Treffer: Food word processing in Chinese reading: A study of restrained eaters.

Food-related attentional bias refers that individuals typically prioritize rewarding food-related cues (e.g. food words and food images) compared with non-food stimuli; however, the findings are inconsistent for restrained eaters. Traditional paradigms used to test food-related attentional bias, such as visual probe tasks and visual search tasks, may not directly and accurately enough to reflect individuals' food-word processing at different cognitive stages. In this study, we introduced the boundary paradigm to investigate food-word attentional bias for both restrained and unrestrained eaters. Eye movements were recorded when they performed a naturalistic sentence-reading task. The results of later-stage analyses showed that food words were fixated on for less time than non-food words, which indicated a superiority of foveal food-word processing for both restrained and unrestrained eaters. The results of early-stage analyses showed that restrained eaters spent more time on pre-target regions in the food-word valid preview conditions, which indicated a parafoveal food-word processing superiority for restrained eaters (i.e. the parafoveal-on-foveal effect). The superiority of foveal food-word processing provides new insights into explaining food-related attentional bias in general groups. Additionally, the enhanced food-word attentional bias in parafoveal processing for restrained eaters illustrates the importance of individual characteristics in studying word recognition.
(© 2023 The British Psychological Society.)

AN0162972099;bjp01may.23;2023Apr11.03:39;v2.2.500

Food word processing in Chinese reading: A study of restrained eaters 

Food‐related attentional bias refers that individuals typically prioritize rewarding food‐related cues (e.g. food words and food images) compared with non‐food stimuli; however, the findings are inconsistent for restrained eaters. Traditional paradigms used to test food‐related attentional bias, such as visual probe tasks and visual search tasks, may not directly and accurately enough to reflect individuals' food‐word processing at different cognitive stages. In this study, we introduced the boundary paradigm to investigate food‐word attentional bias for both restrained and unrestrained eaters. Eye movements were recorded when they performed a naturalistic sentence‐reading task. The results of later‐stage analyses showed that food words were fixated on for less time than non‐food words, which indicated a superiority of foveal food‐word processing for both restrained and unrestrained eaters. The results of early‐stage analyses showed that restrained eaters spent more time on pre‐target regions in the food‐word valid preview conditions, which indicated a parafoveal food‐word processing superiority for restrained eaters (i.e. the parafoveal‐on‐foveal effect). The superiority of foveal food‐word processing provides new insights into explaining food‐related attentional bias in general groups. Additionally, the enhanced food‐word attentional bias in parafoveal processing for restrained eaters illustrates the importance of individual characteristics in studying word recognition.

Keywords: attention allocation; eye movement; food word superiority; restrained eating

INTRODUCTION

Human beings have a typical attentional pattern that prioritizes the rewarding food‐related cues, such as actual food and food words or images, over non‐food cues (Sui et al., [50]; Werthmann et al., [57]). This phenomenon is a cognitive bias known as food‐related attentional bias (Hagan et al., [15]). It is thought that the cognitive bias towards food may be a potential mechanism contributing to unhealthy eating behaviours and obesity (Werthmann et al., [58]).

Restrained eaters (i.e. people who strictly control their food consumption in order to lose or maintain body weight) are usually unsuccessful in their dietary restraint (Herman & Polivy, [16]). Their special cognitive bias towards food (e.g. food‐related attentional bias) may explain such failure (Kong et al., [20]). Prior studies have investigated how individual characteristics, such as restrained eating, modulate food‐related attentional bias. However, research findings on food‐related attentional bias in restrained eaters have been inconsistent (Hollitt et al., [17]; Polivy et al., [40]; Werthmann et al., [59]). Furthermore, Devoto et al. ([10]) have suggested that other individual characteristics, such as BMI (Body Mass Index) and hunger levels, could also modulate individuals' cognitive processing. Soetens and Braet ([47]) reported that overweight young adults have a memory bias for high‐calorie food cues. Werthmann et al. ([58]) found overweight individuals have an approach‐avoidance pattern of attention allocation towards high‐fat food images. However, results have been inconsistent in that some studies have reported that neither BMI nor hunger level can modulate individuals' food‐related attentional bias (Coricelli et al., [9]). The selection of paradigms (e.g. probe detection, Stroop and visual search paradigms) and food‐related cues (e.g. food words, food images and real foods) may have contributed to these inconsistencies (Jiang & Vartanian, [19]; Luo et al., [29]).

