Effect of Typicality in Abstract Word Categories on N400 ERP

Mansoureh Fahimi

and Marc Van Hulle

Laboratory of Neuro- and Psychophysiology, KU Leuven University, Leuven, Belgium

1 OBJECTIVES

We investigate the effect of word typicality −the

degree of membership of a word to its superordinate

category− on the N400 event-related potential (ERP)

using a single-trial detection approach based on

spatiotemporal beamforming. Unlike the norm in

N400 studies, where mostly concrete categories are

used (imaginable objects), we considered a total of 6

basic categories: three abstract and unimaginable

(emotion, event, illness), one abstract yet clearly

imaginable (colour), and two concrete categories,

one coherent (mammals) and one incoherent

(furniture). Our results show that, independently of

word abstractness or concreteness, word typicality

has a clear effect on N400 both in terms of

amplitude and scalp localization, which in turn is

indicative of differences in difficulty of word

processing.

2 METHODS

Stimuli were developed to belong to 6 basic

categories. For each of the 6 categories, 15 word

stimuli were chosen, and a subsequent group of

about the same size chosen as “fillers” from random

categories of mainly concrete words (out of category

word-pairs). The typical and atypical category

members, and also the filler words (non-category

members), were matched for word length,

orthographic neighborhood size, and frequency of

occurrence, using the Dutch CLEARPOND software

(Marian et al., 2012).

Table 1: means (m) and standard deviation (std) of

considered word properties.

In-category Non-category

Word length m=6.7, std=2.2 m=6.7, std=1.5

Orth.Neighb. size m=2.6, std=3.6 m=1.7, std=2.2

Freq.of occurrence m=19.2, std=31.7 m=12.1, std=17.2

We recruited 17 volunteers to score, on a scale of

1-5, 90 words based on how typical they thought an

exemplar of each category was.

Table 2: Example words for each category.

typical atypical nonmember

gebeurtenis

feest kindertijd basketbal

(event)

(party) (childhood) (basketball)

kleur

blauw amber acteur

(color)

(blue) (amber) (actor)

meubel

stoel kapstok galerij

(furniture)

(chair) (coat rack) (gallery)

ziekte

epilepsie verslaving ooievaar

(illness)

(epilepsy) (addiction) (stork)

zoogdier

olifant vleermuis vuilnis

(mammal)

(elephant) (bat) (garbage)

emotie

droefheid verwarring vliegtuig

(emotion)

(sadness) (confusion) (plane)

The experimental paradigm was a simple word-

pair experiment (semantic priming). The prime word

was always chosen to be the label of the

superordinate category (i.e., the name of the

category). The target is randomly chosen to be either

a non-member (“filler”) or one of the 15 words

chosen as member of that category.

EEG data was recorded using 64 active Ag/AgCl

electrodes (SynampsRT, Compumedics, France),

according to the international 10-20 system. Two of

these electrodes served as ground (AFz) and

reference (FCz). The EEG signal was recorded at a 2

KHz sampling rate and downsampled to 250 Hz.

We recruited for our pilot study 12 subjects (5

males, two left-handed, average age was 21.6, std=

1.9). Ethical approval for this study was granted by

an independent ethical committee (“Commissie voor

Medische Ethiek” of UZ Leuven, our University

Hospital). This study was conducted in accordance

with the most recent version of the Declaration of

Helsinki.

Data Analysis

The EEG data was re-referenced offline from the

original ground and reference to a common average

reference (CAR), and filtered using a 4

order

Butterworth filter in the range of 0.1-30 Hz. A

second filtering in the range of 0.1 to 30 Hz was

applied to further demote possible remaining

artefacts. The data was epoched using windows

starting from 100ms prior to the presentation of the

stimulus of interest (target) until 1000ms post-onset.

The baseline was removed using the average signal

Fahimi, M. and Hulle, M.

Effect of Typicality in Abstract Word Categories on N400 ERP.

In Extended Abstracts (NEUROTECHNIX 2016), pages 6-8

in a 100ms interval prior to stimulus onset. Trials in

which the signal exceeded ±150µV where excluded

from the analysis. Trials in which mouse button

press responses were incorrect for non-member and

typical member targets were also omitted, yet

incorrect button press responses for atypical words

were not omitted.

LCMV Beamformer for Single-trial ERP

Detection

The linearly constrained minimum variance (LCMV)

beamformer (van Vliet et. Al 2016) is a

spatiotemporal filter that relies on spatial- and

temporal templates of the ERP collected during a

training session (using a proportion of the dataset for

training). These templates are formed by subtracting

the average EEG recordings of two experimental

paradigms both in time (between 350 and 500ms after

stimulus onset) and space (electrodes). As our experi-

ment involved three possible outcomes (typical, atypi-

cal, and nonmember), we used trials of nonmember

and typical targets to maximize the N400 effect. This

template is optimized two satisfy two criteria: a)

maximal correlation with the actual amplitude of our

component of interest (here N400) and b) minimal

correlation with interfering signals, such as noise or

other ERP components. The template is then applied

to each epoch separately (single trial) and the

beamformer returns a single value, which indicates

the presence of an N400 response for that epoch.

Statistical Analysis

Since we have unbalanced data, a linear mixed effect

model was used with N400 response (the output

resulting of the beamformer, cf. supra) as an indepen-

dent variable, and with the following fixed effects for

several analyses: relatedness (whether or not our

target was a member of the category, irrelevant of

typicality), typicality (labels of the targets divided into

typical, atypical, and nonmembers), and concreteness

(labels of the targets divided based on whether they

are members of the concrete or the abstract category).

