Chapter 2. Experiment 1: The influence of verb bias on ORC processing
2.1 Materials and Methods
2.1.5. EEG recording and data analysis
The electroencephalogram (EEG) was recorded using SymAmP2 produced by
NeuroScan with 64 electrodes from the 10-20 system. All scalp electrodes were
referenced to a common vertex reference located between CZ and CPZ and
re-referenced offline to the average of the right and left mastoids. Vertical eye movements
(VEOG) were recorded via electrodes placed on the supraorbital and infraorbital ridge
of the left eye, and horizontal eye movements (HEOG) were recorded via electrodes
placed at the outer canthus of both eyes. Electrode impedance was kept below 5Ω.
The EEG data were continuously recorded and digitalized at a sampling rate of 1000
Hz.
All the trials were included in ERPs preprocessing and further statistical analysis.
Given that the trials were relatively few, the EEG data were decomposed by applying
the ensemble empirical mode decomposition (EEMD) for data analysis. Previous
studies have suggested that EEMD can improve the signal-to-noise ratio (Subari,
Al-Baddai, Tomé, Goldhacker, et al., 2015; Al-Subari, Al-Al-Baddai, Tomé, Volberg, et al.,
2015 ; Chen, Chao, Chang, Hsu, & Lee, 2016 ; Cong et al., 2010 ; Hsu, Lee, & Liang,
2016 ). For instance, Hsu, Lee, and Liang (2016) applied EEMD to reanalyze the the
dataset of Cheng et al. (2013) and showed demonstrated that only one third of the
original trials were required to replicate the mismatch negativity (MMN) effect. Chen et
al. (2016), Tzeng et al. (2017) also applied EEMD for N400 measurement. This study
applied the same analytic procedure for N400, late positivity and frontal negativity
measurement, as described as follows:
(1) The time range for EEG segments for EEMD analysis was from 200ms before
stimuli onset to 1000 after the onset.
(2) The EEMD analysis was performed with 10 times of sifting and 40 ensembles.
The amplitude of Gaussain noises used in the EEMD procedure was 10% of
EEG signal’s standard deviation.
(3) Each EEG segment was decomposed into eight Intrinsic Mode Functions
(IMFs).
(4) According to the previous studies, the summation across IMF6, IMF7, and
IMF8 which cover the frequency range from 0.5 to 6.5Hz were used to extract
N400 signal, and IMF7 and IMF8 ranging from 1.65-3Hz for late positive or
negative waveform (Chen et al., 2016, Roehm, Bornkessel-Schlesewsky, &
Schlesewsky, 2007; Roehm, Schlesewsky, Bornkessel, Frisch, & Haider,
2004). Then, the summation was averaged over all trials for each condition and
each channel for each participant to yield the event-related modes (ERMs) to
represent the original ERPs.
2.2 Result
One participant was excluded from further analysis due to the insufficient valid
trails; therefore, this study analyzed the total data of twenty-nine participants.
2.2.1 Accuracy of comprehension test
The overall accuracy of comprehension test was 87% (SD= 0.06, range: 68%
-100%), showing that participants did not have difficulty understanding the sentences
and had pay attention in the experiment.
2.2.2 Statistical analysis for ERPs data: cluster-based random permutation analysis
To evaluate the temporal and topographical differences between conditions (SC vs.
DO, SC vs. EQ, and DO vs. EQ), the cluster-based random permutation analysis was
conducted on each of the following critical regions – embedded RC noun, RC verb, DE
and head noun – for the mean amplitudes of two epochs, N400 from 250 to 500ms and
late component (frontal positivity/ negativity) from 500 to 1000ms. First, for each
contrast, a simple dependent-samples t test was performed at each electrode. Electrodes
that exceeded a significance level (alpha = 0.1) were identified and formed as either
negative or positive clusters. For each cluster, the cluster-level test statistics was
calculated by taking the sum of all the individual t statistics within that cluster. Next, a
null distribution was created by computing 100 randomized cluster level test statistics.
The observed cluster-level test statistics was compared against the null distribution. The
clusters fell into the highest or lowest 2.5th percentile were considered significant. Then,
the cluster-based random permutation analysis was also performed on each time point to
identify and form the time cluster. Finally, 1000 randomized cluster-level test statistics
was conducted for each cluster on the basis of spatial and temporal adjacency. Thus, this
procedure yielded the significant cluster that displayed the contrast between conditions.
