Interaction effects of aging, word frequency, and predictability on saccade length in Chinese reading

Introduction

It has been well documented that word processing may vary among young and older adult readers of Western languages such as English and German. Specifically, recent evidence has revealed a subtle decline in word identification among older adults when they read; that is, older adults spend more time fixating on target words, and regress to them more often than younger readers (Kemper, Crow & Kemtes, 2004; Laubrock, Kliegl & Engbert, 2006; Rayner, Castelhano & Yang, 2009; Rayner, Castelhano & Yang, 2010; McGowan et al., 2014; Whitford & Titone, 2017; Paterson, McGowan & Jordan, 2013a; Paterson, McGowan & Jordan, 2013b; Paterson, McGowan & Jordan, 2013c). Studies have also revealed that frequency and predictability effects, which are closely related to how easily any word can be processed, are also impacted by aging. For example, Kliegl et al. (2004) revealed that older readers yielded larger frequency effects, younger readers tended to skip more predictable words, and older adults re-fixated less on such words while reading in German. Rayner et al. (2006) also reported larger frequency effects for native older adult English readers than young adult readers; however, they failed to observe differences in predictability effects between the two age groups. Recently, Steen-Baker et al. (2017) reported a different contextual sensitivity in the regression patterns of older readers. A study conducted by Choi et al. (2017) revealed larger predictability effects on fixation time measures of target words based on age groups. Thus, it seemed that the eye movements of the two age groups might be affected by frequency/predictability factors in slightly different ways.

The visual and linguistic demands of written Chinese are very different from those of alphabetic texts. Even so, researchers have observed a similar difference in eye movement behaviors between young and older readers of Chinese; that is, studies have consistently revealed that older Chinese readers fixate on target words for a longer time than their younger counterparts (Zang et al., 2016). Wang and her collaborators replicated larger frequency and predictability effects on fixation time measures for older readers of Chinese when they read two-character-words than for their young counterparts (Wang et al., 2018a; Wang et al., 2018b; Zhao et al., 2019). However, there were still distinctions based on impact of age on eye movement control between readers of Chinese and alphabetic languages; this implies that older English adult readers demonstrate longer forward eye movements, and skip words more often than their younger counterparts (see, Rayner et al., 2006; Rayner, 2009), although, recent evidence has revealed that there is no difference between the two age groups in terms of skipping rate and saccade length (Choi et al., 2017). Growing evidence on the topic of Chinese reading has indicated that older readers make shorter forward saccades and skip words more infrequently than young adult readers (Wang et al., 2018a; Wang et al., 2018b; Zang et al., 2016; Li et al., 2018). Thus, it seems that older Chinese readers employ a more careful strategy for eye movements, as compared to older readers of alphabetic texts. Therefore, based on the findings of the aforementioned studies, it is explicit that the effect of age on eye movement behavior is both language specific and universal.

As an important aspect of eye movement control during reading, saccade targeting to word seems to differ across Chinese and alphabetic languages. For most alphabetic writing systems, readers project their eyes to a specific location of a word, which is called preferred-viewing location (PVL; Rayner, 1979). It has been confirmed that linguistic characteristics of a word, such as predictability, did not impact the initial landing position of the eyes for target words (Rayner et al., 2001). In contrast, previous studies failed to find reliable evidences that Chinese readers target saccade to PVL during reading. However, evidence has shown that a target word can easily be processed, in that their frequency and predictability modulate the lengths of saccades targeting. Specifically, Liu and his collaborators have confirmed that target-word frequency modulates the lengths of incoming and outgoing saccades for target words, with a shorter saccade to/from low-frequency word (Liu, Reichle & Li, 2015; Liu, Reichle & Li, 2016; Liu et al., 2017a; Liu et al., 2017b); when compared with less predictable target words, highly predictable target words elicited longer saccades (Liu et al., 2018). According to their findings, saccade amplitude is adjusted dynamically using lexical processing. Readers lengthen an impending saccade if the processing of the parafoveal or foveal word is easy (Liu, Reichle & Li, 2015; Liu et al., 2017a; Liu et al., 2017b).

