Eye Tracking in Measuring the Relation of Anaphor Processing, Reading Fluency and Working Memory (WM) Capacity among University Students

This paper aimed at investigating (i) eye movement behaviour as indication of reading fluency during anaphor processing of young adults; (ii) and the relation between eye movement behaviours and Working Memory (WM) among young adults. This experiment used eye-tracking device to study the eye movements and Auditory Working Memory subtest to measure WM capacity. Tobii Eye-tracker was used to measure the movements of eyes for the participants. The results showed that young adults have similar eye movement behaviours, which were inverse typicality effect, as children. However, the score of their WC was similar to past studies whereby a larger WC capacity indicates better accessibility of resources and thus faster resolution of anaphors and antecedents. These findings were analyzed to determine the implications on the research development of eye-tracking towards students’ reading fluency during anaphor processing and its relation to working memory.


Introduction
Past empirical studies have associated the relationships between reading fluency, anaphor processing and Working Memory (WM) through eye-tracking. Joseph, Bremner, Liversedge and Nation (2015) revealed that the magnitude and time course of children during anaphoric processing was seen to be greatly affected by WM capacity. One important finding fro m the study was in the antecedent region whereby the interactions among reading fluency, WM and typicality of the anaphor resolution revealed a difference in time course for those with h igh and low WM capacity. On the other hand, Garrod and Sanford (1977) in a study stated that adults exhibited shorter reading times when the antecedent was typical and presented longer first-pass reading times when anaphors were further fro m their antecedents. In addition, a large d ifference was observed due to individual d ifferences in the WM capacity which affect reading fluency in adults (Carpenter, M iyake, & Just, 1994;Daneman & Carpenter, 1980;M ihat, Azman & Soh, 2018;Hasrul, Hazita & Azizah, 2018;Philip, Tan & Jandar, 2019) and in ch ildren (Cain, Oakhill, & Bryant, 2004;Pimperton & Nation, 2010).
Reading fluency is the ability to read accurately and smoothly through recognition of individual words and then the integration of the meaning of words (Gernsbacher, 1990;Johnson & Laird, 1983;Van Dijk, Kintsch, & Van Dijk, 1983) as well as the ability to associate between current and previous information, such as in anaphor processing (Garrod, O'Brien, Morris, & Rayner, 1990;Garrod & Terras, 2000). McDonald (1995) described anaphor processing as the process of relating an antecedent to its anaphor. Antecedent (e.g. Mary) is usually a noun phrase which is mentioned earlier and provides the interpretation for a second expression, the anaphor (e.g. she). During anaphor processing, readers have to determine the potential antecedent in the preceding text and then deduce the relationship between the antecedent and the anaphor (Garrod & Terras, 2000) which indicates a high demand of underlying complex cognitive skills and resource, specifically WM. WM as stated by Cowan (2010), "is the small amount of information stored in a readily accessible state in order to perform any cognitive tasks (p.447)" which can be used to reason out limitations in processing capacity (Leonard, Weis mer, M iller, Francis, Tomb lin, & Kail, 2007) and has captured the interest of Reading Fluency and Working M emory (WM ) Capacity among University Students researchers to establish its link to eye movement behaviours.
There were a considerable amount of studies researching on eye movements, anaphor processing and WM but majority have not co mbined all the elements together. In addition, there is a scarce of existing studies in this context among young adults. Hence, this study aims at examining anaphor processing, reading fluency and WM capacity among young adults by referring to the framework of Joseph, Bremner, Liversegde and Nation (2015) and incorporating eye movements. This study focused on: i). eye movement behaviour as the indicat ion of reading fluency during anaphor processing of young adults. ii). relationships between eye movement behaviour and WM among young adults.