We believe that the trend of attentional bias (i.e. approach bias, avoidance bias or no bias) might change with cognitive processing over time. However, the studies described above did not focus on the changing trends of attentional bias at different processing stages. Eye‐tracking techniques can help us understand attentional processing in human beings by recording individuals' eye movements, and it has been widely used in studies of eating‐related behaviour, such as nutrition label processing (Ma & Zhuang, [34]) and food choice (Van Loo et al., [51]). Eye‐tracking techniques can effectively capture participants' attention to food cues, even food cues that occur on TV (Alblas et al., [1]).

In the present study, we used a popular gaze‐contingent paradigm to test restrained and unrestrained eaters' food‐related attentional bias while reading. This paradigm provides a naturalistic condition to investigate the attentional patterns of food‐word recognition, including both foveal and parafoveal food‐word processing. The foveal region (the central fixated region) comprises the central two degrees of the visual field, while the parafoveal region extends from the foveal region to about five degrees on either side of it (Rayner, [43]). Although word recognition mainly occurs in the foveal region, under certain conditions additional information can be extracted using parafoveal vision before the reader fixates on a target word (Schotter, [45]). Foveal word processing reflects the later stage of attention allocation, while parafoveal word processing reveals its early stages (Ma & Li, [30]). Therefore, the exploration of parafoveal and foveal food‐word processing in Chinese reading may enhance our judgement of food‐related attentional bias in the early and the later stages of cognitive processing.

The recognition superiority of food words relative to non‐food words may exist in foveal processing. The processing speed of foveal word recognition is influenced by many word properties in Chinese reading, such as word frequency, predictability and word length. Specifically, words with high frequency, high predictability or short length tend to be fixated on for a shorter time than words with low frequency, low predictability or long length (Li et al., [23]; Li & Pollatsek, [25]). Moreover, the semantic preview benefit in Chinese reading has been observed using the gaze‐contingent paradigm when the semantic‐related words are synonyms (Zhu et al., [66]). For example, the processing of the target word 'nurse' will be facilitated if readers can preview the semantic‐related word 'caregiver'. This facilitation of word recognition might derive from connections among the mental representations of semantic‐related words. According to the spreading‐activation theory of semantic processing (Collins & Loftus, [8]), the human lexicon network consists of a large number of words arranged by word availability (i.e. words with higher availability are processed faster). Subjective experiences with a word may also affect how individuals shape its meaning, which may in turn impact the availability of the word in an individual's mental lexicon and alter its speed when reading.

As food is known to be of great importance for human survival, we posit that the special meaning of food words might contribute to unique mental representations in human lexicon network, which should facilitate foveal food‐word processing. However, popular reading models to date have used word frequency, predictability and word length to predict eye movement control in reading (Li et al., [23]; Li & Pollatsek, [25]). Individual differences, such as personalized word representation, have rarely been considered. Therefore, in the current study we chose to examine foveal food‐word processing in both restrained and unrestrained eaters. We assumed that both restrained and unrestrained eaters would have faster foveal processing when reading food words, which may be represented by a faster or more sensitive response loop in the human brain, since rewarding food cues are strongly related to life.

The superiority of food‐word processing may also exist in parafoveal processing. Parafoveal information processing in reading has been studied for over 30 years, which reflects the early‐stage attention allocation of word recognition. Parallel attention allocation models, such as the SWIFT model, assume that readers can process all words in the perceptual span (Engbert et al., [12]). Serial attention allocation models, such as the E‐Z Reader model, also suggest that readers will occasionally obtain lexical information from the parafovea if the words are sufficiently easily processed (Reichle et al., [44]; Schotter, [45]).

