Professional Documents
Culture Documents
2011 - Lachaud - EEG, EYE and KEY - Three Simultaneous Streams of Data For Investigating The Cognitive Mechanisms of Translation
2011 - Lachaud - EEG, EYE and KEY - Three Simultaneous Streams of Data For Investigating The Cognitive Mechanisms of Translation
net/publication/264382314
EEG, EYE and KEY: Three Simultaneous Streams of Data for Investigating the
Cognitive Mechanisms of Translation
CITATIONS READS
15 666
1 author:
SEE PROFILE
All content following this page was uploaded by Christian Michel Lachaud on 12 February 2016.
Introduction
The psycholinguistic study presented in this article was realized for the European
Research Project “EYE-to-IT”,1 in collaboration with the National Centre for Epilepsy,
eye tracking (EYE) and keystroke logging (KEY), EYE-to-IT aimed to develop
technical solutions for studying the cognitive mechanisms of translation and for
his task. The prompting technique, already in use for online documentation and
feedback systems (Cavero et al., 1995; Ferreira & Atkinson, 2009), could be adapted
for priming transcoding and translation (Duyck & Warlop, 2009; Finkbeiner et al.,
not restrict the translator’s freedom, fluidity and creativity by locking a given
formulation in his mind, prompting should not consist in delivering a list of possibilities
1
from which the translator would choose the one he prefers. Rather than providing a
translators’ mental processes, supporting them when they fail or are being
challenged. Prompting can only achieve this goal by offering a fuzzy, even subliminal
influence, using prompts different enough from the targeted solution to avoid focusing
solution is to prompt formal and semantic representations via formal and semantic
instance, the Norwegian word “mannekeng” (fashion model) can be used to prompt
the Norwegian transcoding of the English source word “male”, the target word
“mannlig”, because both prompt and target share sublexical features (i.e. the first four
a specific representative for the category “male”. With such inputs, the translator
would be left dealing with his own mental processes, the dynamics of which are now
subtly influenced.
systematic approach intended for future technical implementation requires that the
starting at the word level before using more complex linguistic stimuli in experimental
settings, like phrases, sentences and texts. The study presented in this article
involved the combining of EEG, EYE and KEY for investigating cognitive loading
below for an explanation of each of these). Our objective was to be able to provide
helpful prompts to translators, based on what we could find about cognitive load in
2
transcoding challenge existing in each situation. We will then discuss the advantages
of each measurement technique (EEG, EYE, KEY), as well as the advantages and
disadvantages in combining them. Finally, we will describe the method, present our
challenges and for starting to explore the psychological nature of these challenges. It
was the first in a series of four experiments which were designed for (i) identifying the
nature of these challenges, (iii) defining the type of information that could helpfully
prompt the translator in order to cope with transcoding challenges, and (iv) evaluating
false friends (DC), True Cognates (TC) and Non-cognates (NC) (Chamizo
(i) A DC relation exists between two words from two languages when they share
the same form but not the same meaning, like “gift”, meaning “present” in
(ii) A TC relation exists between two words from two languages when they share
both the exact same form (contrary to cognates whose form may differ slightly
in a language pair) and the same meaning, like “egg” for the
English/Norwegian pair.
(iii) A NC relation exists between two words from two languages when they share
the same meaning but not the same form, like “ant” in English and “maur” in
Norwegian.
