Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
MirrorBot
IST-2001-35282
Bio-mimetic multi-modal learning in a mirror neuron-based robot
Category Specificity In The Processing Of Colour And Form Related Words
Authors: Friedemann Pulvermüller, Fermin Moscoso-del-Prado-Martin, Olaf HaukCovering period 1.9.2003-25.5.2004
MirrorBot Report 13
Report Version: 1
Report Preparation Date: 25. May. 2002
Classification: Restricted
Contract Start Date: 1st June 2002 Duration: Three Years
Project Co-ordinator: Professor Stefan Wermter
Partners: University of Sunderland, Institut National de Recherche en Informatique et en Automatique at Nancy, Universität Ulm, Medical Research Council at Cambridge, Università degli Studi di Parma
Project funded by the European Community under the “Information Society Technologies Programme“
0. Overview 3
1. Background 5
2. Method 8
3. Results 11
3. Discussion 14
4. Summary and Outlook 16
5. Implications for Computational Neuroscience and 18Brain-Inspired Robotics – Future Interaction with MirrorBot Partners
5. References 19
2
0 OVERVIEWThe Cambridge group in the Mirrobot project investigates the brain correlates of words and
their semantic and conceptual relationships to referent actions and objects. Multimodal
imaging techniques, including functional magnetic resonance imaging (fMRI), electro-
encephalography (EEG), magneto-encephalography (MEG) and trans-craneal magnetic
stimulation (TMS), were used to investigate words from different lexico-semantic categories.
In the first year of the project, action words were in the focus. We could demonstrate that
aspects of the meaning of action words are reflected in specific activation along the motor
strip and the premotor cortex. For example, words related to the hand (e.g., “pick”) activated
hand motor cortex, whereas words related to the leg (e.g., “kick”) activated leg motor cortex
(Hauk, Johnsrude, & Pulvermüller, 2004). This brain correlate of word meaning is evoked
already within 200 ms after information in the input is sufficient for word recognition (Hauk
& Pulvermüller, 2004). The patterns of semantic activation distinguishing between arm- and
leg-related words were present even if subjects did not attend to the stimulus words,
indicating a high degree of automaticity of semantic processes at the single word level
(Shtyrov, Hauk, & Pulvermüller, 2004). These findings are best explained by the existence of
widespread cortical cell assemblies that bind the neuron populations that process language
with those processing perception and actions, in which mirror neurons play a crucial role
(Pulvermüller, 2002; Rizzolatti, Fogassi, & Gallese, 2002). We envisage these cell assemblies
to include neurons in perisylvian cortex along with neurons in different parts of the motor
system. The proposal that category specific associative networks in the brain are the basis of
word, action and perceptual processing receives support from different lines of modern
neuroimaging research (cf., Pulvermüller, 2003) and could be implemented in
neurocomputational models (Wermter et al., 2004). In collaboration with Parma, TMS
experiments were performed to further study the functional relationships between language
and action. These efforts are detailed in Mirrobot Report 4 of WP2, Reports 7 and 8 of WP1,
Report 11 of WP 10 and mentioned in several others.
In the second year, the Cambridge group continued their research efforts to uncover the
cortical basis of action word processing, an area which had proven to be extremely fruitful
during the first year of the project. In parallel, the group investigated the processing of fine-
grained subcategories of words whose meanings relate to different types of visual
information. Words related knowledge about colors and shapes, for example “brown” and
“round”, were investigated using EEG, MEG, and fMRI. We expected word category
differences to be present in the inferior temporal “what”-stream of visual processing (Ishai,
Ungerleider, Martin, Schouten, & Haxby, 1999; Martin & Chao, 2001b; Mishkin,
Ungerleider, & Macko, 1983). In particular, the middle temporal gyrus has been described to
be involved in the association of objects with their particular colours (Zeki & Marini, 1998).
3
Therefore we expected an greater activation of middle temporal gyrus upon presentation of
colour-related words than following presentation of form-related words. In a similar manner,
because shapes, but not colours, relate to abstract action knowledge – an abstract shape can
always be sketched by distinctive movements which can be performed with different parts of
the body – we expected higher prefrontal and premotor activation during the processing of
form-related words than in the processing of colour-related words. The evaluation of fMRI
and MEG results is still in progress. We present here results of an EEG study and a modeling
effort undertaken to calculate the cortical activation underlying the processing of colour and
form related words. Further work was done to study the neural basis of serial order and
syntax. These research efforts will be discussed in a later report.
4
Arm word Leg word
Figure 1: Model of the cortical distribution of neuron ensembles linking word and action representations; possible cortical networks for arm and leg related wirds are illustrated (top). Sources of magnetoencephalographic activity seen for words semantica
1. BACKGROUND
Among the most intensely debated issues in the cognitive neuroscience of language is the
question of the cortical “seat” of word meaning (Martin & Chao, 2001a; Pulvermüller, 1999).
Although there is little doubt that areas in left inferior frontal and superior temporal cortex –
sometimes referred to as Broca’s and Wernicke’s regions – play a major role in language
processing, the location of additional areas possibly contributing to semantic processing
remains controversial. Most theories localize meaning-related mechanisms in areas anterior,
inferior, and posterior to Wernicke’s area in the left temporal lobe (Hickok & Poeppel, 2000;
Hodges, 2001; Price, Warburton, Moore, Frackowiak, & Friston, 2001; Scott & Johnsrude,
2003). However, since most studies investigating the issue have focussed on the cortical
processing of highly imageable concrete nouns and concepts related to their meaning, it is
possible that other word types engage semantic representations in other cortical regions.