It is known that older Chinese readers adopt a more careful saccade strategy during reading, which may be owing to their poorer word processing, possibly due to a visual decline with age and/or additional visual requirements for the Chinese text (Owsley, 2011; Zhang et al., 2012; Wang et al., 2018a; Wang et al., 2018b; Zang et al., 2016; Li et al., 2018). Previous studies have revealed that word length produced a similar landing position, probability, and duration of eye movement in both young and older adult readers of English (Rayner et al., 2006; Paterson et al., 2015). Von der Malsburg, Kliegl & Vasishth (2015) also observed an insignificant interaction effect of aging with word length on scan-path, whereas a Chinese study recently revealed a lager word length effect on older Chinese readers than their young counterparts both in terms of fixation time and saccade length (Li et al., 2018). As has been noted earlier, growing evidence on how age impacts Chinese reading has consistently shown that older readers yield larger frequency and predictability effects on fixation time (i.e., gaze duration and total reading time). However, little is known about whether or not this impact of age could be categorized into saccades behaviors (i.e., length of impeding incoming/outgoing saccades for target words). Do these two effects of linguistic characteristics on saccades length change with age in Chinese readers? Exploring this would be undeniably helpful for revealing the factors responsible for the effect of age on eye guidance in Chinese readers. In the present study, we focus on how age mediates the effects of word frequency and predictability on the lengths of impending incoming/outgoing saccades for a target word. How is the length of the impending saccade related to parafoveal and foveal processing (Liu, Reichle & Li, 2015; Liu et al., 2018)? Despite research conducted by Liu and his collaborators investigating the individual effects of word predictability and frequency on the length of impending saccades among a group of young adults (Liu, Reichle & Li, 2015; Liu, Reichle & Li, 2016; Liu et al., 2017a; Liu et al., 2017b), questions regarding how these two factors contribute, additionally or interactively, to saccade length, and whether or not age impacts these effects, still persist. Prior literature has suggested the need to predict differences in basic eye movement in terms of age; currently, it suggests that older adults need more time, fixate longer, and exhibit short saccade lengths during sentence comprehension as compared to young adults. However, we have mainly focused on the different effects of frequency/predictability or their interaction on the length of impending saccades for old adult readers. Encouraged by previous studies, we manipulated predictability and frequency orthogonally. The impact of age on lexical processing produced larger effects of frequency/predictability or their interactions on the fixation time of older readers. As the focus of present study, it predicates reliable interaction effects of age with two linguistic characteristics of words (i.e., frequency & predictability) and their interaction on the length of impending saccades.

Methods

Results

Discussion

The present study assessed how the effects of frequency and predictability, and their interaction in eye movement behavior, especially in case of saccades, on Chinese readers were mediated by age. This study contributes to the growing evidence on the impact of age on word processing in several ways. First, the results of the word-level measures analyses replicated those of previous studies, which state that older readers need more fixation time for comprehending text. However, we could not determine the effect of age on average saccade length, which is also consistent with the scope of the findings of a recent study by Zhao et al. (2019). Second, with regards to the fixation time measures on target words, older readers fixated on these words for a longer time and re-fixated on them more often than younger readers, which also mirrors previous findings of studies on both Chinese and alphabetic text reading (Kliegl et al., 2004; Rayner et al., 2006; Rayner, 2009; Rayner, Castelhano & Yang, 2010; Kemper & McDowd, 2006; Kemper & Liu, 2007; Paterson, McGowan & Jordan, 2013a; Paterson, McGowan & Jordan, 2013b; Paterson, McGowan & Jordan, 2013c; McGowan et al., 2014; Zang et al., 2016; Wang et al., 2018a; Wang et al., 2018b). Third, we replicated age-related differences in skip measure; specifically, older Chinese readers skipped target words less often than young adult readers (Zang et al., 2016; Wang et al., 2018a; Wang et al., 2018b). Moreover, we observed the reliable impact of age on frequency and predictability, and their interaction effects on certain eye movement measures, which notably reveal eye movement control and word processing in Chinese reading.