3
Figure 1 shows the logical structure of L2-L1 correspondences at the lexical
and semantic levels for DC, TC and NC, based on the architecture of Conceptual
Mediation Hierarchical Models (French & Jacquet, 2004; Frenck & Pynte, 1987;
2 1 2/1 2/1
Semantic
Lexical
C2 S2/1 C1 S/C2/1 S2 C1
L2 L1 L2 L1 L2 L1
Figure 1. Logical structure of lexical and semantic L2-L1 relations for Deceptive
Cognates, True Cognates and Non-cognates. S stands for “Stimulus” (English word
tongue (Norwegian), into which he must transcode. The numbers 1 and 2 refer
4
transcoding procedure. In theory, transcoding an English NC Stimulus S 2 like “ant”
into its Norwegian correspondence C1 “maur” will require four main steps: (i)
recognizing the stimulus (retrieving and activating in long-term memory the mental
1), (ii) accessing the meaning (activating in long-term memory the mental
marked 2/1 in Figure 1), (iii) retrieving the correspondence in the target language
Norwegian, marked C1 in Figure 1), and (iv) producing the solution (speaking, typing,
writing). In the case of TCs, because both the formal and the semantic
representations are identical in L1 and L2, no transcoding will happen. Therefore, the
processing of a TC stimulus will be limited to three steps: (i) recognizing the stimulus,
(ii) accessing its meaning, and (iii) producing the solution. Transcoding English DC
input S2 (like “gift”) into its Norwegian correspondence C1 (“gave”) will require
(Group, 1986), whose principles are well formulated in TRACE (McClelland & Elman,
1986), a psycholinguistics model of word recognition, the stimulus will not only
activate meaning 2 (“present”) in the English lexicon, it will also activate meaning 1
(“poison”) in the Norwegian lexicon (see also (Bölte & Coenen, 2002). Activation will
automatically propagate in the neural network, spreading back from the semantic
activated, one in Norwegian from the English meaning (C1 “gave”), and one in
English from the Norwegian meaning (C2 “poison”). The consequence will be a
5
conflict between the two lexicons during selection, both at the semantic level between
meanings 1 and 2 and at the lexical level between representations C1 and C2. Facing
a confusing situation, the translator may have four types of reactions: (i) He might
automatically produce the correct Norwegian formulation (“gave”) for the English
meaning (“gift”), although probably with greater difficulty than for a NC; (ii) Confused,
mind than the English language. He may then realize his mistake and try to correct it,
in which case two scenarios may occur (iii and iv). A successful correction would
require him (iii) to go back from C2 (“poison”) to meaning 1 (“poison”), to S2/1 (“gift”),
successful, the time needed to produce the correct Norwegian transcoding (“gave”) of
the English stimulus (“gift”) will be further increased and may have other behavioural
manifestations, like typing corrections. But the translator may also be unable to fully
correct his mistake, especially under time pressure, (iv) if he only reverses his
transcoding, and does not transcode again. In this case, he will simply be back to the
experiments.
such as electric potentials (EEG) and behaviour (eye movements and keystrokes).
The EEG technique measures, usually on the surface of the scalp, the electric
2005), including that of cognitive activity, at the moment they occur in the cerebral
6
cortex (“online” measurement). EEG can be used to track cognitive load variations as
an index of processing difficulty, to locate where in the brain the activity occurs for
behaviour which provides indices about visual information intake. Eye movements
can reveal the progression of syntactic processing and meaning building in the
reader’s mind, and in the cases where information intake is necessary for the
information intake, eye-tracking may also reveal information about the cognitive
recording typing behaviour on a keyboard, can be used for studying the mental
processes during translation by studying how the text was produced (also considered
measurement), in which case keystroke logging only serves as a means to record the
Despite the fact that each technique imposes its own constraints on the
experimental situation and is generally seen as exclusive of the two others, EYE-to-IT
engaged researchers in the challenge of simultaneously using EEG, EYE and KEY
prepare the participants adequately. Furthermore, the risk of producing bad quality
7
The main technical difficulty in this study was presented by EEG and EYE.