When haemodynamic and neurophysiological imaging studies compared words referring to
objects with words that have a clear semantic relationship to actions, typically action verbs
(Dehaene, 1995; Kellenbach, Wijers, Hovius, Mulder, & Mulder, 2002; Preissl, Pulvermüller,
Lutzenberger, & Birbaumer, 1995; Pulvermüller, Preissl, Lutzenberger, & Birbaumer, 1996),
or nouns referring to tools (Chao, Haxby, & Martin, 1999; Ishai et al., 1999; Martin, Wiggs,
Ungerleider, & Haxby, 1996), the latter elicited strong frontal activation including premotor
cortex, suggesting that the frontal activation might reflect aspects of the action-related
meaning of action words (Martin & Chao, 2001a; Pulvermüller, 1996). If so, the cortical
locus of meaning processing could be, in part, determined by the general neuroscientific
principle of Hebbian learning according to which neuronal correlation is mapped onto
connection strength: (Hebb, 1949; Tsumoto, 1992). If word forms frequently co-occur with
visual perceptions (object words), their meaning-related activity may be found in temporal
visual areas, whereas action words frequently encountered in the context of body movements
may produce meaning-related activation in the fronto-central motor areas (Braitenberg, 1992;
Pulvermüller, 1996; Martin, 2001; Pulvermüller, 2003). In earlier studies (Hauk et al., 2004;
Hauk & Pulvermüller, 2004; Pulvermüller, Härle, & Hummel, 2000; Shtyrov et al., 2004), we
provided the first compelling evidence that word-meaning processing elicits specific action-
related activity patterns in fronto-central areas, including motor and premotor cortex. Action
words referring to different parts of the body lighted up fronto-central sensorimotor areas that
were also active when actions were performed with the word-related body parts. This
demonstrates a close link between the cortical substrate of language and action in the
processing of semantic meaning of words that refer to human actions.
5
Whether or not frontal areas also contribute to the processing of other word categories is a
matter of debate. According to the cell assembly model of Pulvermüller and Preissl (1991),
words of all kinds are organized by distributed neuron populations distributed over superior
temporal and inferior frontal perisylvian cortex, which processes acoustic and articulatory
information. This view, although well-supported by empirical data (Pulvermüller, 2003), is
questioned by some traditional models from aphasiology, according to which separate
modular processors in frontal and temporal lobe selectively contribute to language
comprehension and production, respectively (Hickok & Poeppel, 2000; Wernicke, 1874).
Word reading would, accordingly, only activate posterior areas but leave frontal areas
unaffected. This type of model is falsified by findings about frontal cortex activation elicited
by action words. However, action words form a small subset of the vocabulary and it may be
asked whether activation of specific frontal lobe areas is a more general phenomenon shared
by other word types, too. Here, we investigate word categories that relate to visual
perceptions: Words that were rated by experimental subjects to have a strong semantic
association to a colour, and words rated to be semantically related to a distinctive form or
shape. We hypothesized that, although both word categories would activate specific inferior
temporal areas in the ventral where-stream of visual processing, form-related words would
elicit additional activity in frontal cortex while colour-related words should elicit a higher
activation in the middle temporal gyrus. The additional activation in frontal lobe would reflect
the fact that an abstract form is always related to a distinct action pattern, a sequence of
movements that can be performed with different body parts to outline the shape. Similarly,
the higher involvement of the middle temporal gyrus in the processing of words with strong
associations to a colour would relate to the involvement of this area in the perception of
colours.
To investigate whether such a differentiation could be observed we performed an EEG
experiment similar to the one described by Hauk and Pulvermüller (2004). The usage of event
related potentials (ERPs) extracted from multi-channel EEG recording allows us to
investigate the precise time course of the differentiation of the words according to their
semantic categories. Source localisation techniques were used to pinpoint the cortical origin
of word-evoked neurophysiological activity.
All word types alike were assumed to activate areas in left inferior temporal lobe. It has been
claimed that this activation could reflect the engagement of a centre processing the meaning
of words, similar to Lichtheim’s “concept area” (Price, 2000). However, it has also been
postulated that an area in posterior inferior temporal lobe specialises in processing visual
word forms (VWF area; Cohen et al., 2000; 2002; Dehaene et al., 2002; Fiez & Pedersen,
1998; McCandliss et al., 2003; Polk & Farah, 2002) and, finally, there is the claim that
inferior temporal areas house word category specific processes (Martin & Chao, 2001). Given
6
the limited spatial resolution of EEG, we expected a general word related activation in
inferior temporal lobe initially, followed by spreading of activity either in temporal lobe
exclusively (colour words) or from temporal to frontal cortex (form words). Interestingly, the
activation of the VWF area was reported to occur between 150 ms and 200 ms following the
presentation of visual stimuli (Bentin et al., 1997; Helenius et al., 1999; McCandliss et al.,
2002; Tarkianen et al., 2002), which temporally overlaps with the activation of category
specific cortical areas reported by Hauk and Pulvermüller (2004) to happen at around 200 ms
from stimulus onset. In fact, even earlier activation of areas related to the meanings of words
has previously been reported by our group (Pulvermüller et al. 2001). This suggests that the
early activation of inferior temporal areas, might in fact correspond not only to the activation
of the orthographic features of the words, but also to their meanings. In this respect, the ERP
methodology allows us to investigate whether there are distinct temporal windows in which
semantic and specifically orthographic processes occur.