Eye Movement Behaviour as the Indication of Reading Fluency during Anaphor Processing of Young Adults
The typicality of an antecedent to its anaphor affects reading fluency wh ich can be observed through eye movement behaviours, in terms of fixat ion durations. Results from previous studies among adult readers (Rosch, Mervis, Gray, Johnsen & Boyes -Braem, 1976;McKoon & Ratcliff, 1980;Duffy & Rayner, 1990;Soh, 2017;Sarah Yusri & Soh, 2019;Soh, Hazita, Noremy & Shamita; implied that reading fluency was disrupted when the typicality of an antecedent was reduced and thus lengthened reading times as fixat ion durations increased. One of the study conducted by McKoon and Ratcliff (1980) revealed that shorter fixat ion duration of the antecedent was observed when the anaphor was directly repeated or explicit; whereas Duffy and Rayner (1990) found that longer fixat ion durations denoting longer processing time occurred when the antecedent of a category name anaphor (e.g. b ird) had a low typicality of the category (e.g. goose) as compared to a high typicality member (e.g. robin). It is presumed that repeating of nouns and antecedents with high typicality provided greater sematic overlap, therefore reduced fixation durations and increas ed reading fluency in anaphor processing.
A study by Joseph, Bremner, Liversedge and Nation (2015) investigated on the effects of typicality on eye movements among children and observed effects of typicality relat ively early during anaphoric processing, fo r instance, on the anaphor itself. They found no reliable effects of typicality in eye movement behaviours, such as first fixation durations, gaze durations, go past times or total reading times. Ho wever, they found effects of typicality during regression on the anaphor regions to re-read previous portions of the text. To specify, frequency of regression on the anaphor region was higher when the antecedent was typical than atypical. The result was inconsistent to the hypothesis made and contrary to the eye movement behaviour of adults, showing an inverse typicality effect. The findings can be exp lained by referring to the view of Garrod and Terras (2000) where they stated that the inverse typicality effects reflected children's early anaphoric processing, in which children used bottom-up, context-blind lexical cues to lin k the anaphor to the typical antecedent but not the atypical antecedent; and the closer sematic overlap of typical antecedent activated the memo ry, hence, the inverse effect was justifiable as children, unlike adults, did not process atypical antecedent as it was not being activated. In addition, regressions on antecedent regions were slightly higher when the nouns were atypical members of the semantic category. Th is was in line with the prediction and consistent with the eye movement behaviours demonstrated by adult readers. Therefore, the purpose of this research is to investigate on the effects of reading fluency during anaphor processing by studying the eye movement behaviours of young adults and compare whether their eye movements portray similarity to children or adult readers.

Relationships between Eye Movement Behaviours and WM among Young Adults
In the past decades, previous researchers had highlighted the fact that eye-tracking had been used intensively as the method to study the sentence-level online processing based on the assumption that cognitive processes correlated with eye movement patterns during reading tasks which provided details such as fixation duration, gaze duratio n, first pass, and second pass (Just and Carpenter 1980;Pollatsek, Juhasz, Reichle, Machacek, & Rayner, 2008). They believed that eye-tracking delivered an online measure of the mental operations during reading processing. Just and Carpenter (1992) pointed out that WM capacity has the potential to provide adequate and accurate explanations of individual variations in language comprehension and processing efficiency during reading tasks. In reading tasks, WM capacity is highly influential whereby it must be allocated to the various processes involved in act ivating lexical representations and integrating them into developing syntactic and discourse representations (Caplan & Waters, 1990, 1999MacDonald & Christiansen, 2002;Soh, Hazita & Ho, 2020), such as anaphoric processing. It is presumed that large WM capacit ies increased the ability of individuals to execute more processes in a g iven period of time and thus shorter fixation durations. For examp le, Joseph, Bremner, Liversedge and Nation (2015) examined on the relation of semantic typicality during anaphor processing through eye movements with individual d ifferences in working memo ry capacity and reading comprehension skills. They mentioned that limitations in verbal WM is one of the reasond of poor anaphoric processing during reading.
Furthermore, ch ildren with good vocabulary knowledge were expected to access words more efficiently wh ich allo wed more resources for comprehension processes and present shorter fixat ions and reading times which indicated better reading fluency (Perfetti, 2007;Soh, 2016). Thus, they were able to process an atypical antecedent faster, and thus resulting in an earlier typicality effect. In contrast, studies (Kuperman & Van Dy ke, 2011;Schroeder, 2011) on adolescents and adults demonstrated longer fixat ions and reading times during processing of anaphors with atypical antecedents. These findings raised a question among researchers whether the longer fixations and reading times in individuals with larger WM capacity could be the fact that they are more purposeful readers, thus allo wing themselves more time to process text elements that are difficult. Thus, this research will investigate on the relation between eye movement patterns and WM among young adults and compare the findings with the previous studies to establish the relationship between WM capacity and eye movements of young adults. Thirty young adults participated voluntarily in this eye-tracking study. They were females who majored in English Language Studies in The National Un iversity of Malaysia (UKM), which were of different races -Bidayuh, Kelab it and Malay. The age range of the part icipants are fro m 20 to 21 years old. Tab le 1 showed the distributions of participants. All of the participants had no prior exposure to eye-tracking research.

