The averaged waveforms of different valence, including negative, neutral, and positive
stimuli, are displayed in Figure 2. As demonstrated in Figure 2, the brainwaves contain clear
N1 and P2 elicited by all the three valence types of stimuli, followed by a negative-going wave (N400) peaking at around 300350 ms. In addition, the responses to the negative and
positive stimuli are more positive than those to the neutral ones during 500800 ms after the
onset of stimuli, especially over the center and the right hemisphere. To better capture this
pattern, Figure 3 is plotted to demonstrate the difference waves between emotional and
neutral verbs. The more positive responses elicited by emotional verbs reveal that there may
be an LPC effect in the current experiment task.
Figure 2. The grand average waveforms of negative, neutral, and positive stimuli
Figure 3. The topography of ERP differences between Negative stimuli vs. Neutral stimuli (Bin 1: Negative - Neutral) and between Positive stimuli vs. Neutral stimuli (Bin 2: Positive - Neutral) during 400-800ms interval
Figure 4 illustrates the grand average waveforms for stimuli with different polarization:
stimuli co-occurring with highly or low emotion-polarized nominal collocates. The brainwaves contain clear N1, P2, a negative-going wave during 300450 ms, and a following
positive wave peaking at around 500 ms. It also seems that across the two conditions, the
waveforms overlap throughout the whole epoch.
To further verify the patterns observed in Figures 2-4, and to examine the effects of
valence and polarization, repeated measures ANOVAs were carried out on the mean amplitudes in the time windows of 300450 ms (N400) and 500800 ms (LPC) on the lateral
regions (i.e. lateral analysis) and the midline electrodes (i.e. midline analysis) (See Chapter 3
Data analysis for details).
Figure 4. The grand average waveforms of highly and low emotion-polarized nominal collocates of stimuli
4.2.1 N400 (300-450 ms)
For the midline analysis, a 3 x 2 x 3 ANOVAs with the factors of valence (Negative,
Neutral, Positive), polarization (High, Low) and midline electrode (FZ, CZ, PZ) was
conducted. A main effect of valence was found (F(2,40) = 7.101, p < .005), with negative and
neutral stimuli eliciting larger amplitudes than the positive stimuli (Negative vs. Positive:
t(20)= -3.917, p < .005; Neutral vs. Positive: t(20)= -3.437, p < .01), and no significant
contrast between negative and neutral stimuli (Negative vs. Neutral: t(20)= .101, p = 2.762).
There was a marginal main effect of electrode as well (F(2,40) = 4, p < .05), revealing that
interactions were found (Polarization: F(1,20) = .763, p = .393; Valence x Polarization:
F(2,40) = .526, p = .595; Valence x Electrode: F(4,80) = 1.082, p = .371; Polarization x
Electrode: F(2,40) = .788, p = .419; Valence x Polarization x Electrode: F(4,80)= .215, p
= .867).
As for the lateral analysis, a 3 x 2 x 2 ANOVAs was performed with the factors of valence
(Negative, Neutral, Positive), polarization (High, Low) and laterality (Left, Right). No main
effects or interactions involving these three factors were found (Valence: F(2,40) = 3.425, p
= .042; Polarization: F(1,20) = .311, p = .583; Laterality: F(1,20) = .374, p = .548; Valence x
Polarization: F(2,40) = .416, p = 663; Valence x Laterality: F(2,40) = .195, p = .813;
Polarization x Laterality: F(1,20) = 2.638, p = .12; Valence x Polarization x Laterality: F(2,40)
= .693, p = .506).
Taken together, the polarization of stimuli’s habitual nominal collocates did not modulate
the valence in this time window. However, an emotion effect was found along the midline
channels (FZ, CZ, PZ), with the positive stimuli inducing smaller N400 compared with that
of the negative and neutral stimuli.