As made evident by Hauk and Pulvermüller (2004) and many others, Mininum Norm Current
Estimates calculated on the basis of EEG recordings are a suitable method to detect
differences in cortical activation elicited by words belonging to different semantic categories.
In the same manner, these topographical estimates should allow us to investigate whether the
some of early peaks in the ERPs and MEG signals do indeed correspond to the area where the
VWFA is generally located.
7
2. METHOD
Subjects
Ten monolingual native speakers of English participated in the study. Their age varied
between 18 and 31 years . They spent a minimum of 13 years on basic and higher level
education. All had normal or corrected-to-normal vision and reported no history of
neurological illness or drug abuse. Neuropsychological testing (Oldfield, 1971) revealed that
all of them were right-handed. Informed consent was obtained from all subjects and they were
paid for their participation. This study was approved by the Cambridge Psychology Research
Ethics Committee.
Stimuli
Stimuli were selected from databases using psycholinguistic criteria. A preliminary list of 403
words was evaluated in a behavioral study to assess the words’ cognitive, emotional, and
referential-semantic properties. This is necessary because words differing on corresponding
dimensions are known to elicit different neurophysiological responses (Kounios and
Holcomb, 1992; Pulvermüller, 1999; Skrandies, 1998). Native speakers of English (N = 15,
different from those participating in the EEG study) gave ratings on a 7-point scale answering
the following questions
· “How easily does this word evoke an image or any other sensory impression?”
(Imageability)
· “Do you evaluate this word or its meaning as pleasant or unpleasant?” (Valence)
· “How arousing is this word or its meaning?” (Arousal)
Ratings were given on a scale from 1 to 7. Additionally, participants were asked to answer on
similar 7-point scales whether the words reminded them of a particular colour or of a
particular shape. The results were evaluated statistically using F-tests. On the basis of this
evaluations, we selected 50 colour-related and 50 form-related words for presentation in the
EEG experiment, interspersed with 150 distractor words not related to colours or forms. The
word groups were matched with respect to mean word length, word form frequency according
to the CELEX database (Baayen et al., 1993), imageability (according to the ratings), valence
(according to the ratings), and arousal (according to the ratings). As an additional condition, a
8
sequence of hash marks matched to the length and luminance of each of the words was
included in the materials.
Procedure
Stimuli were presented for 100 ms each in white capital letters on a gray background in the
middle of a computer screen, subtending a horizontal visual angle smaller than 5 degrees. A
fixation cross was always present in the center of the screen between stimulus presentations.
Subjects were instructed to observe the stimuli silently, i.e., no overt response was required.
Stimuli were presented in pseudorandom order with a stimulus onset asynchrony randomly
varying between 2 and 3 sec. Two pseudo-randomized lists of stimuli were created, each
including all stimuli, which were alternated between subjects. Subjects were instructed to
reduce eye-blinks and movements as far as possible, and to restrict unavoidable movements to
the breaks within the experiment. The experimental session contained five breaks of 10 sec
duration.
Data recording
Electrophysiological data were collected in an electrically and acoustically shielded chamber
at the EEG laboratory of the MRC Cognition and Brain Sciences Unit in Cambridge, UK. The
EEG was recorded at a sampling rate of 500 Hz (0.1–30 Hz band-pass filter) from 64
Ag/AgCl electrodes mounted on an electrode cap (QuickCap, Neuromedical Supplies,
Sterling, VA) using SynAmps amplifiers (Neuro-Scan Labs, Sterling, VA). Electrodes were
placed according to the extended 10/20 system. EEG data were recorded against a reference at
AFz and converted off-line to average reference. The EOG was recorded bipolarly through
electrodes placed above and below the left eye (vertical) and at the outer canthi (horizontal).
After the actual experiment, subjects were instructed to blink and to move their eyes to the
left, right, up, and down, as indicated by symbols appearing on the computer screen. Average
responses to these eye movements were used for the correction of corresponding artifacts in
the EEG data (Berg and Scherg, 1994).
Data analysis
The continuously recorded neurophysiological data were divided into epochs of 1 ms length,
starting 200 ms before stimulus onset. Trials with voltage variations larger than 100μV in at
least one channel were rejected, and an eye artifact correction algorithm (Berg and Scherg,
1994) was applied. Data were band-pass filtered between 1–20 Hz. By averaging over
corresponding trials, event-related potentials (ERPs) were computed for every subject,
electrode, and stimulus category (HASH, COLOUR or FORM).
9
Source estimation
We calculated Minimum Norm Current Estimates using the method described by Hauk and
Pulvermüller (2004). We used a three-dimensional source space consisting of four concentric
equidistant “shells” (0.8–0.2 of electrode radius), with 1,965 current sources equally
distributed over these shells. At each location, three orthogonal sources were placed. Their
strengths were estimated separately, and then combined as the Euclidean vector length to
yield the intensity of activity at the corresponding location. For our analysis, we selected only
the uppermost shell at excentricity 0.8, containing 996 current sources (cf., Hauk et al., 1999).