Instruments
The stimuli used were comprised of sixteen sets of short texts in English which were adapted from journal articles to maintain the reliability and validity of the experiment. The texts had an average of 3 to 5 sentences each. Each text had two versions presented in the same slide side-by-side. The first text had a more familiar antecedent (e.g. truck) while the other had fewer typical (e.g. crane). The typical and atypical antecedents did not differ in length. Both versions of texts within the same context had the same anaphor (e.g. truck). The anaphor and antecedents were the area of interest of the texts.
The anaphor regions are underlined and the antecedent regions are italicized.
Participants' eye movements were recorded using a Tobii X120 non-intrusive eye-tracker with 120 Hz sample rate. The apparatus was chosen for its features as it has (i) the flexibility in allowing slight head movements; (ii) the ability to carry out experiments for a long period of time without causing tiredness of the eyes; and (iii) high sensitivity in tracking and detection of word-to-wo rd eye movements. The participants were not required to scroll the page as the texts were fixed to fit the whole screen. The participants had to signal the experimenter to proceed to the next slide. It had been a long day. The builders were exhausted. Eventually a truck arrived to help. They needed the vehicle because the load was so heavy. At last they could start work on the building.
It had been a long day. The builders were exhausted. Eventually a crane arrived to help. They needed the vehicle because the load was so heavy. At last they could start work on the building. It was a sunny afternoon. Judy was walking home from school, daydreaming about herself. She crossed the busy street at the light. She stopped to watch the traffic while she gathered her thoughts. She had just decided she was going to be a doctor when she grew up. The profession seemed to be an exciting one. She knew she would have to work hard at school.
It was a sunny afternoon. Judy was walking home from school, daydreaming about herself. She crossed the busy street at the light. She stopped to watch the traffic while she gathered her thoughts. She had just decided she was going to be a meteorologist when she grew up. T he profession seemed to be an exciting one. She knew she would have to work hard at school. Adam had been driving around the city for thirty-five minutes. He'd been trying to find the train station but it was nowhere in sight. Although the map to the station was old, Adam still couldn' t understand why he was unable to find it. He stopped and asked for directions and realized that he was in the wrong town.
Adam had been driving around the city for thirty-five minutes. He'd been trying to find the gas station but it was nowhere in sight. Although the map to the station was old, Adam still couldn' t understand why he was unable to find it. He stopped and asked for directions and realized that he was in the wrong town.
Eye Tracking in M easuring the Relation of Anaphor Processing, Reading Fluency and Working M emory (WM ) Capacity among University Students

