Procedure for the ERP session

Participants were seated 100 cm from the computer screen in a quiet testing room.

The experiment began with a written instruction along with an 8-trial practice session for the purpose of familiarizing subjects with the task and the experimental environment. The trial procedure is shown in Figure 3.1. At the beginning of each trial, a plus sign appeared in the center of the computer screen for 500 ms to announce an upcoming word pairs.

After a stimulus onset asynchrony (SOA) ranging randomly between 1000 and 1500 ms, each syntactic cue appeared in the center of the screen for 200 ms. The offset of the cue was followed by a 300 ms inter-stimulus interval (ISI) and then the target word was presented for 200ms. After a 1000 ms blank from the offset of the target word, a message xiàyītí /下一題/NEXT TRIAL was presented centrally on the screen. The message remained on the screen for 2500 ms, and the next trial started after a delay of 1500ms.

The whole experiment was divided into 4 blocks, each lasting about 3.5 min.

Participants were asked to finish a paper-and-pencil word recognition task at the end of each block to ensure they were focused during the experiment. The word recognition task consisted of 12 old phrases appeared in each block, as well as 12 pseudo phrases in which half were ambiguous phrases. Participants were asked to check off each phrase that they

Two neuropsychological tests were conducted separately following the ERP recording session, including assessments of reading ability (Reading experience test:

Acheson & MacDonald, 2008) and executive function (Verbal fluency test: Benton &

Hamsher, 1978).

Figure 3.1. A diagram of trial procedure in Experiment 1

3.3 EEG recording parameters and data analysis

The electroencephalogram (EEG) was recorded using 32 sintered Ag/AgCL electrodes from the 10-20 system (QuickCap, Neuromedical Supplies, Sterling, TX, USA) (see Figure 3.2). All scalp electrodes were referenced to a common vertex reference located between Cz and CPz online, and re-referenced to the average of the right (M1) and left (M2) mastoids offline. Vertical eye movements were recorded via a pair of electrodes placed on the supraorbital and infraorbital ridge of the left eye, and horizontal eye movements were recorded via electrodes placed at the outer canthus of each eye in a bipolar montage. Impedance was kept below 5kΩ for all electrodes. The continuous EEG was amplified by the SYNAMPS2 amplifiers (Neuroscan, Inc., EL Paso, Texas, USA)

The EEG data were segmented offline into 1400 ms epochs, spanning 200 ms pre-stimulus to 1200 ms post-pre-stimulus. Trials contaminated by artifacts from amplifier blocking, signal drifting, muscle activity, eye blinks and movements were rejected offline before averaging. The averaged ERPs had the baseline corrected over the 200 ms pre-stimulus period, and were digitally filtered with a band-pass of 0.1-30 Hz. Only corrected trials were included in the following analysis. Overall, trial loss due to artifacts and incorrect responses averaged 27%. For all participants there were at least 15 trials in each condition.

Figure 3.2. Shown are the locations of the 30 scalp electrodes on the QuickCap used in the present study. The electrodes used for statistical analysis are triangles for frontal electrodes, and circles for central/posterior electrodes. For those electrodes filled in shapes are used for showing the representative waveforms.

3.4 Result

Twenty participants took part in Experiment 1; all have at least 15 valid ERP trials in each condition. Data from these twenty participants are represented as follows.

3.4.1 Behavioral data

Participants’ overall accuracy rate for the word recognition task was 74.4% (SD = 0.07), As this is a very simple task, this relatively low performance suggest that participants might not be fully attending to the stimuli.

3.4.2 ERPs data

Figure 3.3 shows overall ambiguity effect that the ERP responses to unambiguous and ambiguous words at three representative midline electrode sites (FZ, CZ, PZ). An ANOVA with 3 levels of Ambiguity (UA vs. AAD vs. AAS) and 2 levels of Electrode site (anterior:F3, FZ, F4, FC3, FCZ, FC4, C3, CZ, and C4; central/posterior: CP3, CPZ, CP4, P3, PZ, P4, OZ, O1 and O2) was conducted on mean amplitudes of data measured between 250-900 ms after the onset of target words. Analyses were first performed on the overall ambiguity effect. There was no difference in mean amplitude response between 250-900 ms for AAD (ambiguous in dominant-biasing context) and AAS (ambiguous in subordinate-biasing contexts) as compared with unambiguous words [F(2, 38) = 2.48; p

= .09]. To examine whether the ambiguity effect was modulated by biasing context, we then conducted a follow-up comparison. The result revealed the effect of Ambiguity was marginally significant only for the AAS (subordinate-biasing context) [F(1,19)= 3.62, p=.07], but not for AAD (dominant-biasing context) (p values=.89).