The corresponding resolution vectors focus mostly on superficial cortical sources in this case
(Grave de Peralta et al., 1997). Tikhonov regularization (Bertero et al., 1988) was applied
such that the mean residual variance over data sets was 5%.
The left and right pre-auricular points and the nasion were determined for each subject and
used as landmarks for both the standard electrode configuration and the segmented skin
surface of the average magnetic resonance image (MRI) of the Montreal Neurological
Institute. These landmarks were used to co-register both modalities in the software package
ASA. In the spherical source model, the current sources that entered the statistical analysis
were located on a sphere below the electrodes, with 80% of the electrode radius. Co-
registration of the electrodes, therefore, also implied co-registration of the current sources
with the average MRI. The amplitudes of these current sources were then spherically
projected on the surface of the average brain.
Statistics
Peaks in the activation values were identified by plotting the root mean squares (RMS) of the
amplitudes for each of the conditions (see Figure 2).
In order to investigate the topographical differences elicited by the different types of stimuli,
we selected 4 regions of interest (ROIs) for the ERP analyses. These regions of interest were
chosen to provide contrasts in hemisphere (left or right) and frontality (anterior or posterior).
Each ROI included 7 electrode positions (left anterior: AF3, AF7, F3, F5, F7, FC3, and FC5;
right anterior: AF4, AF8, F4, F6, F8, FC4, and FC6; left posterior: P1, P3, P5, P7, PO3, PO9,
and O1; right posterior: P2, P4, P6, P8, PO4, PO10, and O2). For each of the 4 RMS peaks,
mean amplitudes were subjected to by-participant ANOVAs including the factors Lexicality,
Hemisphere, and Frontality. We only report significant interactions including the factor
Lexicality (word or hash marks). In a similar manner, the mean amplitudes to word stimuli
(colour- and form-related words, excluding the sequences of hash marks) were subjected to
ANOVAs including the factors Category (colour- or form-related), Hemisphere, and
Frontality. Once again, we only report those significant interactions of theoretical relevance.
10
3. RESULTS
Identification of peaks in the ERPs
The bottom part of Figure 2 shows the Root Mean Squares (RMS) of the amplitudes at all
recorded electrodes for each of the three types of stimuli in our experiment: hash marks
(green line), colour-related words (red), and form-related words (blue). The graph shows three
clear peaks in the RMS, centred approximately at 106 ms, 154 ms, and 254 ms from stimulus
onset. An additional peak centred at around 202 ms from stimulus onset is also apparent in the
case of the form-related words, and to a lesser degree in the colour-related words. The upper
panel in Figure 1 summarizes the scalp distribution of the amplitudes for the different types of
stimulus at each of the four peaks mentioned above.
96 to 116 ms time window
11
x
y
106.0 ms106.0 ms
Use cursor keys to rotate
5.20
-5.20EEG (uV)
x
y
154.0 ms154.0 ms
Use cursor keys to rotate
5.20
-5.20EEG (uV)
x
y
202.0 ms202.0 ms
Use cursor keys to rotate
5.20
-5.20EEG (uV)
x
y
254.0 ms254.0 ms
Use cursor keys to rotate
5.20
-5.20EEG (uV)
x
y
106.0 ms106.0 ms
Use cursor keys to rotate
5.37
-5.37EEG (uV)
x
y
154.0 ms154.0 ms
Use cursor keys to rotate
5.37
-5.37EEG (uV)
x
y
202.0 ms202.0 ms
Use cursor keys to rotate
5.37
-5.37EEG (uV)
x
y
254.0 ms254.0 ms
Use cursor keys to rotate
5.37
-5.37EEG (uV)
x
y
106.0 ms106.0 ms
Use cursor keys to rotate
5.19
-5.64EEG (uV)
x
y
154.0 ms154.0 ms
Use cursor keys to rotate
5.19
-5.64EEG (uV)
x
y
202.0 ms202.0 ms
Use cursor keys to rotate
5.19
-5.64EEG (uV)
x
y
254.0 ms254.0 ms
Use cursor keys to rotate
5.19
-5.64EEG (uV)
Colourwords
Formwords
Hashmarks
106 ms. 154 ms. 202 ms. 254 ms.
Figure 2. RMS peaks and their scalp distributions.
There was a significant main effect for the factor Frontality [F(1,9) = 10.28, p = 0.0101] and a
marginally significant interaction between the factors Frontality and Word Category [F(1,9) =
3.57, p = 0.0915].
144 to 164 ms time window
There was a significant main effect of the factor Hemisphere [F(1,9) = 9.15, p = 0.0001], and
significant interactions of the factors Hemisphere and Lexicality [F(1,9) = 23.06, p = 0.0010],
Frontality, Hemisphere, and Lexicality [F(1,9) = 13.44, p = 0.0052].
Figure 3 compares the Minimum Norm Current Estimates of words (colour-related and form-
related) with hash marks. Both in the case of colour-related and form-related words, the area
where the difference between the estimated currents between words (red) and hash marks
(blue) is greatest in left posterior temporal areas. In both cases, the difference in favour of
words peaks at a point whose estimated Tallairach coordinates are x = -46, y= -41, z = -13,
located in the left fusiform gyrus. In turn, the difference in favour of the hash marks is
maximal at the right posterior cingulate, with Tallairach coordinates x = 25, y = -69, z = 10.
