre-attentive (default) tuning to meaningful words sets up language-specific preparatory processes that affect change detection in speech sequences. Ó 2003 Elsevier Science (USA). All rights reserved. Keywords: Speech comprehension; Lexical processing; Auditory sensory memory; Mismatch Negativity (MMN); Event-related potentials (ERP) 1. Introduction When engaged in a conversation, listeners tune in to the relevant stream of speech and filter out irrelevant speech input that may be present in the same environ- ment. Nonetheless, attention might be involuntarily di- verted to meaningful items coming from an ignored stream, like in the well-known own-name effect (e.g., Moray, 1959). This brings up the question of to what extent speech is processed in the ignored streams. The present study is concerned with an issue related to this question. Some processing of unattended speech sounds occurs even when one performs an unrelated task. For exam- ple, Service, Winkler, Maury, and N€ a€ at€ anen (submitted) have shown that the phonotactical structure of ignored spoken pseudowords, phonologically legal non-words of a given language, is processed even when the speech sounds are irrelevant for the participantÕs task. Using the event-related brain potential (ERP) technique allows one to assess speech processing with millisecond accu- racy and without the interference of task-related pro- cesses and participant strategies. ERPs are thus frequently used in related studies, in particular, the Mismatch Negativity (MMN) ERP component, which is elicited whether or not participants attend or ignore the sounds. The MMN and its magnetic counterpart, the MMNm, reflect the detection of a change of the cur- rent auditory event from the auditory stimulus repre- sentations extrapolated from the regularities, which have been extracted from the preceding auditory Brain and Language 88 (2004) 54–67 www.elsevier.com/locate/b&l q This study was supported by the DFG, the German-Hungarian Scientist Exchange Program (DAAD-M € OB Grant 53/2001), and the Hungarian National Research Fund (OTKA T034112). The authors thank Terez Balazs, Kinga Gyimesi, and Martina Nemetz for technical assistance. * Corresponding author. Fax: +49-0341-97-35-969. E-mail address: [email protected](T. Jacobsen). 0093-934X/$ - see front matter Ó 2003 Elsevier Science (USA). All rights reserved. doi:10.1016/S0093-934X(03)00156-1
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Brain and Language 88 (2004) 54–67
www.elsevier.com/locate/b&l
Pre-attentive auditory processing of lexicalityq
Thomas Jacobsen,a,* J�aanos Horv�aath,b Erich Schr€ooger,a Sonja Lattner,c Andreas Widmann,a
and Istv�aan Winklerb
a Institut f€uur Allgemeine Psychologie, Universit€aat Leipzig, Seeburgstraße 14-20, Leipzig 04103, Germanyb Institute of Psychology, Hungarian Academy of Sciences, Budapest, Hungary
c Max Planck Institute of Cognitive Neuroscience, Leipzig, Germany
Accepted 22 May 2003
Abstract
The effects of lexicality on auditory change detection based on auditory sensory memory representations were investigated by
presenting oddball sequences of repeatedly presented stimuli, while participants ignored the auditory stimuli. In a cross-linguistic
study of Hungarian and German participants, stimulus sequences were composed of words that were language-familiar, lexical,
meaningful in Hungarian but language-unfamiliar, not lexical, meaningless in German, and words with the opposite characteristics.
The roles of frequently presented stimuli (Standards) and infrequently presented one (Deviants) were fully crossed. Language-fa-
miliar and language-unfamiliar Deviants elicited the Mismatch Negativity component of the event-related brain potential. We found
differences in processes of change detection depending on whether the Standard was language-familiar, or not. Whereas, the lexi-
cality of the Deviant had no effect on the processes of change detection. Also, language-familiar Standards processed differently than
language-unfamiliar ones. We suggest that pre-attentive (default) tuning to meaningful words sets up language-specific preparatory
processes that affect change detection in speech sequences.
� 2003 Elsevier Science (USA). All rights reserved.
56 T. Jacobsen et al. / Brain and Language 88 (2004) 54–67
a separate block, a pseudoword deviant among apseudoword standard (all stimuli had a CVCV struc-
ture). The authors obtained a negative ERP effect for
the word deviant, which peaked between 400 and
450ms after the words onset, which they termed the
‘‘late MMN.’’ The pseudoword deviant elicited a
weaker late-MMN than the word deviant. The authors
concluded that pre-attentive ‘‘auditory processing, . . .,is highly associated with the cognitive meaning of thestimuli’’ (p. 332). However, the late MMN has only
been found in children to now. Therefore, it may
represent some process that does not characterize
adults (the ERP wave-forms go through marked
changes during development; see, e.g., Cheour, Lep-
paenen, & Kraus, 2000). Moreover, Korpilahti et al.
