See

discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/8931460

Grammar Processing Outside the Focus of

Attention: An MEG Study
ARTICLE in JOURNAL OF COGNITIVE NEUROSCIENCE DECEMBER 2003
Impact Factor: 4.69 DOI: 10.1162/089892903322598148 Source: PubMed

CITATIONS

DOWNLOADS

VIEWS

65

99

4 AUTHORS, INCLUDING:
Yury Shtyrov

Friedemann Pulvermller

Aarhus University

Freie Universitt Berlin

92 PUBLICATIONS 2,873 CITATIONS

234 PUBLICATIONS 11,872 CITATIONS

SEE PROFILE

SEE PROFILE

Risto J Ilmoniemi
Aalto University
304 PUBLICATIONS 15,734 CITATIONS
SEE PROFILE

Available from: Risto J Ilmoniemi

Retrieved on: 15 September 2015

Grammar Processing Outside the Focus of Attention:

an MEG Study
a
ta
nen3,4 ,
Yury Shtyrov1 , Friedemann Pulvermu
ller 1 , Risto Na
2,4
and Risto J. Ilmoniemi

Abstract
& To address the cerebral processing of grammar, we used
whole-head high-density magnetoencephalography to record
the brains magnetic fields elicited by grammatically correct
and incorrect auditory stimuli in the absence of directed
attention to the stimulation. The stimuli were minimal short
phrases of the Finnish language differing only in one single
phoneme (word-final inflectional affix), which rendered them
as either grammatical or ungrammatical. Acoustic and lexical
differences were controlled for by using an orthogonal design
in which the phonemes effect on grammaticality was inverted.
We found that occasional syntactically incorrect stimuli elicited

INTRODUCTION
For over a century, the processing of language in the
brain has remained one of the most intriguing issues in
science. Although its many aspects have been tackled by
numerous studies, little is still known of how our central
nervous system goes about processing grammar. Grammar is one of the main intrinsic properties of the human
language. The very presence of a grammatical system
distinguishes our ability of verbal communication from
all signaling systems used by animals. Earlier investigations of the neurophysiology of syntactic processing
(e.g., Hagoort, Brown, & Groothusen, 1993; Osterhout
& Holcomb, 1992; Neville, Nicol, Barss, Forster, & Garrett, 1991) have led to important insights into this issue.
Three major syntax-related phenomena have been
described on the basis of the brains electric responses
to language stimuli. The first one is the so-called early
left anterior negativity (ELAN) occurring, with a latency
of 125 msec, in response to function words1 whose
placement in a phrase violates sentence structure rules
(Neville et al., 1991). Secondly, grammatically related
frontal negativities with longer latencies (over 250 msec)
have been found, for example, for different morphological or syntactic errors (Gunter, Friederici, & Schriefers,
1

Medical Research Council, Cognition and Brain Sciences Unit,

Cambridge, UK, 2 Biomag Laboratory, Helsinki University
Central Hospital, 3 University of Helsinki, 4 Helsinki Brain
Research Center
2003 Massachusetts Institute of Technology

larger mismatch negativity (MMN) responses than correct

phrases. The MMN was earlier proposed as an index of preattentive automatic speech processing. Therefore, its modulation by grammaticality under nonattend conditions suggests
that early syntax processing in the human brain may take place
outside the focus of attention. Source analysis (singledipole
models and minimum-norm current estimates) indicated
grammaticality dependent differential activation of the left
superior temporal cortex suggesting that this brain structure
may play an important role in such automatic grammar
processing. &

2000; Mu
nte, Schiltz, & Kutas, 1998). Finally, a late
positive shift (often termed P600), reaching its maximum at 600 msec and maximal at centro-parietal
recording sites, has been considered the most robust
neurophysiological response to ungrammatical sentences because it could be observed in many linguistic
experiments (for review, see Osterhout, McLaughlin, &
Bersick, 1997; Osterhout & Hagoort, 1999).
When studying the processing of grammar in the brain,
it seems important to control for a number of issues. For
instance, in such studies, subjects are typically asked to
attend to presented sentences (e.g., Friederici, Wang,
Herrmann, Maess, & Oertel, 2000; Osterhout & Swinney,
1993; Neville et al., 1991); often, the task is to judge the
sentence grammaticality. In such cases, as attention is
required, one cannot be sure to what extent the registered responses are influenced by brain correlates of
attention rather than by the language-related activity as
such. Attention-related phenomena are known to modulate a variety of the brains evoked responses in a
substantial span of time after stimulus onset and to
involve a number of brain structures including those
close to, or overlapping with, the core language areas
(see, e.g., Yamasaki, LaBar, & McCarthy, 2002; Yantis
et al., 2002; Escera, Alho, Winkler, & Naatanen, 1998;
Tiitinen, May, & Naatanen, 1997; Woods, Alho, & Algazi,
1993; Woods, Knight, & Scabini, 1993; Alho, 1992; Picton
& Hillyard, 1974). It also appears possible that subjects
pay more attention to incorrect sentences as they try to
Journal of Cognitive Neuroscience 15:8, pp. 11951206

make some sense of them or that they process grammatical and ungrammatical items using different strategies.
These different strategies and attention variation may
find their reflection in the event-related measures, overlapping with true syntax-related activity. Therefore, limiting the attention-related effects seems to be necessary
at this stage for further investigation of the neural
processing of syntax.
A methodologically important aspect in such experiments is stimulus selection. Grammatically correct and
incorrect phrases would inevitably differ in many other
features: e.g., sound onset times and durations in
acoustic stimulation, as well as visual geometry and
overall luminance in visual tasks, would change when
words or affixes are added, removed, or relocated to
modulate grammaticality. Differences even in basic
physical features may lead to differential brain activation
(Korth & Nguyen, 1997; Naatanen & Picton, 1987) that
could in principle overlap with or be misinterpreted as
language-related effects. For example, it was shown that
word-elicited responses are highly dependent on the
word length (Osterhout, Bersick, & McKonnon, 1997;
Assadollahi & Pulvermu
ller, 2003), suggesting the necessity to control for such factors. Furthermore, the
physical stimulus parameters are of crucial importance
for obtaining ERP effects of syntactic processes (Gunter,
Friederici, & Hahne, 1999). It seems most logical to us
to apply an experimental design in which physical
differences between the grammatical and ungrammatical items are controlled for by presenting the same
physical contrasts to the subjects both in ungrammatical
and grammatical contexts; this sort of an orthogonal
design would help to disentangle purely syntactic effects
from those related to the physical stimulus features.
So, in sum, it appears crucial to limit the influence of
attention and different processing strategies on the
brain response to grammatical and ungrammatical
strings as well as to apply maximal control over stimulus
features using a counterbalanced stimulus design. Controlling for these issues simultaneously in a single study
can be a challenging task, which, as to our knowledge,
has not been realized so far. The present study represents an attempt to achieve this goal.
Here, we addressed brain activity related to grammatical processing in the absence of directed attention
towards stimuli and without an active task which would
possibly invite subjects to invent a strategy for dealing
with unusual stimuli. We used a counterbalanced stimulus design, which allowed us to control for effects of
the physical and lexical parameters of the stimuli. To
gain high temporal resolution necessary in such an
experiment, we opted for multichannel magnetoencephalographic (MEG) recording of brain responses,
which can also provide information about the spatial
location of the registered activity.
We utilized the so-called mismatch negativity (MMN),
a unique indicator of automatic cerebral processing of
1196