192 to 212 ms time window
There was a significant main effect of the factor Frontality [F(1,9) = 12.40, p = 0.0065] and a
significant interaction for the factors Frontality and Word Category [F(1,9) = 6.96, p =
0.0270].
12
y
z
154.0 ms154.0 ms
Use cursor keys to rotate
0.14
-0.14EEG (uV)
y
z
154.0 ms154.0 ms
Use cursor keys to rotate
0.12
-0.12EEG (uV)
y
z
154.0 ms154.0 ms
Use cursor keys to rotate
0.14
-0.14EEG (uV)
y
z
154.0 ms154.0 ms
Use cursor keys to rotate
0.12
-0.12EEG (uV)
x
z
154.0 ms154.0 ms
Use cursor keys to rotate
0.14
-0.14EEG (uV)
x
z
154.0 ms154.0 ms
Use cursor keys to rotate
0.12
-0.12EEG (uV)
Colour - Hash
Form - Hash
Figure 3: Average Minimum Norm Current Estimates comparing words with hash marks in the 144-164 ms time window. Red corresponds to the points where the estimated current for words is greater than for hash marks, blue corresponds to the areas with a hig
Figure 4 compares the Minimum Norm Current Estimates for colour-related and form related
words. The areas where the activation elicited by form-related words is greater than that
elicited by colour-related words. Red indicates the areas with a higher current estimate for
colour-related than for form-related words.
Notice in Figure 4 that the difference in estimated currents in favour of colour-related words
peaks in the left temporal lobe, with approximate Tallairach coordinates x = -58, y = 1, z = -
14, corresponding to the left middle temporal gyrus, while the difference in favour of form-
related words is maximal at Tallairach coordinates x = -45, y = 48,z = -5 corresponding to the
left middle frontal gyrus, and also at x =55, y = -29, z = -9 which corresponds to the right
middle temporal gyrus.
244 to 264 ms time window
The were no significant effects in the ANOVAs for this time window.
13
y
z
202.0 ms202.0 ms
Use cursor keys to rotate
0.13
-0.14EEG (uV)
x
z
202.0 ms202.0 ms
Use cursor keys to rotate
0.13
-0.14EEG (uV)
y
z
202.0 ms202.0 ms
Use cursor keys to rotate
0.13
-0.14EEG (uV)
Figure 4: Average Minimum Norm Current Estimates comparing colour-related words with form-related in the 192-212 ms time window. Red corresponds to the points where the estimated current for colour-related words is greater than for form-related, blue corresponds to the points where the activation for form words is greater than for colour-words
4. DISCUSSION
The present study investigated the ERPs and Minimum Norm Estimates correlating visual
lexical processing. Our analyses concentrated in the contrast between existing words and
sequences of hash marks, and in the contrast between words with strong colour associations in
their meanings, and words with stron form associations in their meanings.
Differences between Words and Sequences of Hash Marks
We found a deflection on the ERP peaking at around 154 ms. According to the ANOVAs on
the amplitudes, this peak differentiates reliably between words and sequences of hash marks.
The differences in the Minimum Norm Current Estimates indicate that words show a higher
current estimate than the sequences of hash marks. This difference in favor of words peaks at
the left fusiform gyrus, while the hash marks show a maximum advantage on the right
posterior cingulate. The estimated Tallairach coordinates of the area where the activation was
superior for words (both colour-related and form-related) is remarkably close to what has
been reported in functional magnetic resonance (fMRI) experiments to be the centre of
activation of the so-called Visual Word Form Area (VWFA; Cohen et al., 2000; 2002;
Dehaene et al., 2002; Fiez & Pedersen, 1998; McCandliss et al., 2003; Polk & Farah, 2002)
which appears to respond selectively to visually presented words. This same area also appears
to be damaged in patients that suffer from word-form dyslexia (Binder & Mohr, 1992;
Warrington and Shallice, 1980). The deficits suffered by these patients include the impaired
ability to read words (but see also Behrmann et al., 1998). In particular, based on a meta-
analysis of 25 group experiments, Cohen et al. (2002) report that the peak of the VWFA is
located at approximately at x = -43, y = -54, z = -12 in Tallairach coordinates. According to
McCandliss et al. (2003) 90% of the individual subject scans locate the peak to the VWFA
within 5 mm of this location. In our analyses, the peak of the difference between words and
sequences of hash marks was estimated to lie at x = -46, y= -41, z = -13, which is about 13
mm away from the reported peak of the VWFA (and 9 mm away from the area where it is
found in 90% of the cases). Our estimate represents the centre of a larger region and it was
obtained using a Minimum Norm Current Estimate on the EEG amplitudes, which assumes a
single layer of possible activation, not including all ventral cortical areas. Therefore we
consider our estimate to be sufficiently close to correspond to the VWFA. Previous evidence
using ERPs in left posterior areas (Bentin et al., 1997, McCandliss et al., 2002),
magnetoencephalography (MEG; Helenius et al., 1999; Tarkianen et al., 2002), and
intracranial electrical recordings (Nobre et al., 1994), indicates that activation which is
specific to written words peaks between 150 ms and 200 ms from stimulus onset. This is also
14
in line with Assadollahi and Pulvermüller (2001), who reported access to cognitive
representations of words, including sensitivity to word length and word frequency, at the 120-
160 ms interval. Again, our results fit well into this general pattern, with the main peak of
activation at the VWFA occurring at around 154 ms. From this we conclude that the
deflection that we found at 154 ms, corresponds to the VWFA previously reported in the
literature. In our experiments, we found that the activation at this peak is not sensitive to
category specific information, thus providing support for the word-form specificity of this
area. This result is also of interest in that it provides an idea of the reliable spatial resolution
of the Minimum Norm Current Estimation technique by relating it to a previously identified
cortical area.