4.2.2 LPC (500-800 ms)
Similar to the statistical analysis in the N400 time window, two three-way repeated
measure ANOVAs were conducted separately on the midline channels (FZ, CZ, PZ) and the
lateral regions (Left, Right) to examine whether emotion effects within this time window
would be modulated by the polarization of stimuli’s habitual nominal collocates. For the
midline analysis, a 3 x 2 x 3 ANOVAs with the factors of valence (Negative, Neutral,
Positive), polarization (High, Low) and electrode (FZ, CZ, PZ) was performed. There was a
main effect of valence (F(2,40) = 7.857, p < .005), with positive stimuli eliciting larger
amplitude than the neutral ones (Positive vs. Neutral: t(20)= -5.688, p < .001), but no
significant contrasts found between negative and neutral stimuli (Negative vs. Neutral: t(20)=
2.349, p = .087) and between the emotional stimuli (Positive vs. Negative: t(20)= - .993, p
= .996). Apart from this effect of valence, no other main effects or interactions were found
(Polarization: F(1,20) = .424, p = .522; Electrode: F(2,40) = 1.479, p = .242; Valence x
Polarization: F(2,40) = 1.195, p = .313; Valence x Electrode: F(4,80) = .588, p = .619;
Polarization x Electrode: F(2,40) = 1.082, p = .328; Valence x Polarization x Electrode:
F(4,80) = .383, p = 755).
As for the lateral analysis, a 3 x 2 x 2 ANOVAs was performed with the factors of valence
(Negative, Neutral, Positive), polarization (High, Low) and laterality (Left, Right). A main
effect of valence (F(2,40) = 5.934, p < .01), a main effect of laterality (F(1,20) = 6.346, p
< .05) and an interaction between valence and laterality (F(2,40) = 3.777, p < .05) were found.
Follow-up pair-wise comparisons indicated that, in the left hemisphere, positive stimuli
induced larger amplitude than the neutral ones (Positive vs. Neutral: t(20) = -3.443, p < .01),
while no other significant contrasts were found (Negative vs. Neutral: t(20)= 1.559, p = .404;
Negative vs. Positive: t(20)= - .378, p = 2.128). Whereas in the right hemisphere, both
negative and positive stimuli elicited larger amplitudes than the neutral ones (Negative vs.
Neutral: t(20) = 3.305, p < .05; Positive vs. Neutral: t(20) = -4.563, p < .005), and no
significant difference between the emotional stimuli (Negative vs. Positive: t(20) = - .374, p =
2.136). A significance difference between the left and right hemisphere was found when the
stimuli were negative or positive, with the right hemisphere eliciting larger amplitudes than
the left (Negative stimuli, left region vs. right region: t(20) = -2.681, p < .05; Positive stimuli,
left region vs. right region: t(20) = -3.385, p < .01). Apart from the above, no other main
effects or interactions were found (Polarization: F(1,20) = .013, p = .909; Valence x
Polarization: F(2,40) = .757, p = .476; Polarization x Laterality: F(1,20) = 2.926, p = .103;
Valence x Polarization x Laterality: F(2,40) = .624, p = .541).
To sum up, similar to the midline analysis in the N400 time window, a main emotion
effect was found along the midline channels, although the exact pattern differed. In the N400
time window, both the neutral and negative stimuli elicited larger amplitudes than the positive
stimuli; however, in the LPC time window, the positive stimuli elicited larger amplitude than
the neutral ones, showing no significant contrast between the emotional stimuli. Furthermore,
there was an interaction between valence and laterality of the brain. Both the positive and
negative stimuli (the emotional stimuli) induced larger LPCs than the neutral ones in the right
hemisphere, whereas only the positive stimuli induced larger LPC than the neutral ones in the
left hemisphere. For the emotional stimuli, the LPCs were stronger over the right hemisphere
than the left. Finally, no modulation of collocates’ polarization on the effect of valence was
found.
Chapter 5 Discussions
Words are judged to be associated with positive or negative evaluation as they co-occur
mostly with other words that belong to positive or negative semantic set. “Semantic prosody”
is the term first introduced by Louw (1993) to describe this “consistent aura of meaning with
which a form is imbued by its collocates”. As for Sinclair’s definition (Sinclair 2004a),
semantic prosody speaks for a pragmatic function out of a juxtaposition of co-occurring
words, with its prior purpose to convey speaker/writer’s attitudes. Stewart (2010) further
remarked that shared features existing in the regular collocates (e.g. “undesirable things” for
the verb cause) would be acquired by the verb over time, leading to an epistemic reading
imposed on the verb. The different, but at time same time, similar interpretations on semantic
prosody in addressing “the transfer of pleasant or unpleasant messages from the habitual
collocates to the node verb” made we believe that it may be an informative issue if we can
examine the nature of semantic prosody from the cognitive neuroscience perspective.