The competition hypothesis was implemented in Chinese reading based on the E‐Z Reader model (Ma et al., [33]). The competition hypothesis indicates that all the words in the perceptual span can be activated. The activation levels of the foveal words (i.e. words being fixated upon) were different from (and normally higher than) those of the parafoveal words. However, the parafoveal words could be recognized when they were easier to process (i.e. when the parafoveal words had relatively high activation levels). Consequently, a competition for cognitive resources occurs between foveal and parafoveal word processing, and the influence of individuals' parafoveal information processing on foveal processing is called the parafoveal‐on‐foveal effect (i.e. the POF effect). Yan and Sommer ([62]) observed a POF effect when emotional words were presented in the parafoveal region. For example, previewing a positive target word, 'assets', in the parafovea would prolong the fixation on the pre‐target word 'many'. However, previewing a neutral target word, 'clients', in the parafovea would not influence the fixation on the pre‐target word 'many'. As the POF effect is caused by the parafoveal information processing, and thus, it potentially indicates early‐stage attention allocation to parafoveal regions. In addition, Veldre and Andrews ([54]) have reported that higher reading ability facilitates the rapid acquisition of semantic information from the parafovea. This finding suggests that eye movement control in reading may interact with individual characteristics (e.g. reading ability). As we mentioned above, individuals' mental representations of some particular words in their lexicon network might influence the extent of parafoveal extraction of these words. Accordingly, we hypothesized that restrained eaters may demonstrate a superior capacity to process food words in the parafovea, and thus, the POF effect would be observed in them when reading sentences with food words.

The present study aimed to explore the superiority of parafoveal and foveal food‐word recognition (relative to word recognition of non‐food words) and the influence of individual differences (i.e. the effect of restrained eating on the processing of food words in the fovea and the parafovea). The study of parafoveal and foveal food‐word processing may provide new insights into attentional bias towards food words. Thus far, the Goal Conflict Model has been widely used to explain food‐related attentional bias (including food‐word recognition) in restrained eaters (Stroebe et al., [48], [49]). According to the Goal Conflict Model, restrained eaters usually experience a conflict between two incompatible goals, namely eating enjoyment and weight control. The goal of eating enjoyment involves a reactivity framework that encourages individuals to approach food, and the goal of weight control involves a regulation framework that prevents them from overeating (Giuliani et al., [14]). Restrained eaters have been found to show an attentional bias towards food cues (i.e. a longer fixation duration on food cues) when the eating enjoyment goal dominated, but there was no bias when the weight control goal dominated (Papies et al., [38]). In fact, the weight control goal has even been found to lead to an avoidance bias from food cues under some conditions (Weng et al., [56]). Pimpini et al. ([39]) found that the approach bias for food cues was also positively correlated with later food intake when participants had a hedonic mindset (i.e. an eating enjoyment goal). Therefore, the Goal Conflict Model uses the two conflicting motivations to explain restrained eaters' food‐related attentional bias. However, the motivation of eating is not the only factor leading to food‐related attentional bias; as mentioned earlier, the superiority of individuals' food‐word recognition might also contribute to food‐related attention bias. The present study tested whether food words could be accessed faster than non‐food words in the parafovea and fovea. This would improve the shortcomings of the Goal Conflict Model in explaining attentional approach or avoidance to food‐related words, if the food‐word processing superiority was found. Specifically, approach or avoidance attentional bias for food words in restrained eaters might simply occur because the food words were more easily processed, rather than being entirely driven by hedonic or health‐related motivations.