For the interaction effects involving age with the linguistic characteristics of predictability and frequency, our findings further suggest that age promotes the effects of frequency. Previous studies have revealed larger frequency effects on fixation time measures, such as first fixation duration, gaze duration, and total reading time, in alphabetic text reading (Kliegl et al., 2004; Rayner et al., 2006). Conversely, Wang and her collaborators failed to report any larger frequency effects on older Chinese readers when analyzing log-transformed fixation time; however, they reported larger frequency effect on raw fixation time measures, such as gaze duration and total reading time, which was owing to older readers demonstrating very long fixation durations for low frequency words (Wang et al., 2018a; Wang et al., 2018b). The interaction effect between age group and frequency is extremely difficult to examine. Analyses on log-transformed data are likely to distort the ratio scale properties, which can lead to effects impacted by distribution tail to go unnoticed (Wagenmakers et al., 2012; Lo & Andrew, 2015); therefore, as compared to log-transformed fixation data, analyzing raw data is more convenient for detecting such interactions. However, we observed a larger frequency effect on log-transformed gaze duration and go-past time for older Chinese readers, which may partially be owing to a much more potent manipulation of word frequency.

Larger predictability effects on older Chinese readers were observed in fixation time measures, which suggest that older readers make greater use of context information for word identification than their young counterparts (Zhao et al., 2019). Our findings also help understand the role of age in utilizing context cues to promote eye movement and word processing during reading. We could not determine a reliable interaction effect of age and predictability on fixation measures in the first fixation duration and go-past time, but marginal or reliable interaction effects on raw data of gaze duration and total reading were observed. Interaction effects involving older groups with other variables are not easy to examine, but raw data help in detecting such interactions. Therefore, our findings have contributed, at least partly, to the evidence that older readers make greater use of context to identify words during reading (Choi et al., 2017; Zhao et al., 2019). Our results, of course, also further elaborate on how predictability impacts word identification in its interaction with frequency. Despite many observations by studies on eye tracking regarding the contributions of predictability and frequency to fixation durations while reading alphabetic text (Kennedy et al., 2013; Rayner et al., 2004), questions regarding whether these two factors contribute additionally, or interactively, to eye movement control in Chinese reading, and whether age impacts their interactions, still persist. Our results demonstrate the additional contributions of predictability and frequency on all measures of fixation for both age groups. Thus, our findings suggest that, at least for older readers, interaction effects of frequency and predictability on fixation time appears to be similar across scripts and age groups.

The probability of skipping and re-fixating on target words revealed some differences between Chinese and alphabetic text. It was found that word skipping rate was insensitive to the interactions of age and frequency and that of age and predictability for Chinese readers, which is a finding consistent with that of previous studies (Wang et al., 2018a; Zhao et al., 2019). However, this finding is inconsistent with that of studies which focused on reading of alphabetic text. Specifically, in English reading, age also promotes the frequency effect on skipping rate of the target word, but does not interact with predictability (Rayner et al., 2006); however, in German reading, it was found that high predictability increased young adult readers’ probability of skipping target words (Kliegl et al., 2004). We also observed no significant interactions between aging, frequency, and predictability for re-fixating rate of target words. This null finding was only partly consistent with the findings of studies on reading of alphabetic text. In English reading, there is no indication that older readers’ re-fixating rate of words might be more likely to be impacted by frequency and predictability. However, in German reading, Kliegl et al. (2004) found that highly predictable words decreased older readers’ probability of multiple fixations on target words. The results of the present study showed that the effects of frequency and predictability on skipping of and re-fixating rate of target words were generally insensitive to age. Thus, it demonstrates a different pattern of the effects of age on skipping and re-fixating across Chinese and alphabetic text reading.