Differences between Colour-related and Form-related Words
We found a deflection in the ERP peaking at around 202 ms that, according to the ANOVAs,
differentiated significantly between colour-related and form-related words. The differences in
the Minimum Norm Current Estimates between colour- and form-related words showed an
advantage for colour-related words peaking around left middle temporal gyrus. At the same
time, the advantage form-related over colour-related words appeared to peak around left
middle frontal and right middle temporal gyri. This specific difference between words
belonging to different semantic categories occurs early in the processing, barely 50 ms after
the peak of activation of the VWFA, and at around the same time interval where Hauk and
Pulvermüller (2004) reported category specific differences between different types of action
words. Early differentiation with respect to semantic factors is in line with theories that posit
automatic activation of all the cortical sub-networks related to the processing of a word,
inmediately following the activation of the sub-networks that encode orthographic and/or
phonological forms of the words (e.g., Pulvermüller, 2001; 2003). In addition, the possible
implication of the right middle temporal gyrus in differentiating between colour-related and
form-related words, as revealed by the comparision of the Minimum Norm Current estimates
points in the direction of distributed, bi-hemispheric networks being involved in the
representation of the meanings of words (Neininger & Pulvermüller, 2001).. Finally, the
marginally significant interaction between the factors Frontality and Word Category at the
100 ms peak provides an indication of a possible earlier differentiation in the cortical
activation elicited by these two word-categories, even 50 ms before the time when activation
peaks in the VWFA. Activation of semantic properties of words at such an early interval is in
line with the MEG results of Pulvermüller et al. (2001), who reported a differentiation between
words with strong multimodal semantic associations occurring already at 100 ms after stimulus
presentation.
15
4. SUMMARY AND OUTLOOK
Taken together, these results support the view that words semantically related to visual
information do not only differentially activate the temporal lobe, but selectively activate
frontal cortical areas as well. In particular, form-related words were found to elicit higher
prefrontal and/or premotor activation than did colour-related words. This differential
activation distinguishing between semantic word categories followed ~50 ms upon activation
of an inferior temporal area by all words under study. This area may be identified as an area
specialized in processing visual word forms, or, as an alternative, as an area generally
contributing to semantic processing of words. The time course of cortical activation seen here
is therefore compatible with models postulating sequential processing of word form and
semantics, but does not necessarily exclude other models. The present spatio-temporal
activation pattern – general word-related activation at 150 ms and category specific activation
at 200 ms – is also compatible with the view that these partly overlapping and partly distinct
patterns reflect aspects of semantic processing.
Our preliminary fMRI results are in line with the EEG source localisation results and support
the view that colour-related words primarily activate inferior and ventral temporal areas
whereas form-related words lead to stronger activation in left frontal areas, including
premotor and prefrontal sites. It remains to be investigated from which particular aspect of the
meaning of form-related words does their premotor and prefrontal activation arise. As a
candidate process, we propose action programs associated with distinct visual shapes.
16
Figure 5: fMRI activation (p < 0.0001) to words related to colour and form relative to a baseline where strings of hash marks matched in lengths to the words were presented. Methods were the same as those described in Hauk et al. (2004).
The present results provide further support for the view that information about words and
their referent objects and actions is laid down in cortex by distributed cortical cell assemblies
whose cortial topography reveals features of the information stored. Among the cortical areas
participating in lexical and semantic storage, frontal areas seem to be of particular interest,
because, as our earlier results suggest, these areas may contribute to the specific linkage of
knowledge about actions, perceptions and language. The present data now indicate that
prefrontal areas may play a specific role in binding information about abstract action related
concepts possibly tied to words that name abstract forms.
17
5. IMPLICATIONS FOR COMPUTAIONAL NEUROSCIENCE AND
BRAIN-INSPIRED ROBOTICS – FUTURE INTERACTIONS WITH
MIRRORBOT PARTNERS
IMPLICATIONS FOR COMPUTAIONAL NEUROSCIENCE AND BRAIN-INSPIRED
ROBOTICS – FUTURE INTERACTIONS WITH MIRRORBOT PARTNERS
This research provides an additional piece of evidence supporting the view that words are
represented by networks with very specific and sometimes clearly different distributions over
the cortex, and that what information is stored about a word is partially revealed by where it is
stored, and when it is being activated. This “topography of meaning” (Pulvermüller, 2001)
can be captured by models in which language is intimately tied to representations of actions
and perceptions. Such models have been further developed and implemented in WP14, based
on close interactions between the groups in Ulm, Sunderland and Cambridge (see Mirrobot
Report 12.2; see Wermter et al., 2004).