The research question of this study was: Do people process the semantic prosody of
Mandarin-Chinese verbs, particularly the negative or positive emotive information induced
by collocational relation with nouns? Our results showed a main emotion effect in 300-450
ms time window (N400), with both the neutral and negative verbs inducing larger N400 than
the positive verbs. Later, in 500-800 ms time window (LPC), we found that both the
emotional verbs had larger amplitudes than the neutral verbs in the right hemisphere, while
only the positive verbs activated larger amplitude than the neutral verbs in the left hemisphere
and in the midline area of the brain (no significant contrast was found between the negative
verbs and neutral verbs). Additionally, specifically for the emotional verbs, we found the right
hemisphere had stronger responses than the left.
We first discuss the emotion effects to see if our study replicated previous findings about
emotional words. Previous ERP studies reported that a P2 enhancement was found for
emotional words as compared to neutral words, but the difference was less consistent
between positive and negative words (e.g. adjective materials in Herbert et al. 2006). In the
present study, we found a recognizable P2 effect of emotion as well. However, the brain
responses to the emotional verbs (positive and negative verbs) and the neutral verbs were all
very similar, showing that the emotion effect of verbs was not apparent in the early time
window, 200-300 ms post stimuli.
Despite the lack of the early emotion effect in P2, emotion effects were found in later
time windows: the N400 and LPC. As reviewed in Chapter 2, N400 was proposed as a brain
component modulated by emotional semantics, with decreased amplitude found for emotional
words as opposed to neutral words, because when the content of a word is involved with
emotion, it facilitates the access into long-term memory and hence leads to easier semantic
processing on emotional words (Kanske & Kotz 2007). The current study observed smaller
N400 for the positive verbs compared with the neutral ones, indicating that when verbs were
expressive with positive emotion, they were easier for semantic processing. Interestingly,
different from previous studies, no significant difference was found between the negative and
the neutral verbs. The unique status of positive verbs (i.e. being different from the neutral and
negative verbs) may be due to a general preferred processing towards positivity in humans
(Kissler et al. 2006). According to Ito & Cacioppo (2005), the positive motivational approach
system tends to be activated more strongly than the negative motivational withdrawal system.
The emotion effect was also found in the LPC: the LPC amplitudes were more enhanced for
the emotional (especially the positive) verbs than for neutral ones, replicating previous
findings (Schapkin et al. 2000; Fischler & Bradley 2006; Herbert et al. 2006). Furthermore,
we discovered that the emotion effect appeared in both hemispheres but was stronger over the
right, which was in line with Atchley et al.’s (2003, 2007) argument that the right hemisphere
is particularly involved in processing affective content of emotional words. The fact that there
was larger LPC amplitudes for emotional verbs than neutral ones may reflect more active
cognitive analysis on words with distinctly negative and positive emotion. That is, both
negative and positive verbs might strike more vivid mental images than the neutral ones,
whose emotionality was rather weak or uncertain.
Knowing that emotion effects in the literature were replicated in the current study, we
now turn to the polarization effect of the nominal collocates. Our behavioral results revealed
the polarization effect of the nominal collocates: verbs in the Low conditions were responded
faster than those in the High conditions. It was tempting to believe that this RT disadvantage
in the High conditions might come from the semantic influence of the nominal collocates.
However, our ERP data refuted this speculation and revealed no difference in the
“collocates-to-verbs” semantic influence between the High vs. Low conditions.
To further examine the existence of the polarization effect of collocates in the behavioral
data but its absence in the ERP data, we took a closer look at the two most frequent nominal
collocates (based on corpus’s frequency rating) of the verbs in the High conditions (namely,
conditions NH, PH, and NUH) since we found the accuracy rate for the neutral verbs was
significantly lower than the emotional ones within the High conditions. We found that these
collocates aligned their emotion valence with that of the verbs in the emotional verb (NH and
PH) conditions. This could imply that the negative or positive feature of the high-frequency
collocates might affect ERP participants in their judgement for the verbs being negative or
positive in the NH and PH condition, respectively. That is, semantic prosody might have long
been memorized as an important message of the verbs, and had been acquired through everyday repeated uses of “verb + collocates” constituent. Before pressing the button to
submit response, these ERP participants had already formed a clear picture of the verb’s
affective content in head, and therefore, no clear impact from the invisible collocates (our
manipulation on High vs. Low conditions) can be found on the online processing of verbs.