EEG requires situations which do not promote eye movements, blinks, motor activity,
and any cerebral activity other than the one being studied. Therefore, reading and
typing while translating represent sources of artefacts that one would like to avoid,
making it meaningless not only to use EYE and KEY together with EEG, but to use
allows us to extract useful information from EEG with artefacts, opening its
the subject regularly gazes beyond the screen (typing, consulting a dictionary) is a
eye movements, between actual reading activity and the search behaviour that
occurred when the translator transferred attention back to the source text. However,
eye-tracking proves to be at least a very useful tool for the online detection of the
difficulties that the translator is facing (longer fixation time or larger number of
fixations). Subsequently, eye-tracking provides inputs for analyzing the nature of the
linguistic challenges and knowing when to deliver a prompt and where on the screen.
complementary way with other techniques. This requires clever designs and proper
In this study, we investigated (i) the amplitude variations through time in the
load variations, (ii) the verification effort during the mental matching between L 2 and
L1, which manifests in the total amount of time spent gazing at both words (Total
8
Fixation Duration), (iii) the attention demand required for processing the stimulus, the
variations of which are manifested in pupil diameter through the pupillary reflex (Pupil
maximum sizes), (iv) the decision difficulty, which is manifested in the amount of time
required before pressing the answer key (Reaction Time), and (v) the confusion of
cognitive load, verification effort, attention demand, difficulty and confusion), while
Method
Independent variables
Two independent variables were manipulated in the experiment:
L1 do not correspond).
The two variables were crossed according to a factorial design, generating six
experimental conditions.
Dependent variables
The dependent variables, measured while both the English and the Norwegian words
9
(i) Time-Frequency Power Difference (EEG) between the Match and
Operational Hypothesis
EEG: Time-Frequency Power Difference will be positive with DCs (greater cognitive
load for processing the Match condition than the Mismatch condition), negative with
TCs (smaller cognitive load for processing the Match condition than the Mismatch
EYE: decreasing gradient of Total Fixation Duration and Pupil Size Amplitude
Experimental materials
120 English words and 240 Norwegian words were used. The English word list
consisted of 40 English/Norwegian DCs like “art” (“art” means “sort, kind, or species”
like “ant”. 120 of the 240 Norwegian words were the Norwegian correspondence of
the English words (respectively “kunst”, “arm” and “maur”), the other 120 being
unrelated. 240 items were created by associating each one of the 120 English words
once with its Norwegian correspondence (120 items) and once with one of the 120
unrelated Norwegian (120 items). The stimulus words were all nouns. In order to
10
avoid ambiguous situations, NC words were excluded if they could be a cognate with
etc.). Spelling in English and Norwegian were strictly identical for TCs and DCs. As
shown in Table 1, the average length of the English words was approximately
frequency index of word occurrence in written language (Wilson, 1988). Other indices
represented with a drawing (Juhasz, 2005). Although these ratings were not available
for all the English words we had selected, we report in Table 1 the information
available in the MRC psycholinguistic database4 (Wilson, 1988) for a random subset
hence that these parameters could not cause the variations observed in the results.
the above psychological dimensions were impossible to control for the Norwegian
words. However, their formal homogeneity (length and frequency) could be computed
(Blair, Urland & Ma, 2002) and was high between conditions (Table 1). All materials
11
Table 1
N Lex
AoA Fam Concr Imag
Letters Frq
10-
383
3-8 1790 424-620 311-614 384-634
150-
Min-Max 3-7 1-1430 441-644 482-635 412-630
406
3-7 20- 401-626 323-623 397-632
166-
1323
511
4.33/1. 324/38
01 3
282/52 552/52 500/92 537/67
4.75/0. 385/38
Mean/SD 268/67 548/53 589/32 574/43
99 0
304/64 542/48 519/83 543/60
5.03/1. 347/31
15 3
40/40/4 40/40/4
N words
0 0
Norwegian stimulus word
0.004-
6.3
3-10
0.032-
Min-Max 3-7
8.4
3-9
0.009-
NA
6.3
0.87/1.
5
5/1.5
1.23/1.
Mean/SD 4.75/1
8
5.1/1.6
0.77/1.
3
Note. N Letters: The amount of letters composing a word, or word length. Lex Frq:
Lexical frequency. For the English words, Thorndike-Lorge written frequency index
provided by the MRC Psycholinguistics Database. For the Norwegian words, index
computed by the author with the Google search engine (amount of pages written in
Norwegian, located in Norway and containing the target Norwegian word at least
Concr: Concreteness. Imag: Imageability. AoA, Fam, Concr and Imag are ratings
12
between 100 and 700, corresponding to a scale of 1 to 7. N words: Amount of words
per category. Values in each cell are given in this order: DC, TC and NC.