Perhaps the most important contribution of the present study is that it helps understand how text processing affects saccades targeting. We suppose that the reason we could not determine reliable effects of frequency and predictability, and their interaction, on both the length of incoming/outgoing saccades (although reliable effects of frequency and predictability were observed on log-transformed data in terms of length of outgoing saccades, with a shorter outgoing saccade when predictability or frequency was low), is that each Chinese character subtended a larger visual angle than that used in the study by Liu and his colleagues (1.32° vs 1°, Liu, Reichle & Li, 2015; Liu, Reichle & Li, 2016; Liu et al., 2017a; Liu et al., 2017b; Liu et al., 2018), which may eliminate some effects of parafoveal word processing. Nevertheless, for both raw and log-transformed length measures, a reliable interaction effect of age and predictability on incoming saccade length was observed. Specifically, older readers yielded a different predictability effect on incoming saccade length. Previous research found that young Chinese adult readers could use predictability information for promoting the processing of parafoveal characters of the target word (Su, Liu & Cao, 2016). The predictability of words presumably facilitates the preprocessing of these words (Balota, Pollatsek & Rayner, 1985; White, Rayner & Liversedge, 2005a; Schotter et al., 2015), and the interaction effect of age and predictability on incoming saccade length suggests that older readers either experience a decline in the preprocessing of low predictability target words or they make greater use of context to identify words.

Reliable interaction effects of age and frequency on fixation time (i.e., gaze duration, go-past time, and total reading time) indicated that older people were more susceptible to foveal load manipulated by word frequency (see Henderson & Ferreira, 1990; Henderson & Ferreira, 1990; Kennison & Clifton Jr, 1995; Kliegl, Nuthmann & Engbert, 2006; White, Rayner & Liversedge, 2005b; Kliegl, 2007, for word frequency manipulating foveal load). Corresponding to this, reliable interaction effects of age and frequency on outgoing saccades length was observed. We found a negative frequency effect on outgoing saccades length for older readers, which may be a result of their longer fixation on the infrequency of target words. It should be noted that foveal load affects information that can be extracted from the parafoveal; moreover, word frequency modulated outgoing saccades length also reflect the functional role of parafoveal processing in determining saccade targeting (Liu, Reichle & Li, 2015). Thus, the reliable interaction effects of age and frequency on outgoing saccades length also suggest a decline of word processing in determining eye movement during reading. This discrepancy observed in both older and young adult readers when considering the effect of predictability on incoming saccade length and effects of frequency on outgoing saccades length may suggest decline in word processing of older Chinese readers. Future studies must investigate why age impacts the effects of frequency on outgoing saccades length, which is a measure of relatively later word processing, rather than on incoming saccade length. Our findings suggest that word identification in Chinese text depends on readers first identifying characters that might combine to form a word (Ma, Li & Rayner, 2015; thus, frequency effects of two-character words may emerge slowly in Chinese reading.

Taken together, the interaction of age with frequency and predictability affect the amplitude of saccades, which is an important finding as it suggests that age impacts word processing of, possibly both, in parafoveal and foveal vision. This study particularly aimed to assess the impact of age on interactions involving word frequency and predictability in Chinese reading. The absence of interaction effects between word frequency and predictability on eye movements measures in young adult readers indicated that word frequency and predictability impacted a different stage of word processing; this finding is consistent with previous findings of studies on reading alphabetic text (Ashby, Rayner & Clifton, 2005; Gollan et al., 2011; Hand et al., 2010; Slattery, Staub & Rayner, 2012). Conversely, the reliable interaction effect of the three variables on outgoing saccades length produced an interaction of frequency and predictability that was positive for older readers and negative for young adult readers (notably, the reliable effects of a three-way interaction, namely Predictability × Frequency × Group, on both raw and log-transformed outgoing saccades length were observed). Both parafoveal and foveal lexical processing difficulty is used to determine how far one’s eyes would move during reading (Liu, Reichle & Li, 2015; Liu, Reichle & Li, 2016; Liu et al., 2017a; Liu et al., 2017b; Liu et al., 2018). Significant interaction effects of Group × Predictability on incoming saccade length suggests a decline in parafoveal lexical processing, while reliable interaction effects of Group × Frequency on outgoing saccades length suggests a decline in foveal lexical processing for older readers. Therefore, this three-away interaction effect on outgoing saccades length may result from a decline in both parafoveal and foveal lexical processing for old readers.