Interestingly, we discovered that two most frequent collocates in the NUH condition held
salient negative or positive emotion as well, with 13 neutral verbs’ collocates having negative
emotions and 19 neutral verbs’ collocates having positive emotions (; as for the remaining 3
NUH verbs: one with a negative collocate, another one with a positive collocate, the other
one with both a negative and a positive collocates). A question then arises: If the
“collocates-to-verbs” influence were prominent, as in the NH and PH conditions, why were
the neutral verbs, especially those in the NUH condition, evaluated “neutral” by the
population recruited in the pilot test?
This discrepancy about the valence of neutral verbs may be due to different time pressure
in the pilot test vs. the ERP experiment. In the ERP experiment, the participants’ brain
response to the verbs was captured with millisecond resolution under time pressure (i.e.
within 3 sec). Like the emotional verbs, the valence of the neutral verbs also came from the
valence of the high-frequency collocates; hence, the judgment of the neutral verbs did not
differ from that to the emotional verbs. In contrast, in the pilot test of valence-rating,
participants were totally free from time limitation, leading to a possible change in their
judgment because not only high-, but also low-frequency collocates were activated and
contributed to how participants rated the verbs. So, even if the participants did have some initial “emotional feeling” about the neutral verbs, it might have been cancelled out by other
co-activated, lower-frequency, collocates.
Therefore, the existence of the polarization effect in the behavioral data but its absence in
the ERP data might actually come from the same reason. That is, the judgment of the
positivity/negativity of the verbs is based on the valence of the high-frequency collocates, not
on how diverse/consistent the valence of the collocates are. The “frequency effect” of the
valence of the high-frequency collocates may result from years of language use by the
speakers themselves. Therefore, the Valence x Polarization interaction in the accuracy data
might simply reflect the possibility that the neutral verbs in the ERP experiment were
perceived with emotion carried by their high-frequency collocates, and such “misperception”
was even stronger when the collocates were highly emotional (i.e. mostly positive or
negative), as demonstrated by the 38.1% accuracy rate in the NUH condition.
In sum, despite the observation about semantic prosody of the “verb + collocate”
constituents in the linguistics literature, our experiment failed to find an online computation
of semantic prosody, as indicated by the existence of the emotion effect of the verbs and the
lack of the polarization effect of the invisible collocates. However, we did observe that the
valence of the verbs may come from the valence of the high-frequency collocates. We argue
that the semantic prosody might have been gradually formed during the process when a
speaker learns and uses the combination of the verb and its collocates, and later be
strengthened by the most frequent collocations between the verb and the following noun.
Once the semantic prosody is consolidated, it is directly associated with the verb and cannot
be easily altered by the valence of other collocates in online processing.
Chapter 6 Conclusion
In the linguistics literature, semantic prosody is collocational influence exerted on a target
word by its habitual or close surrounds. This study attempted to capture the online
computation of semantic prosody. Focusing on the emotion aspect of the target verbs
(Mandarin Chinese two-character transitive verbs), we further manipulated the verbs with
high or low emotion-arousing degree of their invisible nominal collocates. The experimental
design employed in our study was a 3 x 2 design, based on the interaction between two main
factors: Valence (Negative verbs, Neutral verbs, Positive verbs), and Polarization (Highly
emotion-polarized invisible nominal collocates, Low emotion-polarized invisible nominal
collocates). The results showed emotion effects for the emotional verbs and the neutral verbs
in the N400 and the LPC time windows. The smaller N400 for the positive verbs compared to
that of the neutral ones demonstrated easier semantic processing of words expressive with
positive emotions. Since no such contrast was found for the negative verbs (as well as a
distinct contrast between positive and negative verbs), we suggested this uniqueness of
positive verbs may due to human’s general bias towards positivity. The enhanced LPC for the
emotional (especially the positive) verbs may suggest the vivid mental images activated by
the verbs and reflect more active cognitive analysis on words with distinctly negative and
positive emotion. Furthermore, stronger LPC responses in the right hemisphere may indicate
right-hemisphere involvement in processing the affective content of words.