Experimental plan
Four lists, each containing one fourth of the 240 items, were created such that in
each list, half of the items were transcoding correspondences, while the remaining
half were not. Each participant was tested with two complementary lists, in order to
provide data on each of the 120 English words. Additionally, a training set containing
Stimulation procedure
Items were selected randomly from each list and presented automatically, delivered
visually at 800 by 600 screen resolution and normal pixel density (96 DPI) on a high
quality 21” cathode ray tube screen (DELL P1130 Trinitron). The screen was
refreshed at 100 Hz, stimulus delivery being synchronized with screen refreshment
for improving timing precision of the stimulating procedure. The distance between the
surface of the screen and the subject’s eyes was 55 cm. Words were written with the
fixed-width font Courier New, in size 24 (9 mm height per letter, i.e. 0.94° visual
angle). The two words composing a stimulus were separated by an empty space
which varied between 9 and 18 characters depending on the length of the words.
This visual layout allowed us to keep the total item length constant (24 characters,
13
Figure 2. Items’ visual layout (character boxes were not visible in the stimulus)
Furthermore, each item’s length was made large enough to make it impossible
to read a word while gazing at the other one, whatever the word length. Therefore,
participants were obliged to make a saccade (a fast movement of the eyes) to read
each word, an essential constraint for analyzing the data: saccadic movements
produce a specific artefact in the EEG which could be used for clear time
The stimulation sequence started with the English word (WL 2 - see Figure 3)
appearing randomly on the left or right half of the screen for 400 ms. Displaying WL 2
for this short duration served as a signal attracting the gaze of the subject to ensure
that the English word was always read first. After 400 ms, the Norwegian word (WL 1)
was displayed on the other half of the screen, while the English word remained on
the screen at its position. The two words were then displayed together until the
response key was pressed or for a maximum duration of two seconds. Two additional
seconds separated the end of an item’s presentation from the beginning of the next
(termed “inter-stimulus interval”). Every 20 items, a resting time was offered to the
14
Figure 3. Summary of the experimental situation during the stimulation sequence on
one item
Task
Participants performed a “go, no-go” transcoding validation task, consisting of
checking if the English and the Norwegian words (DCs, TCs, NCs) presented in the
box if the Norwegian word was the correct transcoding of the English word or doing
nothing otherwise. Placed under performance pressure both for speed and accuracy,
subjects had to provide a decision as fast as possible and were encouraged to avoid
mistakes (see the Experimental Session section below for more details about the
experimental protocol).
Recording
Three computers were used. The first computer, piloted with a script written for E-
Prime (2008b),5 presented the stimulus and measured participant’s decisions (i.e.
15
key pressed and reaction time measured from the appearance of the Norwegian
word - see Figure 3). The second computer recorded the EEG continuously at 512
Hz with the ASA software (2008a), through a 64 channels shielded cap.6 The third
computer recorded eye-movements from the appearance of the English word and
until the subject had given his answer or until 2 seconds had elapsed. The eye-
tracker (EyeLink 1000, SR Research Ltd., Canada, used in remote mode with a 500
Hz monocular sampling rate) was piloted directly from the first computer by additional
Parallel port and Etherlink), EEG and EYE data being time-stamped with triggers
generated by E-Prime for the various stimulation events and for the participant’s
Population
We were advised by the European Union Officers following EYE-to-IT to use
bilinguals only at this early stage of research development. Translators are a rare and
expensive population, whose use was not scientifically justified for this study. Indeed,
any bilingual could transcode isolated words as well as any expert translator. 52
participants were tested (52 per cent female), aged 27 on average (min 18, max 65,
SD 9). Norwegian native speakers who were also fluent bilinguals in English were
English was 6.5 years, min. 0 for real bilinguals, max. 10, SD 3.3). Those with an
adequate profile were given the online language assessment test “Dialang” (Freie
16
Universität Berlin et al., 2003), evaluating their proficiency in English. Participants
with the highest reading level and a vocabulary level score above 900 out of 1,000
were selected. After signing a consent form, candidates were included in the pool of
neurological disorders. Those with corrected vision were accepted provided that they
wore glasses or contact lenses during the experiment. Participants were paid 500
Norwegian Kroner.