(2001) did not separate the lexical status variable from
acoustic/phonetic factors either.Based on their EEG and MEG results, Pulverm€uuller
et al. (2001) suggested that task-irrelevant words un-
dergo lexical analysis. In their EEG experiments, a word
and a pseudoword deviant were presented amongst a
pseudoword standard. In the MEG experiment, isolated
syllables were presented in random succession at a 450-
ms stimulus onset asynchrony (SOA). On 16% of the
trials, a succession of two of these syllables resulted ei-ther in a word or a pseudoword deviant. Responses to
the codas from these word and pseudoword deviants
were compared with those elicited by standard isolated
syllables in order to assess the parameters of the MMN
components. In all critical comparisons, larger MMNs
were elicited by word deviants than by pseudoword
deviants. The authors interpreted their results as re-
flecting the ‘‘presence of memory traces for individualspoken words in the human brain.’’ It should be noted
that Pulverm€uuller et al.�s results seem to contradict the
results of Diesch et al. (1998).
Finally, Wunderlich and Cone-Wesson (2001) re-
ported that CVC word deviants and CV syllable devi-
ants were less likely to elicit MMN and that the
amplitude of the MMN response is smaller for these
types of stimuli than when the oddball protocol iscomposed of tones. These results contradict the re-
mainder of the literature on MMN elicited by speech
sounds (in addition to the papers already mentioned,
see, e.g., Cs�eepe, Osman-S�aagi, Moln�aar, & Gosy, 2001;
T. Jacobsen et al. / Brain and Language 88 (2004) 54–67 57
other, currently irrelevant sources are also active in theenvironment.1
In addition to testing the elicitation and parameters
of the MMN components, differences in processing
language-familiar and language-unfamiliar Standards
were be investigated. The present experiment can detect
lexical analysis processes for ignored speech sounds (1) if
these processes affect the ERP responses elicited by the
repetitive speech stimuli or (2) if these processes affectthe regularity representations (which include the repre-
sentation of the speech sounds) extracted from the re-
peating speech sounds, or (3) if these processes directly
affect the processing of the infrequently presented speech
stimuli, the Deviants. The second and third effects would
appear as changes in the MMN component, which re-
sults from detecting some mismatch between the deviant
sound and the representation of the regular aspects ofthe speech sequence. Winkler et al. (1999b) suggested
that the MMN elicited by deviant speech-sounds may
have two separate sub-components: one elicited by
acoustic deviance and another elicited by deviance in
some categorical form of information, e.g., classes of
word stimuli. The present study focuses on the com-
parison of MMN effects that can be attributed to the
experimental variation of lexicality, i.e., whether thegiven spoken word items were meaningful (words for
the given participant group) or not (pseudowords). Thus
we expect to obtain MMN components to be elicited in
all four conditions of the present study due to acoustic
differences between the Standards and Deviants. The
answer to our questions will be found (a) by comparing
the responses elicited by the same speech items when
they serve as words and when they serve as pseudowords(i.e., across the two language groups) and (b) by ex-
amining the modulation of the MMN responses with
respect to the lexical familiarity of the Standards and
Deviants.
We shall consider four hypotheses, which have been
put forward by the previous studies. The lexical trace
hypothesis, as advocated by Pulverm€uuller et al. (2001)
predicts that word deviants should elicit a larger MMNthan pseudoword deviants irrespective of the lexical fa-
miliarity of the standard stimuli. The supplementary
processing hypothesis, as suggested by Diesch et al.
(1998), assumes that attempts at lexical access are re-
peatedly made by the system even for highly repetitive
sound items. It predicts larger MMN to be elicited by
deviants presented among meaningless than by mean-
ingful word standards, irrespective of the deviants� lex-ical status. A third hypothesis can be derived from the
1 The term ‘‘default’’ has been taken from the computer literature,
in which it denotes processes performing a given function even when
the operator does not interact with them. The details of the process
can, however, be modified to a certain extent by explicit instructions
from the user.
results showing that familiar contexts enhance the pro-cesses of change detection (see Section 1). Applying this
notion to the present experiment, one can suggest that a
sound environment containing mostly meaningful words
sets a more familiar context than the repetition of
pseudowords. The familiar context hypothesis then sug-
gests MMNs of higher amplitude to be elicited by de-
viants appearing in the more familiar than in the less
familiar context, irrespective of the lexical status of thedeviant item. Finally, the lexical context hypothesis is a
version of the familiar context hypothesis, which focuses
on linguistic processes. The dominance of meaningful
words in a speech sequence creates a language context in
which potentially relevant speech events are likely to
occur. Pseudoword standards, on the other hand, create
an unknown-language, or perhaps even no-language
context. The lexical context hypothesis thus predicts thatthe default processing of meaningful and pseudoword
standards will differ and, further that the results of the
lexical analysis of the standard words will result in an
additional deviance-detection component to be elicited
within the MMN response (analogously to the results of
Winkler et al. (1999b) for phonetic categories). The
general familiar context and the lexical context hy-
potheses are closely related, though the specific predic-tion of the latter hypothesis concerning the processing of
the standards can be tested by comparing the ERP re-
sponses elicited by the standard speech items. Also the
three hypotheses are not mutually exclusive either and,
therefore, interactions between the postulated processes
may occur. These will be tested in the present study.