Journal of Cognitive Neuroscience

acoustic stimuli, which can be used to investigate the

cerebral processing of speech and language (Shtyrov,
Kujala, Palva, Ilmoniemi, & Naatanen, 2000; Shtyrov &
Pulvermu
ller, 2002a, 2002b; Naatanen & Winkler, 1999;
Naatanen, 2001; Pulvermu
ller et al., 2001). MMN, with
its major sources of activity in both the supratemporal
and frontal cortices, is a brain response elicited by rare
(deviant) stimuli occasionally presented in a sequence
of frequent (standard) stimuli (Opitz, Rinne, Mecklinger,
von Cramon, & Schro ger, 2002; Picton, Alain, Otten,
Ritter, & Achim, 2000; Alho, 1995). It can be registered
as a negative deflection in the scalp electroencephalographic recording (EEG), or, as a component in the
brains magnetoencephalogram. Importantly, MMN (and
its magnetic counterpart, MMNm, or mismatch field),
can be elicited in the absence of the subjects attention
(Naatanen, 1995; Na
atanen & Escera, 2000). It is considered to reflect the brains automatic discrimination of
changes in the auditory sensory input and, furthermore,
to provide an index of experience-dependent memory
traces in the human brain (Naatanen, 2001; Shtyrov
et al., 2000; Kraus, McGee, Carrell, & Sharma, 1995).
Recent evidence suggests that the MMN reflects brain
processing of such language elements as phonemes
(Naatanen et al., 1997; Na
atanen, 1999, 2001), syllables
(Shtyrov et al., 2000; Alho et al., 1998), and words
(Shtyrov & Pulvermu
ller, 2002b; Korpilahti, Krause,
Holopainen, & Lang, 2001; Pulvermu
ller et al., 2001).
Importantly, we have recently found a specific pattern
of MMN responses to inflectional affixes (Shtyrov &
Pulvermu
ller, 2002a), which is best explained by the
activation of distinct cortical memory traces for such
affixes realized as distributed strongly connected populations of neurons (Pulvermu
ller, 1999, 2001). Inflectional (functional) affixes are important tools utilized in
many languages for realizing syntax and conveying
grammatical information. This suggests that the MMN
could be a sensitive tool for probing neural elements
responsible for the processing of grammar in the brain.
We have therefore set out to investigate the cerebral
processing of syntax using a whole-head high-density
MEG system, and recorded MMNm elicited by either
grammatically correct or ungrammatical phrases, while
the subjects were distracted from the auditory stimulation by engaging in a different activity. The stimuli were
short pronounverb phrases differing only in their final
phoneme, which rendered them either as grammatical
or as grammatically incorrect (see Figure 1, Table 1, and
Methods section for details). Any possible effects of
acoustic (phonetic) differences were ruled out by using
an orthogonal design reversing the effect of the phoneme and word combination on grammaticality.

RESULTS
Magnetic MMN responses were elicited in all conditions.
The overall analysis of single equivalent current dipole
Volume 15, Number 8

Figure 1. Spectrograms of acoustic Finnish-language stimuli used in the four experimental conditions. The standard and deviant stimuli in each
pair are indiscriminable up to the divergence point (marked with white arrowheads) at their end. The standarddeviant contrasts are identical
across all experimental conditions. Grammatically incorrect phrases are marked with an asterisk (*). See Table 1 for more details.

(ECD) sources showed that the grammaticality of the

deviant stimuli had a significant effect on the MMNm
dipole moment, F(1,8) = 6.35, p < .04. Separate
analyses for each hemisphere showed that the grammaticality had no effect for the right hemisphere alone
( p > .65), whereas in the left hemisphere its effect was
significant, F(1,8) = 6.67, p < .04: syntactically incorrect phrases elicited stronger MMNm activity than did
the correct ones (see Figures 2, 3, and 4a). This effect

also became manifest as a significant Pronoun Suffix

interaction ( p < .04). It neither depended on the
direction of the acoustic contrast between the suffixes
( p > .12) nor on the preceding pronoun ( p > .63) as
such, but solely on the correctness of the phrase as a
whole (Figure 4a).
The L1 minimum-norm current estimate (MCE)
analysis of the MEG signal suggested activation of
distributed cortical sources spread out over the left

Table 1. Auditory Finnish-Language Stimuli Used in the Four Experimental Sessions

Deviant

Standard

Condition 1a

Condition 1b

Condition 2a

Condition 2b

Ma
tuon

*Ma
tuot

*Sa
tuon

Sa
tuot

(I bring)

(*I bring)

(*you bring)

(you bring)

*Ma
tuot

Ma
tuon

Sa
tuot

*Sa
tuon

(*I bring)

(I bring)

(you bring)

(*you bring)

English translations are given in parentheses, ungrammatical phrases are marked with (*). The standard and deviant stimuli in each pair are
indiscriminable up to the divergence point at their end. The standarddeviant contrasts are identical across all experimental conditions. Critical
contrasts are in bold.

Shtyrov et al.