To examine the above hypotheses, gaze‐contingent paradigms and eye‐tracking techniques were used in this study. Recording eye movements during sentence reading was deemed more suitable to study individuals' word recognition than using visual search or probe paradigms (Veenstra & Jong, [53]; Werthmann et al., [57]), because such a measurement is more naturalistic and allows the direct capture of participants' attentional patterns at each stage of word processing. Additionally, the gaze‐contingent paradigm has been used for over 30 years in the relevant literature and has been proven to be an effective method for studying attention allocation in reading. In the gaze‐contingent paradigm, an invisible boundary is placed before the target words. Under valid preview conditions, readers can preview the identical target words (i.e. the target words will not change either readers fixate on them or not). Under invalid preview conditions, readers can only preview pseudowords (nonwords) before they fixate on the target words. The preview text will automatically change to the target words if the reader's eyes (i.e. fixation) cross the invisible boundary. Due to saccade suppression (Martin, [35]), participants cannot notice the changes to the manipulated target regions; therefore, this paradigm is natural and similar to regular reading. In this study, the matched target food and non‐food words were embedded in the same sentence frame (see Figure 1). The manipulated two‐character target regions were considered the main regions of interest, and both foveal and parafoveal processing were examined. Eye movement measurements on these target regions were used to examine the foveal processing of food words, while eye movement measurements on the two‐character pre‐target regions were used to test parafoveal processing. Since attention allocation overlaps with word recognition when reading, the study of parafoveal and foveal food‐word processing can shed light on food‐related attentional bias. The specific hypotheses are as follows.

For foveal processing, we proposed the following hypotheses:

H1

Standard preview benefits should be observed for all participants in this gaze‐contingent paradigm. Compared with the invalid condition, the fixation time should be shorter and the skipping probability should be greater in the valid condition, as valid previews should improve the recognition of target words. The validation of H1 would illustrate the validity of the gaze‐contingent paradigm.

H2

Food‐word processing superiority should be observed for both restrained and unrestrained eaters. As mentioned before, food is imperative to living and food words may have priority in the human mental lexicon. Therefore, food words should be generally accessed and recognized faster than non‐food words, regardless of dietary restraint. We hypothesize that food words should be fixated on for a shorter time and be skipped more frequently than non‐food words on the target regions. The validation of H2 would improve the Goal Conflict Model in explaining attentional approach or avoidance of food cues (food words in this study). Restrained eaters may leave food words faster (e.g. shorter fixation durations on food words) because of their faster lexical access, and not because of their motivation for eating.

For parafoveal processing, we proposed the following hypothesis:

H3

The POF effect should be observed in restrained eaters rather than unrestrained eaters. The first fixation duration and the gaze duration on the pre‐target region should be longer in the food condition than in the non‐food condition, especially for restrained eaters, since they are more sensitive to food cues. The validation of H3 would be more consistent with the food‐word processing superiority hypothesis, since a prior study (Ma et al., [33]) has already demonstrated that easier‐to‐process parafoveal words lengthen the processing time of the currently fixated upon region. By contrast, the Goal Conflict Model might predict shorter fixation durations on the pre‐target region, since restrained eaters should approach parafoveal food words faster than non‐food words. Thus, the validation of H3 (i.e. the POF effect for restrained eaters under food‐word preview conditions) would also improve the Goal Conflict Model by demonstrating that the longer latency to food words might be caused by the POF effect rather than by the eating goals.

METHODS

Participants

A power calculation using the G*power 3.1.9.4 software (Faul et al., [13]) showed that each group required at least 24 participants to detect a medium effect size (i.e. f = 0.25 [Cohen, [7]], power = 0.8, alpha = .05). We thus recruited 64 Chinese college students for the formal experiment (Table 1). All participants had normal or corrected‐to‐normal vision and were unaware of the purpose of the study. They were divided into restrained and unrestrained eating groups by their scores on the restraint eating subscale of the Dutch Eating Behavior Questionnaire (DEBQ) (van Strien et al., [52]). The proportion of females in the restrained group was 0.77, and in the unrestrained group was 0.59. This study was conducted in accordance with the Declaration of Helsinki, and all procedures were approved by the ethics committee.