Experimental session
Experimental sessions were individual and lasted 2 hours, including the testing on all
experiments. The participant was prepared (installation and plugging in of the electro-
cap), instructed on how to behave with the equipment (be as still as possible, blink
breaks to relax the eyes and to move if necessary), and instructed about the task to
perform (read the stimulus, decide if the Norwegian word is the correct translation of
the English word, and press the button of the button box if the answer was yes, do
nothing otherwise. Perform the task as fast as possible and, at the same time, avoid
making mistakes). Instruments were then adjusted and tested (lowest possible
electric resistance for all electrodes, calibration of the eye-tracker). Before starting to
train on a set of training items, the subject was asked to summarize what he had
experimenter. The subject was then trained until his performance reached a minimum
threshold of average speed (1200 ms) and accuracy (80 per cent). The testing part
which followed was extremely brief (five minutes long) because of its high intensity.
The experiment was run along with two other experiments in the same session, and
17
took place at the ERP Laboratory of the National Centre for Epilepsy, Rikshospitalet
Data processing
EEG: Time-Frequency analysis (Harmony et al., 2009; Hoechstetter et al., 2004;
Koenig, Hubl & Mueller, 2002) used the data from 44 participants only, the EEG
quality from eight participants being poor. Signal processing and analysis was done
with the software BESA (2009). Time-Frequency analysis considered EEG segments
of 575 ms (-100 to 475) around the first horizontal eye movement after the
and L1). Longer segments could not be used (appearance of the first answers). For
each Stimulus Type condition, the analysis compared power differences existing in
the EEG between the Matching condition and its corresponding Mismatching
interval between -100 and +475 ms. Time-frequency matrices of p-values7 were
matrix per channel, repeated for each one of the three Stimulus Type conditions. The
cranial locations was analysed with multilevel modelling (Clark, 1973; Hox, 2002;
EYE: Eye data from all subjects were used. Raw measurements on matching
items were isolated from those on mismatching items. Items with blinks or no eye
detection were eliminated. The overall quality, however, was very good, because
subjects were instructed to blink systematically between items and to pay great
attention to the stimulus, which naturally reduced their blinking rate during item
18
Tampere, Finland), analysed the raw data and extracted Total Fixation Duration and
Pupil Size Amplitude Variation. Outliers were filtered with Median ± 3 Median
Absolute Deviation applied by subject and by item in each condition. Missing values
were not replaced. Three linear multilevel models, one per dependent variable,
crossed subjects and items as random variables and included Stimulus Type as a
fixed variable.
KEY: Data from only 41 subjects were used, 11 subjects having performed the
task incorrectly during part of the experiment. The procedure described above for eye
movements was followed for filtering outliers and statistical analysis. Because Error
Probability was binary, a logit binomial multilevel model was used. The model’s
For EEG, EYE and KEY, the test statistic was a large sample Chi squared test.
Results are graphed in Figures 4 to 8 with 95 per cent confidence interval bars.
19
Results
EEG
Stimulus Type
20
Time-Frequency Power Difference was smaller by 0.311 Time-Frequency units
with TCs than with DCs (2(1, N = 2736) = 36.8, p < .0001), and smaller by 0.238 Time-
Frequency units than with NCs (2(1, N = 2736) = 21.5, p < .0001). The difference
EYE
21
Total Fixation Duration with DCs was 368 ms longer than with TCs (2(1, N = 2827)
= 50.2, p < .0001), and 289 ms longer than with NCs (2(1, N = 2827) = 30.9, p < .0001).