1197

Figure 2. ECD models of MMNm activity in the left hemisphere (grand average). Both ungrammatical deviant stimuli produced stronger MMNm
source than the corresponding correct phrases regardless of the acoustic, phonetic, phonological, or lexical properties of the stimuli. Red and blue
contour lines indicate positive and negative values of the normal component of the magnetic field on the helmet-shaped MEG-array surface,
respectively. The arrows show the dipole location and are proportional in relative size to the response magnitude. The view is from the left side of
the head.

temporal cortex (see Figure 5). This activation was

significantly stronger (Figure 4b) for the syntactically
incorrect than for correct deviant phrases, F(1,11) =
9.46, p < .011. Again, this difference was clearly
affected by the grammaticality, but neither by the pronoun context ( p > .25) nor by the suffix ( p > .8).

No differences between conditions were found in the

right hemisphere.
As one could notice in Figure 5 (upper left panel),
some additional activation could be suggested to appear
also in the left inferior frontal cortex, which would be
consistent with earlier results (Neville et al., 1991). We

Figure 3. Waveforms
(magnetic field gradient) of
MMNm responses in the left
hemisphere (grand average).
Both ungrammatical deviant
stimuli produced stronger
MMNm than the corresponding
correct phrases regardless of
the acoustic, phonetic,
phonological, or lexical
properties of the stimuli.
Temporal gradiometer with the
most prominent responses
(channel 0242) was used for
producing this diagram.

1198

Journal of Cognitive Neuroscience

Volume 15, Number 8

Figure 4. Grand-average mean

and standard error of mean of
the MMNm dipole-moment
magnitude (nanoamperemeter,
nAm) in the left hemisphere:
(A) calculated for single
equivalent current sources
(ECD), (B) calculated for L1
MCE (calculated as a sum of
dipole moments of all
simultaneously activated
adjacent current sources). Both
analyses indicated that the
ungrammatical deviant stimuli
produced a stronger MMNm
source than the corresponding
correct phrases regardless of
the acoustic, phonetic,
phonological, or lexical
properties of the stimuli.

analyzed activity in this area as well, but although we

found that, similarly to the supratemporal source, it was
somewhat increased for the ungrammatical deviant
phrases, this effect was not significant.
In the current study, the grammaticality affected the
MMNm responses, but had no significant effect on the
responses to the standard stimuli only (cf. Figures 3
and 6). We also found no effects of grammaticality on
the latency of the MMNm responses, the mean latency in
the left hemisphere being 213 msec after the divergence
point (mean SEM 19 msec). We also looked for significant
differences between the conditions in later time intervals

(>350 msec) but did not find any. Specifically, we would

like to note that no late deflection of the magnetic
response, which could be related to the P3- or P600-like
components of the EEG signal (Osterhout, McLaughlin
et al., 1997; Donchin, 1981), could be detected (see
Figures 3 and 6). We also noticed that the [n]-ending
stimuli produced larger MMNm dipole moment than did
the [t]-ending ones. However, this independent of the
grammaticality effect (which could in principle be due
to differences in attack and voicedness between the two
phonemes) was only significant for comparison between
Conditions 2a and 2b ( p < .02; but p > .95 for

Figure 5. Grand-average L1
MCEs of MMNm in the left
hemisphere. The MCE solutions
are projected unto
triangularized gray matter
surface. Both ungrammatical
deviant stimuli produced
stronger MMNm source than
the correct phrases regardless
of the acoustic, phonetic,
phonological, or lexical
properties of the stimuli.

Shtyrov et al.

1199

Figure 6. Waveforms
(magnetic field gradient)
of auditory responses to
standard stimuli in the left
hemisphere (grand
average). Although the
MMNm responses were
affected by grammaticality,
there was no significant
effect on the standard
responses (cf. Figure 3).
Temporal gradiometer
with the most prominent
responses (channel 0242)
was used for producing
this diagram.

Conditions 1a and 1b) and only in ECD but not in MCE

analysis, so it does not seem to present a solid finding.
Finally, we found no topographical effects of any of the
factors.

DISCUSSION
In the present study, mismatch responses were elicited
by grammatically correct or incorrect pronounverb
phrases. The choice of the single final phoneme in all
phrases rendered the whole structure as syntactically
correct or as ungrammatical. To exclude confounding
effects, the phonetic and lexical contrasts were counterbalanced over the experimental conditions using a fully
orthogonal stimulus design. The subjects were instructed
not to attend to the stimuli; they did not perform any
stimulus-related language task.
We found that MMNm responses to the syntactically
incorrect phrases were always larger than those to the
correct phrases ending in the same word regardless of the
direction of the acoustic contrast or of the preceding
pronoun. That is, neither the choice of the final phoneme
nor that of the pronoun in the sentence altered this effect,
which was present in both conditions in which the
deviant stimuli were grammatically inconsistent.
The present data are consistent with earlier results on
the cerebral processing of syntactic violations (see Introduction), which suggested negative-going components of brain responses at time delays of 125 msec or
more following syntactic errors. The present modification of the MMN may be related to this early negativity.
There are, however, important differences between the
present study and the earlier ones. Importantly, we were
able to find syntactically related effects in the absence of
directed attention towards the stimuli as subjects were
engaged in a different task and were instructed to ignore
the stimulation. The violations in this study modulated
the MMN response, which is considered to reflect a preattentive automatic level of auditory processing in the
cerebral cortex (Naatanen & Alho, 1995; Naatanen &
Escera, 2000; Tiitinen et al., 1997). This suggests that the
1200

Journal of Cognitive Neuroscience

brain may be capable of automatic syntactic analysis of

the incoming language signals already at relatively early
stages of speech processing. This conclusion is supportive of the earlier suggestion that early negativities, such
as ELAN, may reflect automatic stages of syntactic processing (Friederici, 2002; Gunter et al., 2000; Hahne &
Friederici, 1999).
The ELAN/N125 component reported in the previous
studies was found in response to phrase structure
violations (example: mathematicians *of proof the
theorem) and has been difficult to replicate (Takazawa
et al., 2002; Neville et al., 1991), whereas in the present
study the MMN modulation was elicited at a similar
tuo*t, sa
tuo*n),
latency by agreement violations (ma
suggesting that a larger variety of grammatical violations,
including agreement violations, could be reflected by
early left-lateralized neurophysiological responses.
An established view in psycholinguistics is that sentence comprehension is incremental, that is, a substantial
amount of processing occurs immediately after a word in
a sentence is perceived, prior to the perception of the
following word (Pulvermu
ller, 2001; Traxler & Pickering,
1996; Marslen-Wilson & Tyler, 1975; Marslen-Wilson,
1987). This implies that by the time a violation occurs, a
certain amount of analysis has been carried out by the
parsing system, a position largely shared by several
syntactic parsing models that may substantially differ
otherwise (e.g., Friederici, 2002; van Gompel & Pickering,
2001; Spivey & Tanenhaus, 1998; MacDonald, Pearlmutter, & Seidenberg, 1994; Trueswell, Tanenhaus, & Kello,
1993; Ferreira & Henderson, 1990; Frazier, 1987; Mitchell,
1987). During such analysis, an expectation for syntactically possible subsequent morphemes may have built
up. When this expectation fails to realize and the parser
therefore encounters a syntactic error (i.e., a mismatch
between preceding and incoming morphemes), it sends
an error signal. Such an error signal may be due to a
lack of neuronal links between lexical representations
which results in a failure to provide priming between the
representations of morphemes that do not match syntactically (Pulvermu
ller, 2002). In our case, the memory trace
Volume 15, Number 8