Several limitations of the present study must be acknowledged. First, the eye-tracker we used has a relatively low spatial and temporal resolution than others. Although this may not be a fatal defect, upon comparison with the methods used in previous studies, we believe that this needs be acknowledged. Second, comprehension scores were lower for older adult readers. It is unlikely that older adult readers did not comprehend sentences because they were all retired teaching staff from the university; they were not very old. Unfamiliarity with computer equipment could have also affected their manual responses. The final limitation is related to validity of the findings. The launch site of impeding saccade may systematically vary across young and older adults, results of saccade length are also likely be systematically affected by higher re-fixation rates of the older adults. But we focused on the interactions of predictability and frequency with age on these two kind of impeding saccade length, the results of reliable interactions must be, at least partly, resulting from these systematical differences. We tested our hypothesis multiple times, and came to the conclusion that using Bonferroni corrections was the most effective method to validate our results (Von der Malsburg & Angele, 2016). Considering our hypothesis that age interacts with linguistic characteristics based on saccade length in Chinese reading, we observed three interactions for each of the two saccade length measures; therefore an alpha of 0.0083 is appropriate in this context (new α =0.05∕6 ≈ 0.0083). Barring the three-away interaction effects on log-transformed outgoing saccades length, and Group × Predictability interaction effect on raw data and incoming saccade length, we have interpreted various null results of interactions in our discussion. Although the extent to which extent statistic power has been reduced by Bonferroni corrections is unclear, a reliable three-way interaction may be owing to the two interactions of age with frequency and with predictability. Discussing these “null” results have has helped clarify the interactions effects of age on impeding saccade length.

The present study, and other similar studies, may contribute to a better understanding of language specific characters and how reading behaviors change with age, specifically in readers of the Chinese text. In summary, we found that effects related to age, word frequency, and predictability appear to be similar across alphabetic and non-alphabetic text, when assessed with measures of fixation time. Measures of saccade length maybe the most informative for understanding how age impacts word processing in Chinese readers. Specifically, a unique pattern of results was observed in terms of measures of incoming and outgoing saccades length. Thus, present study provides evidence suggesting that these age-related effects differ qualitatively across alphabetic text reading. Previous research has already elaborated on this qualitative difference. Paterson et al. (2015) found that young and older adult readers of alphabetic text demonstrated similar patterns of landing position on both short and long words. This finding suggests that eye guidance is preserved in older adult readers during alphabetic text reading. We suppose that this was due to differences in the behaviors of alphabetic text and Chinese text readers. Specifically, reading alphabetic text requires the use of visual cues for guidance on where to move one’s eyes, and requires the reader to focus on a target word regardless of word length (Rayner, 1998; Rayner, 2009). Whereas, reading Chinese text requires eye movements, that is saccade targeting, to be modulated by the parafoveal and foveal lexical processing (Liu, Reichle & Li, 2015; Liu, Reichle & Li, 2016; Liu et al., 2017a; Liu et al., 2017b). Our findings demonstrate that age interacts with lexical processing to effect saccade length. Ultimately, observing effects of age on eye movements while reading in Chinese may help in further understanding of the characteristics that underlie changes in reading behaviors over time.

Conclusions

Results of present study suggest that the word processing function of older Chinese readers in terms of saccade targeting declines with age.

Supplemental Information