2.2.3 ERPs result (n = 28)
The cluster-based permutation analysis
Contrasts on embedded RC noun
Figure 2.1 presented the grand-averaged ERMs waveforms elicited by embedded
RC noun in three types of verb bias conditions. It showed that embedded RC noun
in the sentence with DO-bias verb elicited larger positive-going waveform in the late
time window than that in the sentence with SC-bias verb. The cluster-based random
permutation analysis (Figure 2.2) revealed that no significant negative cluster was found
in DO-SC contrast in time window of 250-500ms, but significant negative cluster in
DO-EQ (321-460ms; p<0.01) and SC-EQ (330-464ms; p<0.01) contrast. In the late time
window of 500-1000ms, DO-SC contrast (512-1000ms; p<0.01) elicited a significant
negative cluster in left frontal-to-central regions.
Figure 2.1. Grand averaged ERMs of the embedded RC noun for DO-bias, SC-bias, and EQ-bias conditions.
Figure 2.2. Topographic maps of embedded RC noun for DO-SC, DO-EQ, and SC-EQ contrasts in N400 and late time window (500-1000ms).
Note. Asterisks represent the significant differences for the contrasts.
Contrasts on RC verb
Figure 2.3 presented the grand-averaged ERMs waveforms elicited by RC verb.
RC verb in the sentence with SC-bias verb elicited larger positive-going waveform than
that in the sentence with DO-bias verb in the late time window. The cluster-based
random permutation analysis (Figure 2.4) revealed that no significant negative clusters
at DO-SC and DO-EQ contrast in the time window of 250-500ms were found. Yet, in
597-987ms
in frontal regions (617-1000ms, p<0.01)
Figure 2.3. Grand averaged ERMs of RC verb for DO-bias, SC-bias, and EQ-bias conditions.
Figure 2.4. Topographic maps of RC verb for DO-SC, DO-EQ, and SC-EQ contrasts in the N400 (250-300ms) and late time window (500-1000ms).
Note. Asterisks represent the significant differences for the contrasts.
Contrasts on DE
Figure 2.5 showed the grand-averaged ERMs waveforms elicited by DE. DE in the
sentence with DO-bias verb elicited negative-going waveform than that in the sentence
with SC-bias verb in the time window of 250-500ms. The cluster-based random
permutation analysis (Figure 2.6) revealed the significant negative clusters in central
and posterior sites in DO-SC contrast (318-455ms, p<0.01), significant negative clusters
in right central and posterior sites in DO-EQ contrast (304-500ms, p<0.01), and
significant positive clusters in central and posterior regions in SC-EQ contrast
(320-440ms, p<0.01) in the time window of 250-500ms.
Figure 2.5. Grand averaged ERMs of DE for DO-bias, SC-bias, and EQ-bias conditions.
Figure 2.6. Topographic maps of DE for DO-SC, DO-EQ, and SC-EQ contrasts in the N400 (250-300ms) and late time window (500-1000ms).
Note. Asterisks represent the significant differences for the contrasts.
Contrasts on head noun
Figure 2.7 showed the grand-averaged ERMs waveforms elicited by head noun.
Head noun in the sentence with DO-bias verb elicited larger N400 amplitude and a late
negative-going waveform than that in the sentence with SC-bias verb. In the time
window of 250-500ms, the cluster-based random permutation analysis (Figure 2.8)
revealed the significant negative clusters in DO-SC contrast (320-480ms, p<0.01) and
DO-EQ contrast (360-500ms, p<0.01) as well as significant positive clusters in SC-EQ
contrast (320-440ms, p<0.01). As for the late time window of 500-1000ms, the result
showed the significant negative clusters in DO-SC contrast (610-839ms, p<0.01) in the
right fronto-to-central regions.
Figure 2.7. Grand averaged ERMs of head noun for DO-bias, SC-bias, and EQ-bias conditions.
Figure 2.8. Topographic maps of head noun for DO-SC, DO-EQ, and SC-EQ contrasts in the N400 (250-300ms) and late time window (500-1000ms).
Note. Asterisks represent the significant differences for the contrasts.
2.3 Discussion- The incremental influence of verb bias on ORC processing
The present study aimed to examine the verb bias effect on the incremental ORC
processing on P600, frontal negativity, and N400 components. The cluster-based
permutation analysis was performed to characterize the spatial and dynamic of the verb
bias effect on ORC processing by contrasting DO-SC, DO-EQ and SC-EQ bias
conditions. The result of DO-SC contrast was summarized in Table 2.6 The marked
ones were specifically pointed out for the following discussion. (For the complete
ERPs result on RC verb, DE, and head noun with the focus on the DO-SC contrast.
Table 2.6. Summary of the ERPs result on DO-SC contrast
Our result indicated that verb bias incrementally influences the ORC processing, as
first indexed by the late frontal positivity effect on RC verb in the DO-SC contrast. RC
verb in DO-bias condition elicited a greater late positivity from 600 to 1000ms in
frontal regions than that in SC-bias condition.