2. Method
2.1. Participants
Ten native speakers of Hungarian (6 male, mean age
of 21.5 years, range 18–31, normal auditory status) and
10 native speakers of German (7 male, mean age of 27.5
years, range 23–33, normal auditory status) participatedin the study for monetary compensation (Hungarians)
or partial fulfillment of grant requirements (Germans).
None of the participants had knowledge of the respec-
tive foreign language of this study or had foreign lan-
guage experience, in general prior to the age of nine
years. All participants acquired at least one foreign
language after this age (mostly English).
2.2. Stimuli
The stimuli consisted of four consonant–vowel–con-
sonant (CVC) words. The Hungarian minimal pair s�aap([Sa:p], engl. transl. ‘‘illegal profit’’) and s�aas ([Sa:S], engl.transl. ‘‘sedge’’), and the German minimal pair Scham
([Sa:m], engl. transl. ‘‘shame’’) and Schaf ([Sa:f], engl.
T. Jacobsen et al. / Brain and Language 88 (2004) 54–67 59
2.5. Electrophysiological recordings
The electroencephalogram (EEG) was continuously
recorded from 11 scalp locations (F3, Fz, F4, C3, Cz, C4,
P3, Pz, and P4) according to the extended 10–20 system
(American Electroencephalographic Society, 1991) and
the left and right mastoids (Lm and Rm, respectively).
The reference electrode was attached to the tip of the
nose. Electroocular activity (EOG) was recorded withbipolar montages, the vertical EOG between a supra-
and an infraorbitally placed electrode and the horizontal
EOG between the outer canthi of the two eyes. Electrode
impedances were kept below 10 kX. On-line filtering was
carried out using a 40-Hz low-pass and 50-Hz notch filter
(DC recording). The signal was digitized with a 16-bit
resolution at a sampling rate of 250Hz.
2.6. Data analysis
All continuous EEG records were off-line filtered
with a band-pass of 1.5–20Hz. Epochs of 900-ms du-
ration, time-locked to the onset of the stimuli, in-
cluding a 100-ms pre-stimulus-onset interval were
extracted and averaged (point-by-point, time-wise)
separately for oddball and control sequences andcondition within the sequence (Standard, Deviant, and
Control vs. language-familiar and language-unfamil-
iar). The ERP responses to the first five stimuli in each
block as well as epochs showing a signal change ex-
ceeding 100 lV on any recording channel were ex-
cluded from further analysis. For visualizing the
results, grand-averages for each ERP response were
computed from the individual average ERPs. Grand-averages were derived according to the language-fa-
miliar versus language-unfamiliar classification of the
stimuli in the conditions, thus collapsing the two par-
ticipant groups (Hungarian and German). This means
that, for example, the grand-average response to the
language-unfamiliar Standard includes the responses
elicited by the Hungarian word ‘‘s�aas’’ in the German
participants and the responses elicited by the Germanword ‘‘Schaf’’ in the Hungarian participants (see Table
2). As a result of this procedure, the two participant
groups as well as the Hungarian and German words
contributed equally to each grand-average response,
thus eliminating the effects of stimulus and group
differences from the grand-averages.
The MMN responses were assessed from Deviant-
minus-Control difference responses. The Control for agiven Deviant was the identical stimulus presented in the
respective Control block. The ERP responses to Stan-
dards were computed while omitting the responses to
those Standards that followed a Deviant by one or two
positions in the stimulus sequence, because these re-
sponses may contain some MMN (see, e.g., Winkler,
Karmos, & N€aa€aat€aanen, 1996).
ERP effects were quantified using 32ms time windowscentered on the peak of the respective wave in the grand-
average ERP or ERP difference (the latter being used for
the MMN component). In order to quantify the full
MMN amplitude, the frontal (Fz) Deviant-minus-Con-
trol difference response was rereferenced to the linked
mastoid leads. This calculation sums the frontally
measured response with its polarity-reversed counter-
part appearing at the mastoid leads (see, e.g., N€aa€aat€aanen,1992; Schr€ooger, 1998).