1 TABLE Demographic data of participants in this study.

<table><thead valign="bottom"><tr><th align="left">Group (<italic>N</italic>)</th><th align="left">Restraint score</th><th align="left">BMI</th><th align="left">Hunger level</th><th align="left">Age</th></tr></thead><tbody valign="top"><tr><td>Restrained eaters (30)</td><td>3.56 (0.47)</td><td>23.04 (5.33)</td><td>4.27 (1.73)</td><td>19.4 (1.87)</td></tr><tr><td>Male (7)</td><td>3.47 (0.24)</td><td>23.81 (2.74)</td><td>4.29 (1.49)</td><td>20.7 (2.43)</td></tr><tr><td>Female (23)</td><td>3.59 (0.52)</td><td>22.80 (5.87)</td><td>4.26 (1.80)</td><td>19.0 (1.45)</td></tr><tr><td>Unrestrained eaters (34)</td><td>2.03 (0.44)</td><td>20.17 (2.25)</td><td>4.38 (1.14)</td><td>18.5 (0.70)</td></tr><tr><td>Male (14)</td><td>1.89 (0.38)</td><td>19.87 (1.60)</td><td>4.50 (1.35)</td><td>18.4 (0.82)</td></tr><tr><td>Female (20)</td><td>2.12 (0.45)</td><td>20.37 (2.59)</td><td>4.30 (0.95)</td><td>18.5 (0.59)</td></tr></tbody></table>

1 Note: The unit of the BMI is kg/㎡. The unit of age is years. Standard deviations are given in parentheses.

Materials and design

Word stimuli

We made 40 groups of word stimuli, in which each group contained one food word, one non‐food word and one pseudoword, with all properties matched. These 40 groups of word stimuli were embedded within 40 sentence frames (see Appendix S1). The sentences were assessed by 10 volunteers using a 7‐point Likert scale (from 1 = very unnatural to 7 = very natural) to ensure that the material felt like natural sentences. To test the general effect of food‐word processing, food words were not categorized based on their healthiness or palatability. All word stimuli were evaluated on both subjective (word predictability and word plausibility) and objective (word frequency, character frequency and stroke number) criteria. There were no significant differences in word predictability (as assessed by 20 volunteers) or word plausibility (as assessed by 11 volunteers) between the food words and non‐food words, as measured on a 7‐point Likert scale (from 1 = very implausible to 7 = very plausible). A published lexicon database (Chinese Linguistic Data Consortium, [6]) was used to provide the frequency information for the word stimuli. There were no significant differences in word frequency, character frequency or stroke number among the food words, non‐food words and pseudowords (see Table 2). None of the volunteers who assessed the materials participated in the formal experiments.

2 TABLE Properties of the food words, non‐food words and pseudowords used in this study.

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Food</th><th align="left">Non&#8208;food</th><th align="left">Pseudoword</th><th align="left"><italic>t</italic>/<italic>F</italic></th><th align="left"><italic>p</italic></th></tr></thead><tbody valign="top"><tr><td>Word frequency</td><td>4.1 (0.8)</td><td>4.1 (1.9)</td><td>&#8211;</td><td>&#60;1</td><td>.982</td></tr><tr><td>Word predictability</td><td>0.01 (0.01)</td><td>0.02 (0.01)</td><td>&#8211;</td><td>&#60;1</td><td>.519</td></tr><tr><td>Word plausibility</td><td>4.8 (0.18)</td><td>4.9 (0.20)</td><td>&#8211;</td><td>&#60;1</td><td>.892</td></tr><tr><td>Frequency of the first character</td><td>479 (115)</td><td>531 (118)</td><td>539 (14)</td><td>&#60;1</td><td>.891</td></tr><tr><td>Frequency of the second character</td><td>499 (134)</td><td>616 (161)</td><td>550 (91)</td><td>&#60;1</td><td>.837</td></tr><tr><td>Stroke number of the first character</td><td>8.7 (0.5)</td><td>9.3 (0.5)</td><td>8.6 (0.3)</td><td>&#60;1</td><td>.535</td></tr><tr><td>Stroke number of the second character</td><td>8.8 (0.6)</td><td>9.0 (0.6)</td><td>9.2 (0.5)</td><td>&#60;1</td><td>.836</td></tr></tbody></table>

2 Note: Word frequency is measured by occurrences per million words. Standard errors are given in parentheses.

Experimental design

The formal experiment was a 2 (preview condition: valid and invalid) × 2 (word condition: food and non‐food) within‐subject design. The gaze‐contingent boundary paradigm was implemented. The preview words were replaced with the target words when the participant's eyes crossed an invisible boundary (i.e. the red line in Figure 1) set before (i.e. to the left of) the target word region (i.e. the region of the underlined two characters in Figure 1). In the two valid preview conditions (included food word and non‐food word conditions), participants saw the valid preview words that were identical with the target words before their eyes crossed the invisible boundary. However, in the two invalid conditions, participants saw the invalid preview pseudowords.