Procedures
A trial experiment were conducted before the actual experiment to determine the best condition to present the stimuli in terms of the font size, text configurat ions, average time taken to read each text and the sequence. The trial was conducted almost similarly to the actual experiment. The participants for the trial experiment were not involved in the actual experiment. At the end of the trial, the participants were interviewed to find out the best sets of condition for the stimuli.
All the participants were gathered for a short briefing before the actual experiment was carried out. Then, they were called individually into the experiment room and underwent a standard 9-point calib ration process before proceeding to the real experiment. Each participants took an average of 15-20 minutes to complete the whole experiment. After the experiment, a standardized test was adapted and carried out to measure the participants' WM capacity using the Auditory Working Me mory subtest fro m the Woodcock Johnson Test of Cognitive Abilities. The experimenter read out a mixed series of words and digits. The participants then rearrange them by first saying the words in order and then the numbers. The task was administered and scored according to manual instructions.

Results
This experiment investigated on the eye movement behaviour as indication of reading fluency during anaphor processing of young adults. Each text was divided into two regions of interest which are the anaphor region and the antecedent region. The eye movement measures that were calculated for the anaphor and antecedent regions were total fixat ion duration and visit count as shown in Table 3 and Table 4 belo w. The eye movement data was also analysed with the score for working memo ry capacity as in Table 5.  Table 3, it can be summarized that there is only a marginal d ifference between typical and atypical anaphor and antecedent regions in mean total fixation durations of the participants. The mean total fixation durations of the participants for both typical antecedents (M=0. 27,SD=0.24) and anaphors (M=0.44,SD=0.34) regions are longer co mpared to atypical antecedents (M=0.40,SD=0.27) and anaphors (M=0.20,SD=0.20) regions and therefore, indicating lo wer read ing fluency. Hence, it can be concluded that the participants did not show difficu lty processing the atypical antecedents which is opposite to that predicted, whereby young adults will exhibit similar eye movement behaviour to adults by producing longer total fixation durations on atypical antecedents than typical antecedents. The data also summarized on the visit count in the anaphor and antecedent regions. The result indicate that the typical antecedents (M=1.63, SD=0.99) and anaphors (M=1.00, SD=0.80) had higher frequency than atypical antecedents (M=1.31, SD=0.84) and anaphors (M=0.81, SD=0.74). It can be presumed that the part icipants did not have much difficulty in processing the atypical antecedents than the typical antecedents.

Eye Movement Behaviour as the Indication of Reading Fluency during Anaphor Processing of Young Adults
Overall, the data summarized that there is only a marginal difference between typical and atypical anaphor and antecedent regions in mean total fixation durations of the participants. The data demonstrated that the mean total fixation durations for both typical antecedents and anaphors regions are longer co mpared to atypical antecedents and anaphors regions and therefore, indicating lower reading fluency. The result opposed to the prediction made whereby atypical antecedents will produce longer total fixat ion durations due to low semantic overlap to its anaphor and disrupt reading fluency. In addition, it is also in contrast to previous studies made among adult readers (Rosch, Mervis, Gray, Johnsen & Boyes-Braem, 1976;McKoon & Ratcliff, 1980;Duffy & Rayner, 1990;Duffy & Rayner, 1990) wh ich revealed that the eye movement behaviour of adult readers during anaphor processing is longer fixation durations when the antecedent has a low typicality of the anaphor category. However, the results fro m this experiment demonstrate inverse typicality effect which is consistent with findings fro m a study by Joseph, Bremner, Liversedge and Nation (2015) on anaphoric processing among children. They mentioned that the possibility of this could be due to the high demand of resource during processing of atypical antecedents which can also be a factor among the participants of this experiment. One other possible justification is that the data from this experiment showed that there is a considerable amount of word skipping on average in the text wh ich contains atypical antecedents among the participants. Therefore, the total fixat ion durations of the antecedent regions will be directly influenced by the frequency of word skipping for that particular region. Next , the data also summarized on the visit count in the anaphor and antecedent regions by looking at the frequency of visit count of the area of interest. This particular met ric was chosen instead due to the unavailability of regression data in the system. The visit duration count is similar to regression whereby it provides data on the number of visits made to that particular area of interest. The result indicate that the participants made more visit count for typical antecedents than atypical antecedents. This is consistent with the findings fro m Joseph, Bremner, Liversedge and Nation (2015) which stated that majority of the children had high frequency of regressions before successfully connecting the atypical antecedents to its anaphor as compared to typical antecedents. They explained that anaphoric processing among children is progressive whereby when the linking process is more difficult, as in the atypical condition, evidence of processing emerges later in reading which can be observed at the post-anaphor region. Therefore, further research can be done on the post-anaphor region to verify the eye movement behaviour of young adults to the behaviour of children and substantiate the findings of this experiment. Moreover, they stated that a higher frequency of regressions may denote an early bonding stage of anaphoric processing. Nevertheless, the participants may also produced lower frequency of visit counts for atypical antecedents because of their word skipping behaviour.