The elicitation of the MMN component was assessed
for each of the four deviance conditions (language-
familiar Standard vs. language-familiar Deviant,
language-familiar Standard vs. language-unfamiliar
Deviant, language-unfamiliar Standard vs. language-
familiar Deviant, and language-unfamiliar Standard vs.
language-unfamiliar Deviant) by comparing the deviantand corresponding control responses using dependent
t-tests (a level was set to p < :01). Further statistical
analyses were conducted using mixed repeated-measures
between-subject analyses of variance (ANOVA) with the
factors lexicality of the Deviant (within-subject; lan-
guage-familiar vs. language-unfamiliar Deviant), lexi-
cality of the Standard (within-subject; language-familiar
vs. language-unfamiliar Standard), and language group(between-subject; Hungarian vs. German). Analyses of
EEG topography additionally included as within-subject
factors the anterior–posterior (F-, C-, vs. P-lines) and
left–right (3-, z-, vs. 4-lines) directions on the scalp,
based on the responses obtained from the corresponding
3� 3 electrode grid between F3 and P4 (see Fig. 1 for a
rough illustration of the topography of the electrode
locations).When applicable, p values reflecting Greenhouse–
Geisser (G–G) corrected degrees of freedom and G–G �values are reported. Normalized data, used for the as-
sessment of topographical effects, were derived using
vector-length scaling, as was proposed by McCarthy
and Wood (1985).
3. Results
An average of 12.7% (STD 7.2%) of the trials per
participant were rejected from ERP analysis (range 3.7–
33.3%). The ERPs to all four Deviant conditions
showed identical, orderly early ERP components (see
Fig. 1 and the left-hand side of Fig. 2).
3.1. ERPs to standard stimuli
A comparison of the language-unfamiliar Standard
and language-familiar Standard ERP responses showed
two differences between the electric brain activity (see
the ERP responses and their difference wave in Fig. 1).
Fig. 1. Grand-averaged (groups collapsed, see Section 2) ERPs from all recording locations for the frequently presented stimuli of the oddball
sequences, Standards: language-unfamiliar, pseudoword, Standards plotted with dashed, language-familiar, word, Standards with thin continuous,
and the difference waveforms of language-unfamiliar minus language-familiar Standards with thick continuous line. The two differences, termed First
and Second time window in the text, are marked by dark and light gray shading, respectively. The time at which the difference between the two word
items (‘‘Schaf’’ and ‘‘s�aas’’) commences is marked by dotted vertical lines.
60 T. Jacobsen et al. / Brain and Language 88 (2004) 54–67
The first difference was a phasic fronto-centrally nega-
tive going deflection (when calculated by subtracting the
language-familiar Standard responses from the lan-
guage-unfamiliar Standard responses) appearing be-tween 550 and 630ms from the stimulus onset (peak at
584ms). The second difference was a phasic broadly
distributed positive-going deflection appearing between
650 and 750ms from the stimulus onset (peak at
672ms). Both differences showed polarity reversal at the
mastoid leads.
3.1.1. First time window
An ANOVA test of the factors Standard (language-
familiar vs. language-unfamiliar), language group
(Hungarian vs. German), and left–right (the 3-, z-, vs. 4-
lines of electrodes) was done for the F-line of electrodes.
(For the results of the full ANOVA and the follow-up
analysis of the interactions, see Appendix A detailing the
full statistical analysis.) There was a significant effect of
Standard, F ð1; 18Þ ¼ 6:6, MSE ¼ :25, p < :05. No othereffect was significant, most F s < 1. The effect of Stan-
dards was also assessed at the mastoids using an AN-
OVA with the factors Standard (language-familiar vs.
language-unfamiliar), language group (Hungarian vs.
German), and left–right (left vs. right mastoid). A sig-
nificant main effect of the Standard was obtained,
F ð1; 18Þ ¼ 10:4, MSE ¼ :09, p < :01. One other effect
was also significant (see Appendix A). In sum, in this
earlier window, there was an effect of language-familiarversus language-unfamiliar Standards at frontal elec-
trodes and, with a polarity reversal at the mastoids.