To know if this ambiguity effect manifests on two behavioral indexes, mean amplitudes between 250-900 ms were subjected to an ANOVA with 3 levels of Ambiguity (UA vs. AAD vs. AAS), 2 levels of Electrode site (anterior vs.

central/posterior), and 2 levels of Group (high vs. low). The results showed a significant main effect of Ambiguity in reading experience test [F(2,36)= 2.81, p=.04], but not in verbal fluency test. We then conducted 2 ANOVAs within the high reading experience

group. One with 2 levels of Ambiguity (UA vs. AAD) and 1 level of Electrode site (F3, FZ, F4, FC3, FCZ, FC4, C3, CZ, C4, CP3, CPZ, CP4, P3, PZ, P4, OZ, O1 and O2), and the other with 2 levels of Ambiguity (UA vs. AAS) and the same Electrode site. There is a significant main effect of AAS [F(1, 8) = 7.46; p = <.05], but not for AAD (p=.44).

Previous studies have shown a slow frontal negativity was observed for syntactically and semantically ambiguous words relative to their unambiguous counterparts beginning around 250 ms when preceding context provides well-specified syntactic information but very little semantic information (Lee & Federmeier, 2006, 2009, 2012) that indicated. As shown in Figure 3.3, the ERP responses were more negative to ambiguous relative to unambiguous words. The difference emerged at around 250 ms post stimuli onset and lasted to the end of the epoch.

Figure 3.3. Grand average ERPs at three midline electrode sites for unambiguous words (black line) and ambiguous words (red line) in Experiment 1

specified contexts elicited a sustained frontal negativity between 250 and 900 ms post-stimulus-onset in prior studies (Lee & Federmeier, 2006; Lee & Federmeier, 2009), the present study, however, displayed a relatively small effect. However, in Figure 3.4, the result of follow-up comparison showed that the slight frontal negativity was only in the subordinate-biasing context but not in the dominant-biasing context. The preliminary findings might correspond to our prediction that the preceding context indeed affects meaning access on Chinese biased homographs, and meaning dominance can interact with the syntactic context effect.

Figure 3.4. Grand average ERPs at three midline electrode sites for unambiguous words (black line) vs. ambiguous words (red line) when the context favors dominate meaning (left column) and subordinate meaning (right column) of the homographs in Experiment 1.

One possible reason is that the task used in this experiment was word recognition task, which did not require participants to integrate the syntactic context and the target word. In other words, participants could process these phrases as list of words. We will follow up on this in the second experiment reported in the next chapter.

3.5 Interim summary and discussion

The result of Experiment 1 provides some implications to our research questions.

We found a hint of the frontal negativity effect. However, the effect was much smaller compared to those reported in past research in English. Tanner et al. (2018) pointed out that grand mean ERP waveforms may not always reflect the central tendency of the population, despite the statistically reliable effects. According to their standpoint, it is possible that the grand mean waveforms are subjected to topographic or temporal distortion. Since previous experiments also revealed that the score on verbal fluency is linked with the effect pattern, we thus look into individual data on the basis of the neuropsychological performances. Intriguingly, such typical ambiguity effect was clearly seen on the majority of subjects with high score of reading experience. This result, on the one hand, provides a preliminary support for the view that the interpretation of ambiguous words can be influenced by experience (Rodd et al., 2016, 2013). On the other, it might explain the inference from the grand average data might be illusions of overlapping different ERP effects behind and thus gave rise to the unrecognizable ambiguity effect in grand-average waveforms across all participants.

After reexamining the data on the basis of high score group of neuropsychological tests, we found the prevalent negativity across the scalp when the context selected the subordinate meanings of the homographs. When the context favored the dominate meanings of the homographs, the ambiguity effect was relatively insignificant, which might denote the dominant meaning of the homographs are more likely to be processed as unambiguous words. Consequently, the meaning activation began earlier around 400-500 ms and needed not to sustain for long as compared with the subordinate meaning of homographs which involved in a meaning selection. Perhaps it provides some evidence

for the view that meaning dominance indeed exert influences on lexical ambiguity resolution.

On the other hand, the comprehension task— word recognition task— used between blocks in present experiment might be the possible reason causing such a relatively slight effect. In fact, prior studies applied a semantic relatedness judgement task to facilitate participants to process the meaning of the whole phrases. In comparison, the word recognition task needed not to integrate the syntactic cues and the target word; instead, it could be done by recalling the target words. It seemed to increase participants’ burden on memory load rather to help integrate the visual information. Since the accuracy rate of the task was also significantly lower, word recognition task might not be that ideal as we thought.