(neuronal assembly) representing the expected correct

affix (Shtyrov & Pulvermu
ller, 2002a) would have been
primed by the preceding context. Language-related
priming effects are known to reduce negative-going
components of the event-related potentials (e.g., Holcomb & Neville, 1990; Bentin, McCarthy, & Wood, 1985).
In ungrammatical conditions, this priming would be
absent which should lead to an increased response as
compared with grammatically consistent stimuli. Thus,
the parsers error signal, the unprimed activation of
morpheme-related neuron ensembles, may be the basis
of syntactically related early left-lateralized responses,
including the larger MMN response to grammatical violations observed here. Interestingly again, this putative
process seems to occur outside the focus of attention
(i.e., at least to some degree automatically).
The strongest activation produced by the syntactic
abnormalities in the current study was located in the left
temporal cortex. We take this as an indication that there
is a distributed neuronal system in the left superior
temporal lobe that contributes to grammar processing.
This area has been repeatedly suggested as a neural
substrate for a variety of language functions such as
phonetic/phonological processing, lexical access, semantic and syntactic processing (e.g., Shtyrov et al., 2000;
Shtyrov & Pulvermu
ller, 2002a, 2002b; Martin & Chao,
2001; Pulvermu
ller et al., 2001; Friederici et al., 2000;
Price, 2000; Helenius, Salmelin, Service, & Connolly,
1998; Naatanen et al., 1997; Binder et al., 1995), which
is in line with the current results. There probably could
be an overlap between these functions with respect to
underlying neuroanatomical structures, but it seems
reasonable to suggest that they are carried out by at
least partially distinct neuronal networks (Pulvermu
ller,
2002). It also appears justified to suggest that the leftlateralized neural systems contributing to the grammatically related MMN in the superior temporal lobe may be
distinct from those responsible for classical bilateral
acoustic-related MMNs.
Interestingly, although we found the strongest activation in the temporal cortex, a left anterior-frontal response was reported earlier (Neville et al., 1991;
Pulvermu
ller & Shtyrov, in press). As we found no
significant grammatically dependent variation for the
inferior frontal activity, we conclude that in the present
data the frontal source did not seem to play a major role
in syntactic processing. There could be several possible
explanations for this: in principle, it could be that the
advanced source analysis algorithm used in this study,
MCE, provides more accurate source localization than
potential mappings used earlier. As a viable alternative
explanation, one could suggest that the anterior source
might not be detectable in MEG due to its spatial
orientation (and/or, vice versa, that the posterior source
may be difficult to image using EEG). Both of these
explanations are supported by earlier MEG research,
which has also demonstrated more temporal than frontal

activity to syntactic violations when subjects were asked

to judge grammaticality of auditory presented sentences
(Friederici et al., 2000). However, a frontal ELAN generator was reported in MEG, although this was also accompanied by a temporal one (Gross et al., 1998). Finally, it is
possible that the absence of directed attention to the
stimuli and of an active task may have reduced the frontal
activity. This would imply that the inferior frontal source
is of smaller importance for automatic syntactic parsing.
To fully answer the question of the cortical generator(s)
for the syntactic MMN modulation observed here and
their relation to ELAN activation, further studies using
similar paradigm would be needed.
Finally, we could not find brain dynamics that might
relate to the late centro-parietal positive shift (P600)
reported earlier in ERP studies of grammatical violations.
Again, one could suggest that cortical generators of late
slow positive shifts are not optimal for MEG recordings,
although, among late positivities, at least the P3 was
successfully registered using MEG (Berg, Kakigi, Scherg,
Dobel, & Zobel, 1999; Mecklinger et al., 1998; Tarkka,
Stokic, Basile, & Papanicolaou, 1995). However, the latter
still does not rule out the possibility that the syntactically
related P600 response has no clear magnetic correlate.
Another possible reason for the absence of a late grammaticality effect in our present data set is the absence of
an attention-demanding task. It has been proposed
earlier that the positive shifts to grammatical violations
reflect attention-dependent processes, for example, attempts at a structural reanalysis of a sentence fragment
(Osterhout & Holcomb, 1992). In this view, the present
absence of clear dynamics in P600 time range (also
confirmed in a separate EEG study, Pulvermu
ller &
Shtyrov, in press) might suggest a role of attention in
the generation of the late positivity. Because subjects did
not focus their attention on language in this study, they
most likely did not engage in re-parsing the deviant
strings, and thus the centro-parietal positive shift did
not occur. This would be consistent with notions that
P600 reflects phrase reanalysis and repair (Friederici,
1997; Osterhout, Holcomb, & Swinney, 1994) and may
not reflect specific syntactic processes per se (Mu
nte,
Heinze, Matzke, Wieringa, & Johannes, 1998).
In conclusion, the present results suggest that the early
syntax parsing of spoken language may not require
focused attention and therefore could probably be automatic to a certain extent. The cortical structures in the left
superior temporal lobe appear to play an important role
in carrying out such automatic grammar processing.

METHODS
Subjects
Twelve healthy right-handed (handedness assessed
according to Oldfield, 1971; no left-handed family members) native Finnish speakers (age 2129, 7 women)
Shtyrov et al.