3.1.2. Second time window
The ANOVA test of the factors Standard (language-
familiar vs. language-unfamiliar), language group
(Hungarian vs. German), left–right (3-, z-, vs. 4-lines),
and anterior–posterior (F-, C-, vs. P-lines) revealed asignificant main effect of Standard, F ð1; 18Þ ¼ 21:8,MSE ¼ :35, p < :001, as well as an interaction of Stan-
dard� left–right, F ð2; 36Þ ¼ 5:9, MSE ¼ :02, � ¼ :97,p < :01 (F ð2; 36Þ ¼ 4:5, MSE ¼ :11, � ¼ :97, p < :05 for
normalized data). The interaction was due to a stronger
effect along the midline than along the two lateral lines,
although the effect was significant for each of the three
anterior–posterior lines (see Appendix A for details onthe statistical analysis). The effect of the standards was
also assessed at the mastoids using an ANOVA with the
factors Standard (language-familiar vs. language-unfa-
miliar), language group (Hungarian vs. German), and
left–right (left vs. right mastoid). A significant main
Fig. 2. Frontal (Fz) grand-averaged (groups collapsed, see Section 2) ERPs (left-hand side) elicited by the Deviant (thick line) and the identical
stimulus Control (thin line) and their difference (right-hand side) at the frontal (Fz, thick line) and the left mastoid (Lm, thin line) leads. The four
conditions are displayed in separate rows. The onset of deviation (the point at which the Standards and Deviants started to differ) is marked by
dotted vertical lines. The full MMN responses (including the frontal as well as the reversed-polarity mastoid response) are marked by gray
shading.
T. Jacobsen et al. / Brain and Language 88 (2004) 54–67 61
effect of the Standard was obtained, F ð1; 18Þ ¼ 6:9,MSE ¼ :07, p < :05. One other effect was also significant
(see Appendix A). In sum, in this later window, there
was a broadly distributed effect of language-familiarversus language-unfamiliar Standards which was most
pronounced along the midline (Fz, Cz, and Pz) and
showed polarity reversal at the mastoids.
The different topography of the earlier and the later
difference suggests that they originate from, at least
partially, different brain areas. The polarity reversal
suggests that both effects were generated near the su-
pratemporal plane of temporal cortex. The presentfindings thus suggest that spoken language-familiar and
language-unfamiliar words invoke partially different
processes in the auditory cortex (see Fig. 1).
3.2. Mismatch negativity
As predicted, all four contrasts (2 types of Devi-
ants� 2 types of Standards) elicited significant MMN
responses (see Fig. 2, right-hand side), which was as-
sessed in the time window of 620–652ms from stimulus
onset (135–167ms after the onset of the auditory devi-
ation in the four items; see Section 2 and Table 1):
language-familiar Deviant in the context of language-familiar Standard, t ð19Þ ¼ 5:6, p < :001; language-
62 T. Jacobsen et al. / Brain and Language 88 (2004) 54–67
familiar Deviant in the context of language-unfamiliarStandard, t ð19Þ ¼ 4:5, p < :001; language-unfamiliar
Deviant in the context of language-familiar Standard,
t ð19Þ ¼ 5:1, p < :001; and language-unfamiliar Deviant
in the context of language-unfamiliar Standard,
Fig. 3. Grand-averaged (groups collapsed, see Section 2) MMN am-
plitudes in the four oddball conditions. Error markings on top of the
bars represent the standard error of the mean.
Fig. 4. Scalp distribution of the grand-averaged (groups collapsed, see Sectio
pseudoword, Standards (thick line), and the difference between MMNs elic
The onset of deviance is marked by dotted vertical lines.
t ð19Þ ¼ 3:9, p < :001. The MMN response had a higheramplitude in the context of language-familiar Standards
than language-unfamiliar Standards, irrespective of de-
viant type or language group (significant effect of
Standard context, F ð1; 18Þ ¼ 10:3, MSE ¼ :3, p < :01;no other effect approached significance, all Fs were <1).
In other words, MMN was larger in the native than in
the foreign language context (see Fig. 3 for the MMN
amplitudes).We asked whether the neural generators of this in-
crement in MMN are the same as or different from the
generators of the MMN elicited in the language-unfa-
miliar context. This was done by comparing the scalp
distribution of the difference between the MMNs from
the language-familiar context and the language-unfa-
miliar context with the MMN of the language-unfa-
miliar context MMN. If these scalp distributions differ,then one of the following two alternative statements is
true: (1) Deviants elicited a process in the language-fa-
miliar contexts which was additional to the processes
elicited by them in the language-unfamiliar contexts or
(2) a part of the deviance-related processes were different
in the language-familiar and language-unfamiliar con-
texts (i.e., the same Deviants were processed by different
sets of neurons depending on the lexicality of the con-text).
A repeated-measures ANOVA was conducted with
the factors condition (language-familiar context MMN
n 2) MMN component elicited in the context of language-unfamiliar,
ited in language-familiar and language-unfamiliar contexts (thin line).
T. Jacobsen et al. / Brain and Language 88 (2004) 54–67 63
minus language-unfamiliar context MMN vs. language-unfamiliar context MMN), anterior–posterior (the F-,
C-, vs. P-lines of electrodes) and left–right (the 3-, z-, vs.