Relationships between Eye Movement Behaviour and WM among Young Adults
To summarize, participants with a larger WM capacity, indicated by higher score of auditory working memo ry subtest, has lower mean total fixat ion durations on both typical and atypical antecedents which also implies better reading fluency. The result wh ich shows that participants with larger WM capacity process antecedents faster is consistent with the pred iction made. It can be seen that participants 1 and 2 took shorter time to resolve the atypical antecedent to its anaphor as they have larger WM capacity. This means they have better resources and therefore are able to link antecedents to its anaphor in a shorter time. Thus, this support the theory (Perfetti, 2007) that individuals with larger WM capacity will be able to process an atypical antecedent faster. Individuals, both children and adult with better reading fluency and large WM capacity resolve the anaphors as soon as their eyes left the post-anaphor region which possibly could be due Eye Tracking in M easuring the Relation of Anaphor Processing, Reading Fluency and Working M emory (WM ) Capacity among University Students to fast anaphor-antecedent lin king with high WM capacity, broad vocabulary and strong semantic links within their lexicons. Nevertheless, the results fro m this experiment do not match the study by Joseph, Bremner, Liversedge and Nation (2015), where the results from the antecedent region show no reliable effects of typicality. Furthermore, the children with good reading fluency and smaller WM capacity demonstrated longer reading times on atypical than typical antecedents. They expressed that the children with s maller WM capacity may require mo re time to process the atypical anaphor completely. Hence, further study is necessary in order to form a firm conclusion on the relation between WM capacity and eye movement behaviour, particularly the total fixat ion durations, among young adults, given the limited size of samp le of this experiment.

Conclusions
In conclusion, this experiment provided an insight on the eye movement behaviour of young adults during anaphor processing which imp lies the fluency of reading and also the relation between eye movements and WM capacity. Overall, the result show that the participants demonstrated inverse typicality effect whereby they have longer fixation durations on typical antecedents than atypical antecedents which resembles children's eye movement behaviour instead of adult readers. Besides, the participants with larger WM were able to process and link antecedents to its anaphor faster than those with smaller WM capacity due to larger resources.

Limitations and Future Recommendations
One of the limitations of th is experiment is the population sample. A larger group of sample with diverse population background should be considered for future research. A larger data will aid in establishing the relation of eye movement behaviour during anaphor processing with read ing fluency as well as WM capacity. Furthermore, a larger population sample will p rovide a more reliable and valid data as a reference for other future research. Next, the data fro m this experiment may be inaccurate due to human factor. For examp le, the word skipping behaviour observed among the participants do influence the result, especially on the eye movement behaviour -the total fixat ion durations of atypical antecedents. The participants may have skipped the texts with atypical antecedents as they were familiar with it. Therefore, future research may present the two texts in different slides. Other than that, the environment of the room was not suitable as one of the participants exp ressed after the experiment that their concentration was affected by the loud noise fro m the renovation work nearby. As for the data analysis, the eye-tracking system used lacked metrics such as regression and saccade, so in-depth analysis on the eye movement behaviour could not be done. Lastly, future research can imp lement the automated version of the Auditory Working Memory Test as it will produce better accuracy of data.