1201

with normal hearing and no record of neurological diseases were presented with four sets of spoken
native-language stimuli in four separate experimental
conditions.
Stimuli
The four short-phrase stimuli (see Figure 1 and Table 1)
were prepared with requirements that the acoustic,
phonetic, and phonological difference between the
standard and the deviant stimuli is identical in each
condition and that the stimuli themselves are as similar
acoustically as possible. These Finnish stimuli were:
(1a) Ma
tuon (I bring, syntactically correct),
(1b) *Ma
tuot (*I bring, syntactically incorrect: the
verb tuot is in the second person inflection form, while
the pronoun ma
, I, requires the first person); here
and throughout we use linguistic convention of marking
ungrammatical phrases with an asterisk (*),
tuon (*you bring, syntactically incorrect:
(2a) *Sa
the verb tuon is in the first person form, while the
pronoun sa
, you, requires the second person);
(2b) Sa
tuot (you bring, syntactically correct). 2
Therefore, it is the single final phoneme, inflectional affix [n] or [t], which may render the whole phrase
as syntactically correct or as ungrammatical. Remarkably,
for ma-stimuli (1a and 1b), the final [n] means that
the phrase is grammatical, while for the sa-stimuli
(2a and 2b) the combination is exactly the opposite:
[t] is needed in the end for the whole phrase to be
correct. This orthogonal design is implemented in
order to control for purely acoustic, phonetic, and phonological effects on mismatch responses as well as
for putative lexical and semantic differences between
the words.
As seen in Table 1 presenting the complete stimulus
design, the stimuli starting with the same pronoun were
always presented in the same experimental condition,
one of them serving as deviant and the other one as
standard stimulus. Thus, the standarddeviant acoustic
phonetic contrast, the critical variable determining the
MMN (Naatanen & Alho, 1997), was the same in all
conditions. However, in each pair of conditions (1a, b
and 2a, b) the responses were elicited either by correct
or incorrect deviant forms. Furthermore, while the
acoustic contrasts were identical in the corresponding
conditions (tuon/tuot in 1a2a and tuot/tuon in 1b2b),
the syntactic contrast (determined by the pronoun) was
the opposite: stimuli ending in tuon were correct in
conditions 1a and 1b, whereas tuot-ending phrases were
correct in Conditions 2a and 2b. This fully orthogonal
design, therefore, allowed us to control not only for
possible acoustic/phonetic effects of the verb contrast,
but even for an unlikely late differential effect of the two
preceding pronouns.
1202

Journal of Cognitive Neuroscience

For stimulus preparation, we recorded multiple repetitions of each word uttered by a female native speaker
of Finnish. With great care we selected a combination of
recordings, whose vowels matched in their fundamental
frequency (F0) and whose overall length and maximal
sound energy were identical both for sa
and ma
and for
tuon and tuot. We then combined the four words in the
four abovementioned phrases with a 50-msec gap between the pronoun and the verb; all phrases were
766 msec in duration. Thus, the two stimuli in each
condition were perceptually identical up to a time point
(referred to as divergence point) shortly before their
end (see Figure 1). In order to determine this divergence point, a gating experiment was performed, in
which initial fragments of increasing duration of both
tuon and tuot were presented to the subjects who were
asked whether they could perceive any difference between the two. In average, they could identify the
difference only when the first 336 msec (SEM 9 msec)
of the words were presented (which equals to 636 msec
after the onset of the complete phrase). This perceived
divergence point is therefore considered as the onset
of the perceptual standard/deviant contrast and a zero
time reference in the current study. Importantly, it is
the same in all pairs of stimuli due to their design.
All stimuli were normalized to have the same peak
sound energy. For the analysis and production of the
stimuli, we used the Cool Edit 96 program (Syntrillium
Software, AZ).
Acoustic Stimulation
All four experimental conditions (Table 1) were performed with every subject, their order being counterbalanced across the subject group. The stimuli were
binaurally presented at 50 dB above the individual
hearing threshold (determined using the experimental
stimuli) via headphones connected to an STIM setup
(Neuroscan Labs, VA). The interstimulus (stimulus
onset asynchrony, SOA) interval was 1450 msec. In each
condition, the deviant stimulus was presented with a
probability of 16.7% among the repetitive standard
stimuli: A pseudorandom sequence of stimuli was created so that there were always at least two standard
stimuli between any two deviants. During the stimulation, the subjects were seated in a magnetically shielded
chamber and instructed to watch a silent video film
of their own choice and to pay no attention to the
auditory stimulation.
Magnetoencephalographic Recording
The evoked magnetic field was recorded (passband
0.03200 Hz, sampling rate 600 Hz) with 204 planar
gradiometer channels of a whole-head Neuromag Vectorview MEG system (Elekta Neuromag, Helsinki) during
the auditory stimulation (Ahonen et al., 1993). The
Volume 15, Number 8

recordings started 100 msec before stimulus onset

and ended 1450 msec thereafter. The responses were
on-line averaged separately for the standard and deviant stimuli in each condition. Epochs with voltage
variation exceeding 150 mV at either of two bipolar
eye-movement electrodes or magnetic-field gradient
variation exceeding 3000 femtotesla per centimeter (fT/
cm) at any gradiometer channel were excluded from
averaging. The recordings for each condition contained
at least 120 accepted responses to deviant stimuli (corresponding to 600 standard responses) in order to
achieve satisfactory averaged signal quality. This requirement led to slight variations in overall recording duration, which in average constituted 2 hr per subject.
MEG Data Processing
The averaged responses were filtered off-line (passband
120 Hz). The period of 50 msec before the divergence
point was used as the baseline. The MMNm was obtained by subtracting the averaged response to the
standard stimuli from that to the deviant ones. The
responses were evaluated separately for each subject
for all experimental conditions. Two approaches were
used for analyzing the neuromagnetic data.
Single-Dipole Fit
Fifty-four gradiometer channels on each side of the
magnetometer helmet were used for assessing cortical
responses in the left and right cerebral hemispheres. By
means of Neuromag sequential single-dipole fitting
software (Elekta Neuromag), the generator sources
(equivalent current dipoles [ECDs]) of the MMNm were
estimated (Ilmoniemi, 1993). Only dipole models explaining more than 65% of the field gradients were
selected for statistical analysis; otherwise, the dipolemoment value was set to 0 nAm. The range of latencies
(as determined from peak dipole moment) accepted for
the MMNm responses was 100350 msec after the
divergence point and the best fit (maximal goodness
of fit/dipole moment) was entered into analysis. The
dipole moments of the MMNm generators were calculated and compared between the experimental conditions. ECD fits were successfully constructed only for
nine subjects, suggesting that the single-dipole model
may not be optimal for explaining the syntactically
related activity. We therefore performed minimumnorm current analysis, which uses a distribution of
multiple current sources for modeling the neuromagnetic signal.
L1 Minimum-Norm Current Estimate
The estimation of the cortical sources of the measured
neuromagnetic activity was performed using L1 MCEs
on the basis of recordings from all 204 gradiometer