4-lines of electrodes) using normalized data to eliminate
the main effect of condition. Significant interactions
between condition and topography were obtained: with
the anterior–posterior factor, F ð2; 38Þ ¼ 5:6, MSE ¼:88, � ¼ :58, p ¼ :024 and with the left–right factor,
F ð2; 38Þ ¼ 4:2, MSE ¼ :12, � ¼ :78, p ¼ :025. In short,the amplitude of the MMNs elicited in the context of
language-unfamiliar Standards was larger at frontal
sites whereas the amplitude of the MMN increment
(from the language-unfamiliar context to the language-
familiar context) was approximately evenly distributed
along the anterior–posterior direction (see Fig. 4). Also,
whereas the MMN elicited in the language-unfamiliar
context was approximately symmetrically distributedacross the two hemispheres, the difference between the
language-familiar and language-unfamiliar context
MMNs was larger over the right than the left hemi-
sphere (see Fig. 4). The observed interactions indicate
that the neural processes invoked by deviance in the
context of language-unfamiliar Standards were at least
partly different from those engaged by Deviants in the
context of language-familiar Standards (including thepossibility of additional neural generators being acti-
vated by change detection in the native-language context
compared with the foreign-language context).
4. Discussion
The effects of lexicality on the processing of spokenword-level items was investigated, when participants
had no task related to the stimuli. We found that the
default processing of frequently presented language-fa-
miliar words, Standards, involves partly different pro-
cesses than that of language-unfamiliar Standards.
Infrequently presented language-familiar words and
language-unfamiliar words, Deviants, presented in the
context of both language-familiar and language-unfa-miliar Standards elicited the MMN component. Neither
the amplitude nor the latency of the MMNs differed
between language-familiar Deviants and language-un-
familiar Deviants. However, the MMN amplitude elic-
ited in the oddball sequences having language-familiar
Standards was larger than that elicited in the sequences
with language-unfamiliar Standards. A topographic
analysis showed that processing Deviants in the lan-guage-familiar context involved at least partly different
processes compared with the deviance-related processes
occurring in the language-unfamiliar context. In the
following, we relate these results to the four hypotheses
described in Section 1.1 (the lexical trace, the supple-
mentary processing, the familiar context, and the lexical
context hypotheses).
4.1. The lexical trace hypothesis
The lexical trace hypothesis holds that deviancy in an
oddball sequence is processed differently for words and
pseudowords by virtue of the lexical status of the devi-
ants. When a word deviant is presented, a lexical
memory representation is pre-attentively activated and
used in the default memory-based comparison process
(reflected by MMN). Because words engage lexicalrepresentation in addition to the acoustic and phonetic
representations activated by pseudowords, the compar-
ison processes result in a stronger mismatch and thus
larger MMN amplitude for word than for pseudoword
deviants, irrespective of the lexical status of the in-se-
quence standard (Pulverm€uuller et al., 2001). However,
with both acoustic and language difference being con-
trolled, we found no evidence for the additional pro-cessing of language-familiar word Deviants as compared
with language-unfamiliar, or pseudoword, Deviants.
Thus the present results do not support the lexical trace
hypothesis.
4.2. The supplementary processing hypothesis
This hypothesis assumes that attempts at lexical ac-cess are (repeatedly) made by the system for every word-
like stimulus (whether it is a word or a pseudoword).
When lexical access fails for pseudoword Standards,
supplementary processing of these items takes place.
Additional processing leads to a more elaborate sensory
memory (or perhaps phonetic) representation for
pseudowords than for words, which results in a larger-
amplitude MMN component being elicited in the con-text of pseudoword than word standards (Diesch et al.
(1998)). The present results, which used appropriate
control for acoustic differences, did not support this
hypothesis, as we obtained opposite results: larger-am-
plitude MMNs to Deviants presented among language-
familiar word than language-unfamiliar, pseudoword,
Standards.
4.3. The familiar context hypothesis
The familiar context hypothesis assumes that the
auditory sensory memory representations created for
familiar sounds are more elaborate, richer in details,
sharper than the ones created for unfamiliar sounds. As
a consequence, when the standard sounds of an oddball
sequence are familiar to a participant, the resultingregularity representations contain more and better-
resolved details (see, e.g., Koelsch et al., 1999) as well as
the possibility of including more types of regularities
(e.g., Brattico et al., 2002) compared to unfamiliar
sounds. This results in a larger perceived difference
between the deviant and the standard sounds, which in
turn elicits MMNs of larger amplitude for deviants
T. Jacobsen et al. / Brain and Language 88 (2004) 54–67 65
of familiar auditory stimuli as opposed to unfamiliarones. This aspect of the present results will be further
investigated.