New implications of the present research on visually-related words for the
neurocomputational and rebotics work in the Mirrobot project are the following: Sometimes,
words have, according to empircal results, no semantic relationship to specific body actions,
but nevertheless activate frontal lobe, in particular prefrontal cortex. It is possible that this
activation – here reported for abstract form words – relates to abstract action plans associated
with these lexical items. In a neural model implemented on a robot, this could inspire
“abstract action concept” areas (AACA) that would model processes possibly taking place in
prefrontal cortex. Such models could serve the same role as the arm-, face- and leg-action
modules in Pulvermüller’s (2001) and Wermter et al.’s (2004) model serve for action verbs.
An AACA could thus represent and process aspects of the meaning of abstract form related
words and possibly other abstract word categories. Clearly, such a model, if correct, has
implications for views on the nature of abstract concepts and the way they are laid down in
our brains and minds and, more generally, on the cortical relationship between language,
concepts and actions (Rizzolatti et al., 2002; Gallese, 2003).
18
6. REFERENCES
Assadollahi, R. and Pulvermüller, F. (2001) Neuromagnetic evidence for early access to
cognitive representations. Neuroreport 12, 207-213
Baayen, R.H. et al. (1993) The CELEX lexical database (CD-ROM). Linguistic Data
Consortium, University of Pennsylvania, Philadelphia, PA.
Behrmann, M. et al. (1998) Visual complexity in letter-by-letter reading: ‘pure’ alexia is not
pure. Neuropsychologia 36, 1115–1132
Bentin, S. et al. (1999) ERP manifestations of processing printed words at different
psycholinguistic levels: time course and scalp distribution. Journal of Cognitive Neuroscience
11, 235–260
Berg, P. and Scherg, M. (1994): A multiple source approach to the correction of eye artifacts.
Electroencephalography Clinical Neurolorgy 90, 229 –241
Bertero, M. et al. (1988) Linear inverse problems with discrete data. II. Stability and
regularisation. Inverse Problems 4, 573–594.
Binder, J.R. and Mohr, J.P. (1992) The topography of callosal reading pathways. a case-
control analysis. Brain 115, 1807–182
Braitenberg, V. and Schüz, A. (1998) Anatomy of the cortex. Statistics and geometry. Berlin:
Springer
Chao, L. L., Haxby, J. V., & Martin, A. (1999). Attribute-based neural substrates in temporal
cortex for perceiving and knowing about objects. Nature Neuroscience, 2(10), 913-919.
Cohen, L. et al. (2000) The visual word form area: spatial and temporal characterization of an
initial stage of reading in normal subjects and posterior split-brain patients. Brain 123, 291-
307
Cohen, L. et al. (2002) Language-specific tuning of visual cortex? Functional properties of
the visual word form area. Brain 125, 1054–1069
19
Dehaene, S. (1995). Electrophysiological evidence for category-specific word processing in
the normal human brain. Neuroreport 6(16), 2153-2157.
Dehaene, S. et al. (2002) The visual word form area: a prelexical representation of visual
words in the fusiform gyrus. Neuroreport 13, 321–325
Fiez, J.A. and Pedersen, S.E. (1998) Neuroimaging studies of word reading. Proceedings of
the National Academy of Sciences U.S.A. 95, 914-921
Fuster, J.M. (1998): Linkage at the top. Neuron 21, 1223–1224.
Gallese, V. (2003). The manifold nature of interpersonal relations: the quest for a
common mechanism. Philosophical Transactions of the Royal Society London, B
Biological Sciences, 358(1431), 517-528.
Grave de Peralta Menéndez, R. et al. (1997): Linear inverse solutions with optimal resolution
kernels applied to the electromagnetic tomography. Human Brain Mapping 5, 454–467.
Hauk O. et al. (2004) Somatotopic Representation Of Action Words In Human Motor And
Premotor Cortex. Neuron 41, 301-307
Hauk, O. and Pulvermüller, F. (2004) Neurophysiological Distinction of Action Words in the
Fronto-Central Cortex. Human Brain Mapping 21, 191–201
Hauk, O. et al. (1999): The minimum norm method as an effective mapping tool for MEG
analysis in T. Yoshimoto, M. Kotani, S. Kuriki, H. Karibe, & N. Nakasato (eds.) Recent
advances in biomagnetism (Proceedings of the 11th conference on biomagnetism). Sendai:
Tohoku University Press, p 213–216
Hebb, D. O. (1949). The organization of behavior. A neuropsychological theory. New York:
John Wiley.
Helenius, P. et al. (1999) Dissociation of normal feature analysis and deficient processing of
letter-strings in dyslexic adults. Cerebral Cortex 9, 476–483
Hickok, G., and Poeppel, D. (2000). Towards a functional neuroanatomy of speech
perception. Trends in Cognitive Sciences 4(4), 131-138.
Hodges, J. R. (2001). Frontotemporal dementia (Pick's disease): clinical features and
assessment. Neurology 56(11 Suppl 4), S6-10.
20
Humphreys, G.W. and Forde, E.M.E. (2001) Hierarchies, similarity, and interactivity in
object recognition: “Category-specific” neuropsychological deficits. Behavioral and Brain
Sciences 24, 453–509.
Ishai, A. et al. (1999). Distributed representation of objects in the human ventral visual
pathway. Proceedings of the National Academy of Sciences U S A 96(16), 9379-9384.