channels equally distributed over the entire scalp. The

minimum-norm method provides a solution to the inverse problem of localizing neural activity in the
brain from its external recordings by revealing the
unique constellation of active neuronal current elements that models the recorded magnetic field distribution with the smallest amount of overall activity
(Uutela, Hama
lainen, & Somersalo, 1999; Ilmoniemi,
1993; Hamalainen & Ilmoniemi, 1984, 1994). Among the
infinitely many current source combinations that can
explain a given scalp topography of the magnetic
field, this technique finds one that does so with the
least amount of overall activity. This minimal solution
should be preferred for the sake of parsimoniousness.
The L1 MCE minimizes the integral of the current
amplitude and reveals a realistic and robust constellation
of localized generators. It is applicable when it can be
assumed a priori that the source distribution consists of
discrete areas of neuronal activity (in contrast with a
situation when no essential a priori information is
available, in which case L2 minimum norm estimate
would be preferable). This assumption of localized
sources is justified in the investigation of language
because neuroimaging results obtained with different
methodologies proved the activation of discrete cortical
areas during a variety of language tasks (Pulvermu
ller,
2001; Price, 2000). A triangularized gray matter surface
was used for projecting MCE solutions for the each
subject individually and for the grand average magnetic
field. The dipole moments of the MMNm generators
were calculated (as a sum of dipole moments of all
simultaneously activated adjacent current sources) and
compared between the experimental conditions. The
MCE source analysis was successfully carried out for all
12 subjects, which suggests that it could be a better
method for analyzing distributed neural activity (as in
case of language-related phenomena) than ECD.
Statistical Analysis
The data were subjected to analyses of variance (ANOVA). Parameters of current sources constructed for
each hemisphere were analyzed separately for ECD
and MCE. We compared them between conditions using
the factors grammaticality (syntactically correct vs. in vs. ma
), and
correct deviant stimulus), pronoun (sa
suffix ([t] vs. [n]).
Ethical Considerations
All subjects gave their written informed consent to
participate in the experiments and were paid for
their participation. The experiments were performed
in accordance with the Helsinki Declaration. Ethical
permission for the experiments was issued by the Research Ethics Committee of Helsinki University Central
Hospital.
Shtyrov et al.

1203

Acknowledgments
We thank Olaf Hauk, Vadim Nikulin, Maritta Maltio-Laine, Simo
Monto, Tuomas Murdoch, Christopher Bailey, Jussi Nurminen,
Johanna Salonen, Anthea Hills, Seppo Kahko
nen, and William
Marslen-Wilson for their contribution at different stages of this
work. We would also like to thank three anonymous referees
for their helpful comments and constructive critique.
Reprint requests should be sent to Dr. Yury Shtyrov, Medical
Research Council, Cognition and Brain Sciences Unit, 15
Chaucer Road, CB2 2EF Cambridge, UK, or via e-mail:
yury.shtyrov@mrc-cbu.cam.ac.uk.

Notes
1. Function words are grammatical words, such as articles,
conjunctions, auxiliary verbs, and so on and so forth, which
have no concrete meaning by themselves and are utilized for
grammatical purposes.
2. These spoken-language stimuli represent the most usual
colloquial pronunciation used in the Helsinki area; therefore,
we only used Finnish subjects who had lived in the Helsinki
area for a minimum of 6 years before the experiment.

REFERENCES
Ahonen, A. I., Hamalainen, M., Kajola, M., Knuutila, J., Laine, P.,
Lounasmaa, O., Parkkonen, L., Simola, J., & Tesche, C.
(1993). 122-channel SQUID instrument for investigating the
magnetic signals from the human brain. Physica Scripta,
T49, 198205.
Alho, K. (1992). Selective attention in auditory processing
as reflected by event-related brain potentials.
Psychophysiology, 29, 247263.
Alho, K. (1995). Cerebral generators of mismatch negativity
(MMN) and its magnetic counterpart (MMNm) elicited by
sound changes. Ear and Hearing, 16, 3851.
Alho, K., Connolly, J. F., Cheour, M., Lehtokoski, A.,
Huotilainen, M., Virtanen, J., Aulanko, R., & Ilmoniemi,
R. J. (1998). Hemispheric lateralization in preattentive
processing of speech sounds. Neuroscience Letters, 258,
912.
Assadollahi, R., & Pulvermu
ller, F. (2003). Early influences of
word length and frequency: a group study using MEG.
Neuroreport, 14, 11831187.
Bentin, S., McCarthy, G., & Wood, C. C. (1985). Event-related
potentials, lexical decision and semantic priming.
Electroencephalography and Clinical Neurophysiology, 60,
343355.
Berg, P., Kakigi, R., Scherg, M., Dobel, C., & Zobel, E. (1999).
Source modelling of the EEG and MEG oddball response in a
subject with a large P300. Electroencephalography and
Clinical Neurophysiology Supplement, 49, 189193.
Binder, J. R., Rao, S. M., Hammeke, T. A., Frost, J. A.,
Bandettini, P. A., Jesmanowicz, A., & Hyde, J. S. (1995).
Lateralized human brain language systems demonstrated by
task subtraction functional magnetic resonance imaging.
Archives of Neurology, 52, 593601.
Donchin, E. (1981). Surprise! ... Surprise? Psychophysiology,
18, 493513.
Escera, C., Alho, K., Winkler, I., & Naata
nen, R. (1998). Neural
mechanisms of involuntary attention to acoustic novelty
and change. Journal of Cognitive Neuroscience, 10,
590604.
Ferreira, F., & Henderson, J. M. (1990). Use of verb information
1204