According to the literature, the less preferred but grammatical structure causes
processing difficulty, as indexed by longer reading time (Lin & Gansey, 2011, Wilson
&Garnsey, 2008) and P600 components (Osterhout & Holocomb, 1992). When reading
the DO-bias verb as the main verb, the parsers would less likely to expect another verb
which indicates the appearance of the clause. Given that the DO-bias verb can be
followed by both direct object and sentential complement, with higher proportion of
taking direct object, the appearance of “another verb” does not violate the sentence
structure. However, DO-biased verb followed by another verb is a “less preferred but ERP
component
Embedded RC noun
RC verb DE Head noun
N400 No difference No difference 318-455ms 322-477ms late
components may reflect this phenomenon, P600 in posterior regions or P600 in frontal
regions. The posterior-distributed P600 is known as an index of grammatical violation
and has been associated with the less preferred but grammatical structure (Osterhout &
Holocomb, 1992). Recent studies have reported that frontal positivity can also reflect
the processing difficulty on the non-preferred but grammatical continuations (Kann &
Swaab, 2003; Leone-Fernandez et al, 2012) and suggested that processing non-preferred
grammatical continuations and ungrammatical continuations involve different
mechanisms (Kann & Swaab, 2003). Kaan & Swaab (2003) manipulated four
conditions to compare the non-preferred and ungrammatical continuations. Their data
showed that, the ungrammatical continuations elicited a typical posterior-distributed
P600, whereas, the non-preferred grammatical continuations elicited a greater positivity
from 500 to 1100 ms in more anterior sites. Leone-Fernandez’s (2012) study also
observed a similar frontally distributed positivity was also elicited by the non-preferred
grammatical continuations. For instance, in Spanish, two different locate predicates (e.g.
estar en and ser en :“to be in” in English) require different subjects, object and events,
respectively. event + ser en was rated as a non-preferable but acceptable continuation.
Comparing with the object + ser en which is the preferable continuations, event + ser
en, the non-preferable but acceptable continuations, elicited positive-going waveform
starting around 400ms till 700ms, especially for central and frontal regions. In our
study, RC verb in DO-bias condition did not elicit a typical posterior-distributed P600,
but a frontally distributed positivity. Our finding of verb bias effect on RC verb
demonstrated that RC verb is not preferable for the DO-bias verb in ORC processing.
This verb bias effect was also shown on the subsequent RC marker DE and head
noun as indexed by the larger N400 responses on RC marker DE and head noun in
DO-bias condition (ex: 他 想起 里長 資助 的 街友。). These findings are consistent
with Lin & Garnsey’s (2011) study which has demonstrated the long-lasting difficulty
on the non-preferred but grammatical continuations, the parsers had difficulty on the
subsequent words – DE and head noun, as indexed by the larger N400 responses on RC
marker DE and head noun in DO-bias condition. When encountering the RC verb
following the DO-bias verb, the parsers became more alert to expect a clause which
may be more likely to be conceptualized as an event. However, when DE appeared, it
signaled that it would be followed by a noun. Thus, the appearance of DE contradicted
to the expectation of an “event”, leading to a greater difficulty of integrating DE to “the
concept of an event” formed earlier and elicited a greater N400. Nevertheless, for the
SC-bias condition, the clause that follows the SC-bias verb can be constructed by
various syntactic structures. Therefore, the parsers had less integration difficulty on the
RC marker DE.
After the appearance of DE, for both DO and SC-bias conditions, the parsers
would expect a noun. Yet, the larger N400 amplitude and a frontal negativity on the
head noun in DO-bias conditions suggested that the processing of the head noun in
these two conditions were different. In DO-bias condition, the following noun was the
actual direct object of the main verb, so the parsers had to integrate the head noun with
the preceding context. This integration difficulty was reflected on N400. During the
integration process, the parsers had to re-assign the object of the main verb. Take one of
the experimental stimuli as an example “他 想起 里長 資助 的 街友。” . The
parsers would first assign “里長” as the direct object of the main verb “想起”, but when
they read “DE”, they would expect the following noun “街友” was the actual direct
object of “想起”. This re-assignment was indexed by a late frontally-distributed
negativity. Previous studies have reported that frontal negativity appears in the process
involved in establishing reference (Barkley et al., 2015; Van Berkum et al, 2007;
Nieuwland et al., 2006). Referentially ambiguous nouns or pronouns has been found to
elicit a sustained frontal negativity, relative to the unambiguous ones. Although the
manipulation of the head noun in this study was not associated with the referential
ambiguity, the process of integrating head noun to the preceding context involved the
referential binding. That is, the parsers looked for the suitable referent (里長 or 街友)
for the main verb. Therefore, frontal negativity elicited by the head noun in DO-bias
condition may reflect another kind of “process of establishing referential binding”.