Kellenbach, M. L. et al. (2002). Neural differentiation of lexico-syntactic categories or
semantic features? event-related potential evidence for both. Journal of Cognitive
Neuroscience 14(4), 561-577.
Kiefer, M. (2001) Perceptual and semantic sources of category-specific effects: event-related
potentials during picture and word categorization. Memory & Cognition 29, 100–116.
Kounios, J. and Holcomb, P.J. (1992): Structure and process in semantic memory: Evidence
from event-related brain potentials and reaction times. Journal of Experimental Psychology:
General 121, 459-479.
Martin, A. et al. (1996): Neural correlates of category-specific knowledge. Nature 379, 649–
652.
Martin, A. and Chao, L. L. (2001a). Semantic memory and the brain: structure and processes.
Current Oppinion in Neurobiology 11(2), 194-201.
Martin, A. and Chao, L. L. (2001b). Semantic memory and the brain: structure and processes.
Current Opinion in Neurobiology 11(2), 194-201.
McCandliss, B.D. et al. (1997) Brain plasticity in learning visual words. Cognitive
Psychology 33, 88–110
McCandliss, B.D. et al.(2002) The visual word form area: expertise for reading in the
fusiform gyrus.Trends in the Cognitive Sciences 7, 293-299
Mishkin, M. et al. (1983). Object vision and spatial vision: two cortical pathways. Trends in
Neurosciences 6 414-417.
Neininger, B. and Pulvermüller, F. (2001) The right hemisphere's role in action word
processing: A double case study. Neurocase 7, 303-317
Nobre, A.C. et al. (1994) Word recognition in the human inferior temporal lobe. Nature 372,
260–263
21
Oldfield, R.C. (1971): The assessment and analysis of handedness: The Edinburgh inventory.
Neuropsychologia 9, 97–113.
Perani, D., et al. (1999): Word and picture matching: a PET study of semantic category
effects. Neuropsychologia 37, 293–306
Polk, T.A. and Farah, M.J. (2002) Functional MRI evidence for an abstract, not perceptual,
word-form area. Journal of Experimental Psychology: General 131, 65-72
Preissl, H. et al. (1995). Evoked potentials distinguish nouns from verbs. Neuroscience
Letters 197, 81-83.
Price, C. J. et al. (2001). Dynamic diaschisis: anatomically remote and context-sensitive
human brain lesions. Journal of Cognitive Neuroscience, 13(4), 419-429.
Pulvermüller, F. (2001) Brain reflections of words and their meaning. Trends in Cognitive
Sciences 5, 517-524
Pulvermüller, F. (1996). Hebb's concept of cell assemblies and the psychophysiology of word
processing. Psychophysiology, 33, 317-333.
Pulvermüller, F. (1999). Words in the brain's language. Behavioral and Brain Sciences, 22,
253-336.
Pulvermüller, F. (2002). A brain perspective on language mechanisms: from discrete neuronal
ensembles to serial order. Progress in Neurobiology, 67, 85-111.
Pulvermüller, F. (2003). The neuroscience of language. Cambridge: Cambridge University
Press.
Pulvermüller, F. et al. (1996). Brain rhythms of language: nouns versus verbs. European
Journal of Neuroscience, 8, 937-941.
Pulvermüller, F. et al. (2001) Neuromagnetic evidence for early semantic access in word
recognition, European Journal of Neuroscience 13, 201-205
Pulvermüller, F., & Preissl, H. (1991). A cell assembly model of language. Network:
Computation in Neural Systems, 2, 455-468.
Pulvermüller, F. et al. (2000). Neurophysiological distinction of verb categories. Neuroreport,
11(12), 2789-2793.
22
Rizzolatti, G. et al. (2002). Motor and cognitive functions of the ventral premotor cortex.
Curr Opin Neurobiol, 12(2), 149-154.
Scott, S. K. and Johnsrude, I. S. (2003). The neuroanatomical and functional organization of
speech perception. Trends in Neurosciences, 26(2), 100-107.
Shtyrov, Y. et al. (2004). Distributed neuronal networks for encoding category-specific
semantic information: the mismatch negativity to action words. European Journal of
Neuroscience, 19(4), 1083-1092.
Skrandies, W. (1998): Evoked potential correlates of semantic meaning: a brain mapping
study. Cognitive Brain Research 6, 173–183.
Tarkiainen, A. et al. (2002) Dynamics of visual feature analysis and object-level processing in
face versus letter-string perception. Brain 125, 1125–1136
Tsumoto, T. (1992). Long-term potentiation and long-term depression in the neocortex.
Progress in Neurobiology, 39, 209-228.
Warrington, E.K. and McCarthy, R.A. (1987): Categories of knowledge: further
fractionations and an attempted integration. Brain 110, 1273–1296
Warrington, E.K. and Shallice, T. (1980) Word-form dyslexia. Brain 103, 99–112
Wermter, S. et al. (2004). Towards multimodal neural network robot learning. Robotics and
Autonomous Systems, in press.
Wernicke, C. (1874). Der aphasische Symptomencomplex. Eine psychologische Studie auf
anatomischer Basis. Breslau: Kohn und Weigert.
Zeki, S. and Marini, L. (1998). Three cortical stages of colour processing in the human brain.
Brain 121, 1669-1685.
23