To make the above-mentioned potential factors clear, we began to do following follow-up studies like analyzing the individual data by grouping the items based on median split of behavior indexes, modifying the experiment by replacing another online comprehension task. we attempted to look for if these factors really make impacts on the result.

3.6 Follow up analysis: Inter-individual variability analyses

In view of inter-individual differences observed in past research (Lee & Federmeier, 2012), we also set out to explore the individual differences in the overall ambiguity effect across all participants. Figure 3.5 plots the mean amplitude differences during a typical time window for the frontal negativity effect (250-900 ms) post stimuli-onset at a representative channel (FZ) for each participant. This analysis revealed a great amount of individual variation within the interested window, with approximately half of the

participants exhibiting an extended negativity and half a positivity, resulting in the cancellation of a significant overall main effect.

Figure 3.5. Effect sizes per participant for ambiguity manipulation at the representative frontal channel (Fz) within 250-900 ms.

To try to account for the source of individual variations, we analyzed the inter-individual variation depending on several neuropsychological indexes. According to past work, the effect patterns might differ due to participants’ cognitive abilities. Verbal fluency test, for instance, has been widely assessed to measure verbal ability including lexical knowledge and lexical retrieval ability (Cohen et al., 1999; Weckerly et al., 2001;

Federmeier et al., 2010) and executive control ability (Henry & Crawford, 2004;

Fitzpatrick et al., 2013). Some related research has also indicated that better performance on verbal fluency is linked with greater amount of frontal negativity elicited by homographs (Lee & Federmeier, 2011). Motivated by previous findings, we attempted to look at group averages based on a median split of participants’ neuropsychological performance to examine if the unapparent overall ambiguity effect was derived from the individual differences.

Two neuropsychological tests were conducted in this study— reading experience assessed by an author and magazine recognition questionnaire and verbal fluency. As a first step to explore the possible influence of these two types of cognitive abilities, participants were divided into high and low score groups which was created by means of median-split method (see Figure 3.6). The means of low and high score groups in reading experience test were 28% and 51% respectively. For verbal fluency test, the mean of low score group was 105.2 and 140.7 for high. We grouped the ERPs according to the high/low group of two neuropsychological tests and observed relations between the brain responses and the cognitive abilities (see Figure 3.7 & Figure 3.8).

Figure 3.6. High/low score group based on a median split for two neuropsychological tests across twenty participants

Figure 3.7. Grand average waveforms to ambiguous words (red line) and unambiguous words (black line) of the low and high score group for verbal fluency test are plotted at 3 representative midline electrode sites (Fz, Cz, and Pz) to observe the overall effect. There is no statistical significance between ambiguous words and unambiguous words in each group.

Figure 3.8. Grand average waveforms to ambiguous words and unambiguous words of the low and high score group for reading experience test are overlaid at 3 representative midline electrode sites (Fz, Cz, and Pz) to highlight the overall effect. There is a prominent statistical significance (p<.05) in high score group, while there is no difference between ambiguous words and unambiguous words in low score group.

The results showed that higher score groups for both verbal fluency and reading experience showed more frontal negativity effect than did the lower score groups.

However, the between group difference was particularly robust for reading experience.

To inspect if the inter-individual variability did result from the relation with reading experience, we plotted a boxplot to compare the brain response of the high and low score group for reading experience test at a representative channel (Fz) (see Figure 3.9). It showed obviously that most subjects with higher score in reading experience were likely to show a negativity to ambiguous compared to unambiguous words over frontal channels,

Figure 3.9. The boxplot plotted on the basis of the high and low score group for reading experience test at Fz to represent the brain response within two groups.

Motivated by this finding, we continue to investigate whether in the high score group of reading experience, these differences would be larger when the context biases the subordinate meaning of the homographs on the basis of the findings in Lee & Federmeier (2009). Figure 3.10 shows in the high group of reading experience, the waveforms to ambiguous vs. unambiguous words when the context favors dominate and subordinate meaning of the homographs, respectively.

Figure 3.10. In the high group of reading experience, ERPs’ responses at three representative electrodes of ambiguous (red line) vs. unambiguous words (black line) when the context favors dominate meaning (left column) and subordinate meaning (right column) of the homographs. The bottom four isopotential voltage maps show scalp distributions viewed from the top of the head for brain responses in both contexts in two time windows (250-550 ms and 550-900 ms). The statistical significance of the difference between unambiguous and ambiguous words is noted only in subordinate-biasing context.

(p<.05) A clear contrast between groups shows that a notable frontal negative effect is elicited only in subordinate-biasing context.