As for the EQ-bias verbs that do not have a clear tendency of taking more direct
objects or sentential complements, our findings showed that it tends to exhibit similar
pattern as the DO-bias verb did. The result probably indicated that when processing the
verbs without clear syntactic or semantic tendency, parsers tended to expect a simplest
grammatical structure — direct object following the main verb. When they found that
the main verb was not followed by a direct object, they experienced difficulty, as shown
a similar pattern of processing DO-bias verb condition.
Nevertheless, out of expectation, differences in embedded RC noun in late time
window were found in DO-SC condition. Since the syntactic structure of noun
following verb is grammatical in either DO-bias or SC-bias condition, no difference
shall be found between noun following DO-bias and SC-bias verb. Yet, the result
showed that embedded noun following DO-bias verb elicited a larger late positivity than
that following SC-bias verb. The possible explanation to that is the limited S+V+O
condition. There are three types of fillers. One is SVO, the other two types are sentential
complement following the main verb. Only nine sentences of SVO type out of total 68
sentences in the experiment. Therefore, the participants would have less expectations on
the sentence that ends with “object”. When encountering the DO-bias verb, the
participant would be more alert that it would not be followed only by the direct object,
but might be the sentential complement, as shown by the sustained negativity in
DO-bias condition. In the light of this, types of filler sentences were better controlled in the
second experiment, with the inclusion of more simple SVO structure of sentences.
To sum, our findings suggested verb bias effect on ORC processing and proposed
the possible explanations for its long-lasting effect. The findings implied that ORC
following the DO-bias verb is more difficult to process than that following the SC-bias
verb and exhibits the similar pattern as ORC following EQ-bias verb
Chapter 3. Experiment 2: The influence of verb bias on SRC processing
Experiment 2 intends to examine the verb bias effect on real-time SRC processing.
Based on the categorization of verb bias defined in the first experiment, three types of
verb bias (DO, SC, and EQ bias verbs) would be followed by an object-modifying
Subjective Relative Clause (SRC) in this experiment. ERP analysis on four regions of SRC sentences, namely the RC verb, embedded RC noun, RC marker DE, and the head
noun, would be conducted to investigate the incremental influence of verb bias on SRC comprehension. Similar to the SC advantage in ORC processing that demonstrated in
Experiment 1, it was also expected to show the SC advantage in SRC processing in
Experiment 2. However, aside from the verb bias effect, the influence of word order in
sentence processing should be considered in the current experiment. Different from
ORCs in which the syntactic structure follows the typical Mandarin word order
Subject-Object-Verb (SVO) (e.g. subject+ main verb+ RC noun: SVO), SRCs (e.g. subject+
main verb+ RC verb: SVV) violates the typical SVO word order. Since ORCs follows
the typical word order, the clear contrasts between conditions in the first experiment are
simply resulted from the verb bias effect. Yet, the violation of typical word order in
SRCs might lead to the influence of both the role of word order and verb bias in SRCs
3.1 Materials and Methods
3.1.1 Participants
Thirty-four right-handed undergraduate and graduate students between the age of
18 and 26 participated in this study. Participants were all native speakers of Mandarin
Chinese in Taiwan with no history of neurological or psychiatric disorders. Each
participant signed the consent for before the experiment and were paid $500 NT of their
participation. All participants were evaluated for their working memory and verbal
memory capacities with working memory test of the Wechsler Adult Intelligence
Scales-fourth edition (WAIS-IV; Weschler, 2008) and reading span test. Detailed
information about the behavioral tests was listed in Chapter 2. Table 1 presents the
result of the behavioral tests.
Table 3.1. The result of behavioral tests
Behavioral tests Mean score Range
WAIS-IV (working memory capacity) 37(3.8) 31-44
Reading span test 2.4(0.7) 1.5-4
Note. Standard deviations are in parentheses.
3.1.2 Experimental Design
This study aimed to manipulate three types of verb bias, Direct Object (DO),
Sentential Complement (SC), and Equilibrium Balanced (EQ) bias verb, which are
followed by a SRC, leading to the following syntactic structure of the target sentences:
Subject + DO/SC/EQ-bias main verb + SRC structure (RC verb + embedded RC noun +
RC marker DE + head noun). In order to make the sentences with SC-bias verb more
complete, in SC-bias condition, the SRC was followed by a verb, leading to the
structure of sentential complement. The current experiment shared the same verbs with
the first experiment. The classification of verb bias has already been defined in the first
experiment.
The construction of target sentences and fillers
The present experiment consisted of 41 target sentences with the manipulation of 14 DO-bias, 13 SC-bias, and 14 EQ-bias verbs as well as 42 fillers (Table2). Each
critical region of the SRC, were matched for word length, word frequency and the
critical region of the SRC, were matched for word length, word frequency and the