Journal of Cognitive Neuroscience

in syntactic parsing: Evidence from eye movements and

word-by-word self-paced reading. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 16,
555568.
Frazier, L. (1987). Sentence processing: A tutorial review. In M.
Coltheart (Ed.), The psychology of reading (Vol. 12,
pp. 559586). Hillsdale, NJ: Erlbaum.
Friederici, A. (2002). Towards a neural basis of auditory
sentence processing. Trends in Cognitive Sciences, 6,
7884.
Friederici, A. D. (1997). Neurophysiological aspects of language
processing. Clinical Neuroscience, 4, 6472.
Friederici, A. D., Wang, Y., Herrmann, C. S., Maess, B., & Oertel,
U. (2000). Localization of early syntactic processes in frontal
and temporal cortical areas: A magnetoencephalographic
study. Human Brain Mapping, 11, 111.
Gross, J., Ioannides, A. A., Dammers, J., Maess, B., Friederici,
A. D., & Muller-Gartner, H. W. (1998). Magnetic field
tomography analysis of continuous speech. Brain
Topography, 10, 273281.
Gunter, T. C., Friederici, A. D., & Hahne, A. (1999). Brain
responses during sentence reading: visual input affects
central processes. NeuroReport, 10, 31753178.
Gunter, T. C., Friederici, A. D., & Schriefers, H. (2000).
Syntactic gender and semantic expectancy: ERPs reveal early
autonomy and late interaction. Journal of Cognitive
Neuroscience, 12, 556568.
Hagoort, P., Brown, C., & Groothusen, J. (1993). The syntactic
positive shift (SPS) as an ERP-measure of syntactic
processing. Language and Cognitive Processes, 8, 439483.
Hahne, A., & Friederici, A. D. (1999). Electrophysiological
evidence for two steps in syntactic analysis. Early automatic
and late controlled processes. Journal of Cognitive
Neuroscience, 11, 194205.
Hamalainen, M. S., & Ilmoniemi, R. J. (1984). Interpreting
measured magnetic fields of the brain: minimum norm
estimates of current distributions. Helsinki University of
Technology, Technical Report TKK-F-A559.
Hamalainen, M. S., & Ilmoniemi, R. J. (1994). Interpreting
magnetic fields of the brain: minimum norm estimates.
Medical and Biological Engineering and Computing, 32,
3542.
Helenius, P., Salmelin, R., Service, E., & Connolly, J. F. (1998).
Distinct time courses of word and context comprehension in
the left temporal cortex. Brain, 121, 11331142.
Holcomb, P. J., & Neville, H. J. (1990). Auditory and visual
semantic priming in lexical decision: A comparison using
event-related brain potentials. Language and Cognitive
Processes, 5, 281312.
Ilmoniemi, R. J. (1993). Models of source currents in the brain.
Brain Topography, 5, 331336.
Korpilahti, P., Krause, C. M., Holopainen, I., & Lang, A. H.
(2001). Early and late mismatch negativity elicited by words
and speech-like stimuli in children. Brain and Language,
76, 332339.
Korth, M., & Nguyen, N. X. (1997). The effect of stimulus size
on human cortical potentials evoked by chromatic patterns.
Vision Research, 37, 649657.
Kraus, N., McGee, T., Carrell, T. D., & Sharma, A. (1995).
Neurophysiologic bases of speech discrimination. Ear and
Hearing, 16, 1937.
MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S.
(1994). The lexical nature of syntactic ambiguity resolution.
Psychological Review, 101, 676703.
Marslen-Wilson, W., & Tyler, L. K. (1975). Processing structure
of sentence perception. Nature, 257, 784786.
Marslen-Wilson, W. D. (1987). Functional parallelism in spoken
word-recognition. Cognition, 25, 71102.
Volume 15, Number 8

Martin, A., & Chao, L. L. (2001). Semantic memory and the

brain: Structure and processes. Current Opinion in
Neurobiology, 11, 194201.
Mecklinger, A., Maess, B., Opitz, B., Pfeifer, E., Cheyne, D., &
Weinberg, H. (1998). A MEG analysis of the P300 in visual
discrimination tasks. Electroencephalography and Clinical
Neurophysiology, 108, 4556.
Mitchell, D. C. (1987). Lexical guidance in human parsing:
Locus and processing characteristics. In M. Coltheart (Ed.),
The psychology of reading. (Vol. 12, pp. 601618). Hillsdale,
NJ: Erlbaum.
Mu
nte, T. F., Heinze, H. -J., Matzke, M., Wieringa, B. M., &
Johannes, S. (1998). Brain potentials and syntactic violations
revisited: No evidence for specificity of the syntactic positive
shift. Neuropsychologia, 36, 217226.
Mu
nte, T. F., Schiltz, K., & Kutas, M. (1998). When temporal
terms belie conceptual order. Nature, 395, 7173.
Naatanen, R. (1995). The mismatch negativity: A powerful
tool for cognitive neuroscience. Ear and Hearing, 16,
618.
Naatanen, R. (1999). Phoneme representations of the human
brain as reflected by event-related potentials. Functional
Neuroscience: Evoked Potentials and Magnetic Fields, EEG
Supplement, 170173.
Naatanen, R. (2001). The perception of speech sounds by the
human brain as reflected by the mismatch negativity (MMN)
and its magnetic equivalent (MMNm). Psychophysiology, 38,
121.
Naatanen, R., & Alho, K. (1995). Mismatch negativityA
unique measure of sensory processing in audition.
International Journal of Neuroscience, 80, 317337.
Naatanen, R., & Alho, K. (1997). Mismatch negativityThe
measure for central sound representation accuracy.
Audiology and Neurootology, 2, 341353.
Naatanen, R., & Escera, C. (2000). Mismatch negativity: Clinical
and other applications. Audiology and Neurootology, 5,
105110.
Naatanen, R., Lehtokoski, A., Lennes, M., Cheour, M.,
Huotilainen, M., Iivonen, A., Valnio, A., Alku, P., Ilmoniemi,
R. J., Luuk, A., Allik, J., Sinkkonen, J., & Alho, K. (1997).
Language-specific phoneme representations revealed by
electric and magnetic brain responses. Nature, 385,
432434.
Naatanen, R., & Picton, T. (1987). The N1 wave of the human
electric and magnetic response to sound: A review and an
analysis of the component structure. Psychophysiology, 24,
375425.
Naatanen, R., & Winkler, I. (1999). The concept of auditory
stimulus representation in cognitive neuroscience.
Psychological Bulletin, 12, 826859.
Neville, H., Nicol, J. L., Barss, A., Forster, K. I., & Garrett, M. F.
(1991). Syntactically based sentence processing classes:
Evidence from event-related brain potentials. Journal of
Cognitive Neuroscience, 3, 151165.
Opitz, B., Rinne, T., Mecklinger, A., von Cramon, D. Y., &
Schro
ger, E. (2002). Differential contribution of frontal and
temporal cortices to auditory change detection: fMRI and
ERP results. Neuroimage, 15, 167174.
Osterhout, L., Bersick, M., & McKonnon, R. (1997). Brain
potentials elicited by words: Word length and frequency
predict the latency of an early negativity. Biological
Psychology, 46, 143168.
Osterhout, L., & Hagoort, P. (1999). A superficial resemblance
does not necessarily mean you are part of the family:
Counterarguments to Coulson, King and Kutas (1998) in the
P600/SPS-P300 debate. Language and Cognitive Processes,
14, 114.
Osterhout, L., & Holcomb, P. J. (1992). Event-related brain