The current experiment failed to find an online computation of semantic prosody, as
indicated by the existence of the emotion effect of the verbs and the lack of the polarization
effect of the invisible collocates. However, we did observe that the valence of the verbs may
come from the valence of the high-frequency collocates. Thus, we argue that the semantic
prosody might have been gradually formed during the process when a speaker learns and uses
the combination of the verb and its collocates. And once the semantic prosody is consolidated,
it is directly associated with the verb and may have well become an epistemic or pragmatic
reading imposed on the verb.
This study may be improved for its future directions. The discrepancy in the emotion
judgment was found between the ERP participants and the valence-rating participants
recruited in the pilot test. Since the so-called “correct answers” were based on the result from
the pilot test in this study, we may reanalyze the ERP data according to the valence answers
made by the ERP participants. This way, different patterns of brain responses may be
revealed and provide a new window into the study of sematic prosody. In addition, the
perfectivity of verbs may considered further controlled in future study since
Mandarin-Chinese verbs may address differently in the perfectivity of the action it denotes:
some include the viewing of the beginning and end of the situation, while others focus on the
middle phase of the action, leaving the end of the situation unspecified.
References
Atchley, R. A., Ilardi, S. S., and Enloe, A. 2003. Hemispheric asymmetry in the processing of emotional content in word meanings: The effect of current and past depression. Brain and language 84.1:105-119.
Atchley, R. A., Stringer, R., Mathias, E., Ilardi, S. S., and Minatrea, A. D. 2007. The right hemisphere's contribution to emotional word processing in currently depressed, remitted depressed, and never-depressed individuals. Journal of Neurolinguistics 20.2:145-160.
Barsalou, L.W. 1999. Perceptual symbol systems. Behavioral and Brain Sciences 22.4:
577-660.
Bednarek, Monika. 2008. Semantic preference and semantic prosody re-examined. Corpus Linguistics and Linguistics Theory 4.2:119-139.
Begleiter, H., Projesz, B., and Garozzo, R. 1979. Visual evoked potentials and affective ratings of semantic stimuli. Evoked Brain Potentials and Behavior, 127-141. New York:
Springer.
Bower, G. 1981. Mood and memory. American Psychologist 36.2:129-148.
Bublitz, W. 1996. Semantic prosody and cohesive company: Somewhat predictable. Leuvense Bijdragen (Leuven Contributions in Linguistics and Philology) 85:1-32.
Clear, Jeremy. 1993. From Firth principles: Computational tools for the study of collocation.
Text and Technology: In Honour of John Sinclair, ed. by Mona Baker, Gill Francis, and Elena Tognini-Bonelli, 271-292. Amsterdam: John Benjamins.
Delorme, A., and Makeig, S. 2004. EEGLAB: An open source toolbox for analysis of single-trail EEG dynamics including independent component analysis. Journal of Neuroscience Methods 134.1:9-21.
Firth, J. 1951. Modes of meaning. Papers in Linguistics 1934-1951, ed. by J. R. Firth, 190-215. London: Oxford University Press.
Firth, J. 1957. Papers in Linguistics. Oxford: Oxford University Press.
Fischler, I., and Bradley, M. 2006. Event-related potential studies of language and emotion:
Words, phrases, and task effects. Progress in Brain Research 156:185-203.
Halliday, M. A. K. 1966. Lexis as a linguistic level. In Memory of J. R. Firth, ed. by C. Bazell, J. Catford, M. A. K. Halliday, and R. Robins, 148-162. London: Longman.
Hauk, O., and Pulvermüller, F. 2004. Effects of word length and frequency on the human event-related potential. Clinical Neurophysiology 115:1090-1103.
Herbert, C., Kissler, J., Junghofer, M., Peyk, P., and Rockstroh, B. 2006. Processing emotional adjectives: Evidence from startle EMG and ERPs. Psychophysiology
Herbert, C., Kissler, J., Junghofer, M., Peyk, P., and Rockstroh, B. 2006. Processing emotional adjectives: Evidence from startle EMG and ERPs. Psychophysiology