Appendix A. Details of the statistical data analyses
A.1. Standards
A.1.1. First time window
The ANOVA test of the factors Standard (within;language-familiar vs. language-unfamiliar), language
group (between; Hungarian vs. German), anterior–pos-
terior (within; the F-, C-, vs. P-lines of electrodes), and
left–right (within; the 3-, z-, vs. 4-lines of electrodes)
revealed a significant interaction between the stan-
dard� anterior–posterior, F ð2; 36Þ ¼ 7:3, MSE ¼ :06,� ¼ :64, p < :01 (F ð2; 36Þ ¼ 2:1, MSE ¼ :19, � ¼ :63,p ¼ :16 for normalized data). The interaction wassubsequently resolved for topography (see below). Fur-
thermore, this initial analysis revealed a main effect of
anterior–posterior, F ð2; 36Þ ¼ 34:8,MSE ¼ :63, � ¼ :59,p < :001, as well as interactions of Standard� group,
F ð1; 18Þ ¼ 7:7, MSE ¼ :78, p < :05, and Standard�group� anterior–posterior, F ð2; 36Þ ¼ 9:0, MSE ¼ :06,� ¼ :64, p < :01. None of these effects required further
analysis. The main effect of ERP amplitude betweensites (anterior, posterior) is not related to the experi-
mental conditions. The two interactions involving group
are confounded with a stimulus difference and are,
therefore, not interpretable. Note that this confound is
controlled in the relevant Standard� anterior–posterior
interaction.
Three separate mixed between-subject repeated-mea-
sures ANOVAs with the factors language group, Stan-dard, and left–right were computed for the F-, C-, and
P-lines in order to resolve the Standard� anterior–pos-
terior interaction of the former analysis for topography.
For the F-line, there was a significant effect of Standard,
F ð1; 18Þ ¼ 6:6, MSE ¼ :25, p < :05. No other effect was
significant, most F s < 1. There were no significant main
effects of Standard in the C- and P-line analyses. Non-
interpretable (see above) interactions of Stan-dard� group were obtained (C-line analysis, F ð1;18Þ ¼ 7:5, MSE ¼ :31, p ¼ :01; P-line analysis,
F ð1; 18Þ ¼ 12:9, MSE ¼ :34, p < :01). No other effects
were significant.
In the ANOVA with the factors Standard (within;
language-familiar vs. language-unfamiliar), language
group (between; Hungarian vs. German), and left–
right (within; left vs. right mastoid), in addition to thesignificant main effect of Standard (F ð1; 18Þ ¼ 10:4,MSE ¼ :09, p < :01; see the text), the main effect
of left–right was also significant, F ð1; 18Þ ¼ 5:8,MSE ¼ :02, p < :05. The latter effect is not related to
the experimental conditions. No other effects weresignificant.
A.1.2. Second time window
The ANOVA test of the factors Standard (within;
language-familiar vs. language-unfamiliar), language
group (between; Hungarian vs. German), left–right
(within; 3-, z-, vs. 4-lines), and anterior–posterior(within; F-, C-, vs. P-lines) revealed a significant main
effect of Standard, F ð1; 18Þ ¼ 21:8, MSE ¼ :35,p < :001, as well as an interaction of Standard� left–
right, F ð2; 36Þ ¼ 5:9, MSE ¼ :02, � ¼ :97, p < :01(F ð2; 36Þ ¼ 4:5, MSE ¼ :11, � ¼ :97, p < :05 for nor-
malized data). This interaction was subsequently re-
solved for topography (see below). Furthermore, the
initial analysis revealed main effects of anterior–poster-ior, F ð2; 36Þ ¼ 23:9, MSE ¼ :25, � ¼ :70, p < :001, andleft–right, F ð2; 36Þ ¼ 19:4,MSE ¼ :04, � ¼ 1:0, p < :001,as well as interactions of Standard� group, F ð1;18Þ ¼ 4:3, MSE ¼ :35, p ¼ :054, anterior–posterior�left–right, F ð4; 72Þ ¼ 10:5, MSE ¼ :008, � ¼ :55, p <:001, and Standard� group� left–right, F ð2; 36Þ ¼ 4:6,MSE ¼ :02, � ¼ :97, p < :001. None of these effects re-
quired further analysis. Main effects and interactions ofEEG amplitude between sites (anterior, posterior; left,
right) are not related to the experimental conditions.
The two interactions involving group are confounded
with a stimulus difference and are, therefore, not inter-
pretable. Note that this confound is controlled in the
relevant standard main effect and standard� anterior–
posterior interaction.