In accordance with our prediction, in the high score group of reading experience, a robust sustained negativity was elicited when the context favors subordinate meaning of an ambiguous word. The effect was quite widespread, with the effect being only slightly larger in the frontal than in the central and posterior channels. In contrast, there is very little difference between unambiguous words and ambiguous words in the dominate-biasing context, except for the N400 effect.

3.7 Follow up experiment: Modification of experimental task

whether the weak frontal negative ambiguity effect to ambiguous words compared with unambiguous words in Experiment 1 was derived from the insensitive comprehension task was of pivotal importance. Therefore, we decided to substitute the online semantic relatedness judgement task used in Lee & Federmeier (2006) for the word recognition task used in Experiment 1 to enhance participants’ attention so that they are more likely to process the cues along with the target words as phrases. For another consideration, participants in Experiment 2 were required to do both experimental lists to examine if there was a list difference which gave rise to such the disparate brain responses. We expect to observe a more robust effect as that in the previous studies with this modification.

3.7.1 Design and prediction

Design were identical with those used in Experiment 1 except that participants were asked to do a semantic relatedness judgment task instead of the word recognition task in Experiment 2. The semantic relatedness judgement task has been widely used to examine how target words are semantically represented in mind. Participant were asked to decide if the two phrases presented on the screen are related or unrelated in meaning. The purpose of this task was to require participants to integrate the syntactic cue and the target word to process the phrase as a phrase but not a pair of words. In so doing, we hope to encourage participants to interpret the meaning of the target based on the given context. In addition, this task can also help to ensure that the participants indeed attend to the stimuli. Also, as data analysis in Experiment suggest a possible role of list in explaining the individual difference, participants in Experiment 2 were tested with both experimental lists, with the order of lists counterbalance among participants.

3.7.2 Method

Six right-handed young adults participated in this ERP experiment (5 females; mean age 22.4 years, age range 21-24) for cash. All were Chinese native speakers and have neither been exposed to other languages other than Taiwanese before the age of five nor had history of neurological or psychiatric disorders or brain damage. All participants were right-handed as measured by the Chinese translated version of Edinburgh inventory (Oldfield, 1970), with the mean laterality quotient being 0.81 (SD = 0.14 range = 0.6-1.0).

No participants had known left-handed blood relatives, as assessed by a familial handedness questionnaire (Lee & Federmeier, 2015). Written consent was obtained from all participants. Participants were randomly assigned to one of the two experimental lists, and also, none had participated in any norming study.

Materials were identical with those used in Experiment 1. In addition, forty-two probes for semantic relatedness judgment were created. Half of the probe trials were semantically related to their target words, and the other half were semantically unrelated.

The former was created on the basis of either the synonym or definition of a target word (e.g., sāndàoliàolǐ/ 三道料理 / three-CL dishes — sāndàocàiyáo/ 三道菜餚 /three-CL meals), and words unrelated to any sense of the target words were obtained for the latter (e.g., fēichángbiànyí/非常便宜/very cheap — fēichángcōngmíng/非常聰明/very clever).

In addition, probes always contained the same syntactic cue as that used in the immediately preceding trial in order not to draw extra attention to either word class or ambiguity of the words. All probes were well designed so that there were no trick questions.

Participants were seated 100 cm from the computer screen in a quiet testing room.

The experiment began with a written instruction along with an 8-trial practice session for

the beginning of each trial, a plus sign appeared in the center of the computer screen for 500 ms to announce an upcoming word pairs. After a stimulus onset asynchrony (SOA) ranging randomly between 1000 and 1500 ms, each syntactic cue appeared in the center of the screen for 200 ms. The offset of the cue was followed by a 300 ms inter-stimulus interval (ISI) and then the target word was presented for 200ms. After a 1000 ms blank from the offset of the target word, a probe for a semantic-relatedness judgement followed one-third of the target words. Once the probe was displayed in phrase on the screen, participants needed to determine whether the phrase was semantically related or unrelated

the beginning of each trial, a plus sign appeared in the center of the computer screen for 500 ms to announce an upcoming word pairs. After a stimulus onset asynchrony (SOA) ranging randomly between 1000 and 1500 ms, each syntactic cue appeared in the center of the screen for 200 ms. The offset of the cue was followed by a 300 ms inter-stimulus interval (ISI) and then the target word was presented for 200ms. After a 1000 ms blank from the offset of the target word, a probe for a semantic-relatedness judgement followed one-third of the target words. Once the probe was displayed in phrase on the screen, participants needed to determine whether the phrase was semantically related or unrelated