The difference between TCs and NCs was marginal (p < .13).
Figure 6. Pupil Size Amplitude Variation (in arbitrary diameter unit) as a function of
Stimulus Type
22
Pupil Size Amplitude Variation with DCs was 5.3 diameter units bigger than
with TCs (2(1, N = 2892) = 10.5, p < .0013) and 3.3 diameter units bigger than with NCs
(2(1, N = 2892) = 4, p < .047). The difference between TCs and NCs was not significant
(p < .22).
KEY
Reaction times with DCs were 189 ms longer than with TCs (2(1, N = 2186) = 99,
p < .0001), and 127 ms longer than with NCs (2(1, N = 2186) = 44, p < .0001).
Additionally, reaction times were 62 ms longer with NCs than with TCs (2(1, N = 2186) =
23
Figure 8. Error Probability (in per cent) as a function of Stimulus Type
Error Probability with DCs was 15 per cent bigger than with TCs (2(1, N = 2460) =
24, p < .0001), and 14 per cent bigger than with NCs (2(1, N = 2460) = 20, p < .0001).
The difference between TCs and NCs was not significant (p < .66).
24
Discussion
Result patterns across dependent variables give a clear, indisputable picture of the
Difference, longest Total Fixation Duration, biggest Pupil Size Amplitude Variation,
longest Reaction Time and biggest Error Probability). The fact that TCs were easier
shortest Total Fixation Duration, smallest Pupil Size Amplitude Variation, shortest
Reaction Time and smallest Error Probability). Although an intermediate pattern was
predicted for NC, analyses do not clearly confirm the hypothesis. Yes, the dependent
variable’s average for NCs is always between DCs and TCs, but the difference
between the three conditions is only significant for reaction times. With other
dependent variables, NCs are either significantly different from TCs and not from
DCs (Time-Frequency Power Difference), or from DCs and not from TCs (Total
If the study had been based on one dependent variable only, the chances are
high that an incorrect conclusion would have been given, e.g.: “NCs are as difficult to
transcode as DCs”, or “NCs are as easy to process as TCs”. It is, however, likely that
NCs are more difficult to process than TCs, but easier to transcode than DCs.
Because each dependent variable measures a different aspect of the mind’s activity,
it is not surprising that varying patterns of results are obtained, hence the
This study reveals that the processing difficulty (Reaction Time) is the highest
for DCs, the lowest for TCs, and medium for NCs. The verification effort (Total
Fixation Duration) and reading attention demand (Pupil Size Amplitude Variation) is
the highest for DCs and the lowest for TCs. The verification effort is marginally bigger
25
with NCs than with TCs, but attention demand is not statistically different. Participants
were the most confused with DCs and less but equally confused with TCs and NCs.
The overall cognitive load is strongly reduced between the TC Mismatching and
Matching conditions, and slightly reduced with NCs, indicating that less computation
mismatching conditions. The overall cognitive load is slightly increased with DCs,
indicating that more computation was needed to make a decision in the matching
supported by this set of results. Because the lexical form is identical between L 2 and
L1 for TCs, verification effort, attention demand, difficulty and confusion are the
lowest and cognitive load is reduced in the matching condition compared to the
mismatching condition. With NCs, the meaning of the two words is identical and
serves as a link between L2 and L1 lexical forms. Thanks to this semantic link, the
overall cognitive load is slightly reduced in the matching condition compared with the
mismatching condition. Verification effort is greater with NCs than with TCs because
two lexical forms are now being compared and mentally transcoded. Consequently,
attention demand, confusion level and the decision-making load are also higher. DC
words have the same lexical form in L2 and L1, but a different meaning in L2 and L1.
DCs therefore induce the highest confusion and difficulty, as well as hesitation in the
decision making process. Thinking about possible meanings, bilinguals will eventually
follow one of the four paths described in the introduction. Due to the additional
computation that was not necessary in the NC condition, it is logical to find an overall
condition.