Three separate mixed between-subject repeated-mea-sures ANOVAs with the factors Standard, group, and
anterior–posterior were computed separately for the 3-,
z-, and 4-lines in order to resolve the Standard� ante-
rior–posterior interaction of the former analysis for to-
pography. The effect of the Standard was significant in
all three analyses. It was strongest along the midline,
F ð1; 18Þ ¼ 25:5,MSE ¼ :17, p < :001. In addition, there
was a main effect of anterior–posterior, F ð2; 36Þ ¼ 23:1,MSE ¼ :06, � ¼ :73, p < :001, and an interaction of
Standard� group F ð1; 18Þ ¼ 4:5, MSE ¼ :17, p < :05.No other effect was significant. For the 3-line the main
effect of the Standard was F ð1; 18Þ ¼ 18:8, MSE ¼ :09,p < :001. There was also an effect of anterior–posterior,
F ð2; 36Þ ¼ 13:3, MSE ¼ :07, � ¼ :70, p < :001. No other
effect was significant, most Fs were <1. At the 4-line the
effect of Standard was F ð1; 18Þ ¼ 16:3, MSE ¼ :13,p < :001. In addition, there was a main effect of
anterior–posterior, F ð2; 36Þ ¼ 33:0, MSE ¼ :06, � ¼ :74,p < :001, and an interaction of standard� group
F ð1; 18Þ ¼ 6:1, MSE ¼ :13, p < :05. No other effect was
significant.
The effect of the Standards was also assessed at the
mastoids using an ANOVAs with the factors Standard
66 T. Jacobsen et al. / Brain and Language 88 (2004) 54–67
(within; language-familiar vs. language-unfamiliar),language group (between; Hungarian vs. German), and
left–right (within; left vs. right mastoid). In addition to
the significant main effect of Standard (F ð1; 18Þ ¼ 6:9,MSE ¼ :07, p < :05, see the text), the main effect of
left–right was also significant, F ð1; 18Þ ¼ 18:9,MSE ¼ :02, p < :001. The latter main effect is not re-
lated to the experimental conditions. No other effects
were significant.
A.2. Mismatch negativity
The repeated-measures ANOVA with the factors
condition (language-familiar context minus language-
unfamiliar context MMN vs. language-unfamiliar con-text MMN), anterior–posterior (the F-, C-, vs. P-lines of
electrodes) and left–right (the 3-, z-, vs. 4-lines of elec-
trodes) using normalized data yielded two condi-
tion� topography interactions (see text). In addition, the
main effects of the topography factors were also signifi-
cant: anterior–posterior, F ð2; 38Þ ¼ 22:0, MSE ¼ :28,� ¼ :80, p < :001, and left–right, F ð2; 38Þ ¼ 7:6, MSE ¼:17, � ¼ :93, p < :01. These effects, however, are notrelated to the present conditions. No other effect was
significant, most F s < 1.
References
Aaltonen, O., Eerola, O., Hellstrom, �AA., Uusipaikka, E., & Lang, A.
H. (1997). Perceptual magnet effect in the light of behavioral and
psychophysiological data. Journal of the Acoustical Society of
America, 101, 1090–1106.
Alho, K. (1995). Cerebral generators of mismatch negativity (MMN)
and its magnetic counterpart (MMNm) elicited by sound changes.
Ear and Hearing, 16, 38–51.
American Electroencephalographic Society (1991). American electro-
encephalographic society guidelines for standard electrode position
nomenclature. Journal of Clinical Neurophysiology 8, 200–202.
Atienza, M., & Cantero, J. L. (2001). Complex sound processing
during human REM sleep by recovering information from long-
term memory as revealed by the mismatch negativity (MMN).
Brain Research, 901, 151–160.
Brattico, E., N€aa€aat€aanen, R., & Tervaniemi, M. (2002). Context effects
on pitch perception in musicians and non-musicians: Evidence
from ERP recordings. Music Perception, 19, 1–24.
Cheour, M., Ceponiene, R., Lehtokoski, A., Luuk, A., Allik, J., Alho,
K., & N€aa€aat€aanen, R. (1998). Development of language-specific
phoneme representations in the infant brain. Nature Neuroscience,
1, 351–353.
Cheour, M., Leppaenen, P. H. T., & Kraus, N. (2000). Mismatch
negativity (MMN) as a tool for investigating auditory discrimina-
tion and sensory memory in infants and children. Clinical Neuro-
physiology, 111, 4–16.
Cowan, N., Winkler, I., Teder, W., & N€aa€aat€aanen, R. (1993). Short- and
long-term prerequisites of the mismatch negativity in the auditory
event-related potential (ERP). Journal of Experimental Psychology:
Learning, Memory, and Cognition, 19, 909–921.
Cs�eepe, V., Osman-S�aagi, J., Moln�aar, M., & Gosy, M. (2001). Impaired
speech perception in aphasic patients: Event-related potential and