potentials elicited by syntactic anomaly. Journal of Memory

and Language, 31, 785806.
Osterhout, L., Holcomb, P. J., & Swinney, D. A. (1994). Brain
potentials elicited by garden-path sentences: Evidence of the
application of verb information during parsing. Journal of
Experimental Psychology: Learning, Memory, and
Cognition, 20, 786803.
Osterhout, L., McLaughlin, J., & Bersick, M. (1997).
Event-related brain potentials and human language. Trends
in Cognitive Sciences, 1, 203209.
Osterhout, L., & Swinney, D. A. (1993). On the temporal course
of gap-filling during comprehension of verbal passives.
Journal of Psycholinguistic Research, 22, 273286.
Picton, T., & Hillyard, S. (1974). Human auditory evoked
potentials: II. Effects of attention. Electroencephalography
and Clinical Neurophysiology, 36, 191200.
Picton, T. W., Alain, C., Otten, L., Ritter, W., & Achim, A. (2000).
Mismatch negativity: Different water in the same river.
Audiology and Neurootology, 5, 111139.
Price, C. J. (2000). The anatomy of language: Contributions
from functional neuroimaging. Journal of Anatomy, 197,
335359.
Pulvermu
ller, F. (1999). Words in the brains language.
Behavioral and Brain Sciences, 22, 253336.
Pulvermu
ller, F. (2001). Brain reflections of words and their
meaning. Trends in Cognitive Sciences, 5, 517524.
Pulvermu
ller, F. (2002). A brain perspective on language
mechanisms: From discrete neuronal ensembles to serial
order. Progress in Neurobiology, 67, 85111.
Pulvermu
ller, F., Kujala, T., Shtyrov, Y., Simola, J., Tiitinen, H.,
Alku, P., Alho, K., Martinkauppi, S., Ilmoniemi, R. J., &
Naatanen, R. (2001). Memory traces for words as revealed by
the mismatch negativity. Neuroimage, 14, 607616.
Pulvermu
ller, F., & Shtyrov, Y. (in press). Automatic processing
of grammar in the human brain as revealed by the mismatch
negativity. Neuroimage.
Shtyrov, Y., Kujala, T., Palva, S., Ilmoniemi, R. J., & Naatanen,
R. (2000). Discrimination of speech and of complex
nonspeech sounds of different temporal structure in the
left and right cerebral hemispheres. Neuroimage, 12,
657663.
Shtyrov, Y., & Pulvermu
ller, F. (2002a). Memory traces for
inflectional affixes as shown by mismatch negativity.
European Journal of Neuroscience, 15, 10851091.
Shtyrov, Y., & Pulvermu
ller, F. (2002b). Neurophysiological
evidence of memory traces for words in the human brain.
NeuroReport, 13, 521525.
Spivey, M. J., & Tanenhaus, M. K. (1998). Syntactic ambiguity
resolution in discourse: Modeling the effects of referential
context and lexical frequency. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 24,
15211543.
Takazawa, S., Takahashi, N., Nakagome, K., Kanno, O.,
Hagiwara, H., Nakajima, H., Itoh, K., & Koshida, I. (2002).
Early components of event-related potentials related to
semantic and syntactic processes in the Japanese language.
Brain Topography, 14, 169177.
Tarkka, I. M., Stokic, D. S., Basile, L. F., & Papanicolaou,
A. C. (1995). Electric source localization of the auditory
P300 agrees with magnetic source localization.
Electroencephalography and Clinical Neurophysiology, 96,
538545.
Tiitinen, H., May, P., & Naatanen, R. (1997). The transient
40-Hz response, mismatch negativity, and attentional
processes in humans. Progress in
Neuro-Psychopharmacology and Biological Psychiatry, 21,
751771.
Traxler, M. J., & Pickering, M. J. (1996). Case-marking in the
Shtyrov et al.

1205

parsing of complement sentences: Evidence from eye

movements. Quarterly Journal of Experimental Psychology,
49, 9911004.
Trueswell, J. C., Tanenhaus, M. K., & Kello, C. (1993).
Verb-specific constraints in sentence processing: Separating
effects of lexical preference from garden-paths. Journal of
Experimental Psychology: Learning, Memory, and
Cognition, 19, 528553.
Uutela, K., Hamala
inen, M., & Somersalo, E. (1999).
Visualization of magnetoencephalographic data using
minimum current estimates. Neuroimage, 10, 173180.
van Gompel, R. P., & Pickering, M. J. (2001). Lexical guidance in
sentence processing: A note on Adams, Clifton, and Mitchell
(1998). Psychonomic Bulletin and Review, 8, 851857.
Woods, D. L., Alho, K., & Algazi, A. (1993). Intermodal selective

1206

Journal of Cognitive Neuroscience

attention: evidence for processing in tonotopic auditory

fields. Psychophysiology, 30, 287295.
Woods, D. L., Knight, R. T., & Scabini, D. (1993). Anatomical
substrates of auditory selective attention: Behavioral and
electrophysiological effects of posterior association cortex
lesions. Brain Research. Cognitive Brain Research, 1,
227240.
Yamasaki, H., LaBar, K. S., & McCarthy, G. (2002). Dissociable
prefrontal brain systems for attention and emotion.
Proceedings of National Academy of Sciences, U.S.A., 99,
1144711451.
Yantis, S., Schwarzbach, J., Serences, J. T., Carlson, R. L.,
Steinmetz, M. A., Pekar, J. J., & Courtney, S. M. (2002).
Transient neural activity in human parietal cortex during
spatial attention shifts. Nature Neuroscience, 5, 9951002.

Volume 15, Number 8