Random effects were targets, primes and subjects.

Repeated measures analysis of variance (ANOVA)

was performed on the outcomes of the linear mixed

effect model. A significance level of 5% was adopted

for all analyses.

3 RESULTS

Beamformer Results

Out of the 64 recording channels, we selected a total

of 31 channels. Given that only a percentage of the

data should be used for training our beamformer

template, we needed to choose a proportion of the

data, where randomly chosen trials would still show

consistent beamformer templates across replications.

When we used 60 percent of the data (which is, 30%

of the typical and unrelated trials respectively), we

achieved an overall stability in both the spatial and

temporal templates. Note that we do not use trials

with atypical targets to develop the beamformer,

because they are expected to be in between the two

extreme cases of typical and unrelated, but also

because atypical trials in general were less

prominent than typical and unrelated ones. To assure

statistical stability across replications, we formed the

beamformer in 100 iterations and analyzed the mean

and variance for the spatial and temporal templates.

An example of both templates is shown in fig. 1.

Figure 1: Spatial (left) and temporal (right) beamformer

templates.

The first hypothesis we tested was on the general

relatedness (target versus nontarget). A one-way

ANOVA of general relatedness (including typical

and atypical members) against unrelated members

revealed a significant difference (p=0.00175,

F=5.6743). When looking for effects of typicality

versus atypicality versus nonmember, a significant

difference of (p=0.0008435, F=4.8172) was found,

both when all groups were included, but also when

the group ‘colour’ was excluded from the analysis

(p=0.004236, F=3.1828). Further pairwise

comparison of the groups revealed a significant

difference between typical versus atypical exemplars

of the categories (p= 0.002725, F=3.7217). An

ANOVA analysis of the effect of concreteness

versus abstractness on the N400 amplitude was also

significant (p=0.002589, F=3.0919). Note that this

result also holds when we eliminate the group

‘colour’ from the analysis (p=0.0045245, F=2.5084),

which shows that our results apply to both cases of

using only abstract unimaginable groups, and when

the abstract category includes both imaginable and

unimaginable words.

Effect of Typicality in Abstract Word Categories on N400 ERP

ERP Analysis

The ERPs of four centrally located (Cz, CPz, CP1,

CP2) electrodes are plotted in fig. 2. They show a

clear distinction between the two different groups of

categories, contrary to the lateralized electrodes

where the abstract and concrete categories resulted

in slightly distorted ERPs.

Figure 2: ERP plots of subjects for a) both categories,

b) only abstract categories and c) only concrete categories.

4 DISCUSSION

In this study we have investigated the effect of an

atypical member of a category on the N400 ERP for

both abstract and concrete word categories. We

observed a bigger N400 effect for trials with less

clear examples of a given category, in accordance

with the prototype hypothesis (Rosch, 1975).

However, this hypothesis was previously tested

mostly for categories of imaginable, concrete words,

such as vegetables, flowers, or birds (Fujihara et. Al,

1998). Categories of abstract concepts were given

less attention. Here, we evaluated the prototype

hypothesis using three abstract categories (illness,

event, and emotion) and one abstract- but still

imaginable category. The typicality effect was

observed in N400 ERP in both the abstract and the

concrete categories. The latter has been shown in

previous studies (Fujihara et. Al, 1998), but not for

the abstract categories. Also it has been shown that

the N400 effect is generally larger for concrete

versus abstract word-pairs (Kounios and Holcomb,

1994, Tolentino and Tokowicz, 2009). In our study

we also found a significant difference between the

two groups, both when the comparison was only

between abstract, unimaginable words versus

concrete ones, and when the abstract category

included imaginable words.

ACKNOWLEDGEMENTS

Mansoureh Fahimi is supported by Hermes Fund, National

Fund for Scientific Research Flanders (SB/151022).

MMVH is supported by PFV/10/008, IDO/12/007,

IOF/HB/12/021, G088314N, G0A0914N, IUAP P7/11,

AKUL 043.

REFERENCES

Fujihara, N., Nageishi, Y., Koyama, S., & Nakajima, Y.,

1998. Electrophysiological evidence for the typicality

effect of human cognitive categorization. Int. J.

Psychophys., 29(1), 65-75.

Kounios, J., & Holcomb, P. J., 1994. Concreteness effects

in semantic processing: ERP evidence supporting

dual-coding theory. J. Exp. Psychol.: Learning,

Memory, and Cognition, 20(4), 804.

Marian, V., Bartolotti, J., Chabal, S., & Shook, A., 2012

CLEARPOND: Cross-linguistic easy-access resource

for phonological and orthographic neighborhood

densities. PloS one, 7(8), e43230.

Rosch, E., 1975. Cognitive representations of semantic

categories. J. Exp. Psychol.: General, 104(3), 192.

Tolentino, L. C., & Tokowicz, N., 2009. Are pumpkins

better than heaven? An ERP investigation of order

effects in the concrete-word advantage. Brain and

Llanguage, 110(1), 12-22.

Van Vliet, M., Chumerin, N., De Deyne, S., Wiersema, J.

R., Fias, W., Storms, G., & Van Hulle, M. M., 2016

Single-Trial ERP Component Analysis Using a

Spatiotemporal LCMV Beamformer. IEEE BME,

63(1), 55-66.

NEUROTECHNIX 2016 - 4th International Congress on Neurotechnology, Electronics and Informatics