26
Conclusion
This study has demonstrated the possibility and the added value of simultaneously
using multiple techniques to measure the cognitive load involved in transcoding TCs,
NCs and DCs. The methodology has allowed us to confirm the existence of a transfer
difficulty hierarchy that translators may face at the word level: DCs are the most
difficult to process, NCs represent an intermediate cognitive load, and TCs are the
easiest. The results allow us to pave the way for a Translation Priming system that
might help translators retrieve forms and meanings from long-term memory. For
example, when processing DCs translators might benefit from prompts related both
semantically and lexically to the target. On the other hand, the processing of NCs
might only require prompts at the lexical level. Finally, when processing TCs
translators might not benefit from prompting as they would not be facing any
transcoding challenge. This information was essential for grounding the next steps of
27
1
Development of Human-Computer Monitoring and Feedback Systems for the Purposes of Studying
th
Cognition and Translation, 2006/01-2009/04, 6 Framework Program, Information Society
Technologies (Contract 517590).
2
According to this type of psychological model, two languages share the same semantic system of
representation in the bilingual mind, but each language is stored as a distinct system of formal
representations. Both languages are therefore linked together through the common semantic system -
Illes, J., Francis, W. S., Desmond, J. E., Gabrieli, J. D. E., Glover, G. H., Poldrack, R., Lee, C. J. &
Wagner, A. D. (1999). Convergent Cortical Representation of Semantic Processing in Bilinguals. Brain
and Language, 70(3), 347-363.
3
Pupil size provided by the EyeLink 1000 is given as integer numbers in an arbitrary unit, with a noise
level of 0.2 per cent of the diameter.
4
The MRC Psycholinguistic database: Cf. http://www.psych.rl.ac.uk/MRC_Psych_Db.html (Last
accessed 06/01/10).
5
E-Prime is software specifically developed for psychological research. It allows the presentation of
visual, auditory, or audio-visual stimuli to human subjects, and records their behavior in automated
experimental sessions.
6 TM
WaveGuard cap (ANT) with a 64 channels 10-20 international layout, using 64 sintered Ag/AgCl
electrodes, Hirose HD connectors.
7
Statistical index giving the probability that the null hypothesis is true. The scientific standard
threshold is p < .05. It means that the probability to observe a given pattern in the data due to chance
is below 5 per cent.
References
GmbH.
Blair, I. V., Urland, G. R. & Ma, J. E. (2002), 'Using Internet search engines to
Bölte, J. & Coenen, E. (2002), 'Is Phonological Information Mapped onto Semantic
Cavero, L., Concejero, P., Gili, J., Yuichiro Anzai, K. O. & Hirohiko, M. (1995), 'Help
28
Chamizo Domínguez, P. J. & Nerlich, B. (2002), 'False friends: their origin and
1849.
Duyck, W. & Warlop, N. (2009), 'Translation Priming Between the Native Language
501.
Finkbeiner, M., Forster, K., Nicol, J. & Nakamura, K. (2004), 'The role of polysemy in
51 (1), 1-22.
Freie Universität Berlin, G., Jyväskylän yliopisto, F., Lancaster University, U., CITO
29
Frenck, C. & Pynte, J. (1987), 'Semantic representation and surface forms: A look at
16 (4), 383-396.
71 (1), 9-16.
Hoechstetter, K., Bornfleth, H., Weckesser, D., Ille, N., Berg, P. & Scherg, M. (2004),
Erlbaum Associates.
Illes, J., Francis, W. S., Desmond, J. E., Gabrieli, J. D. E., Glover, G. H., Poldrack,
Koenig, T., Hubl, D. & Mueller, T. J. (2002), 'Decomposing the EEG in time, space
30
Performance and Cognitive Load Variations after Prompting'. International
Lachaud, C. M. & Renaud, O. (2009), 'A Tutorial for Analyzing Human Reaction
Rasbash, J., Charlton, C., Browne, W. J., Healy, M. & Cameron, B. (2009), MLwiN.
31