Questionnaires

The Dutch Eating Behavior Questionnaire (DEBQ) is a 33‐item questionnaire that comprises three subscales: restrained eating (10 items), emotional eating (13 items) and external eating (10 items) (van Strien et al., [52]). All 33 items of the DEBQ were used and assessed with a 5‐point Likert scale (from 1 = never to 5 = very often), but only the restraint eating subscale was used to classify restrained and unrestrained eaters in this study. The Cronbach's a was 0.87 for the restraint eating subscale. Restrained eaters were defined as having a restraint score higher than 3, while unrestrained eaters were those who scored lower than 3 (Hollitt et al., [17]). The restraint scores were significantly higher for restrained eaters (M = 3.56, SE = 0.08) than for unrestrained eaters (M = 2.02, SE = 0.08), p < .001. Participants' hunger status was assessed by a 7‐point Likert scale (from 1 = very hungry to 7 = not at all hungry), and the hunger level was not significantly different between the restrained (M = 4.26, SE = 0.32) and unrestrained eaters (M = 4.38, SE = 0.19), t (62) < 1.

Apparatus

Each sentence was presented to participants on a 24‐inch LCD monitor with a resolution of 1920 × 1080 pixels and a refresh rate of 144 Hz. All sentences were displayed in Song 24‐point font in black (RGB: 0, 0, 0) on a grey background (RGB: 128, 128, 128). Participants' eyes were positioned about 62 cm from the computer monitor. Based on this setup, each character subtended a visual angle of about 0.8°. We used the Eyelink 1000 plus (SR Research Ltd.) eye tracker to track participants' eye movements, with the sampling rate set to 1000 Hz. A chin rest was used to minimize head movements. Participants read the sentences binocularly, but eye movements were recorded only for the right eye. Similar apparatus, parameters and procedures have also been used in previous studies when investigating other reading topics such as word length effects during Chinese reading (Ma et al., [32]).

Procedure

Upon arrival at the laboratory, participants were given brief instructions about the experiment and the apparatus. They then completed the DEBQ and hunger scales and participated in the eye‐tracking experiment.

The eye tracker was initially calibrated with a 3‐point grid and could be recalibrated during the experiment as necessary. The maximum error of validation was 0.5° in visual angle. There were 10 practice sentences, 40 experimental sentences and eight comprehension questions in the formal experiment. The 10 practice sentences were presented first, after which the 40 experimental sentences and eight comprehension questions were presented in random order. The experimental sentence appeared only if participants successfully fixated on the one‐character‐sized black box at the first character position of each sentence. Participants pressed a response button to move forward after reading a sentence or answering a comprehension question. They were asked to read each sentence silently, and a short break was allowed when needed. The experiment took about 30 min in total to complete, and all participants were rewarded with 20 RMB after completing the experiment.

RESULTS

The accuracy results of the comprehension questions (M = 0.895, SD = 0.094) indicated that all participants understood the sentences well. Fixation durations smaller than 80 ms or larger than 1000 ms were removed from the data analysis (about 1% of the raw data), because these fixations were too short or too long to be typical fixations (Ralph‐Nearman & Filik, [42]). The two‐character target and pre‐target regions were the regions of interest (ROIs) in this study. We reported the following five eye movement measurements based on the ROI: first fixation duration (FFD, the time of the first fixation on the ROI in the first‐pass reading); gaze duration (GD, the sum of all fixation durations on the ROI in the first‐pass reading before moving to other regions); total time (TT, the sum of all fixation durations on the ROI); fixation count (FC, the number of fixations on the ROI during first‐pass reading, not counting instances of skipping); and skipping probability (SKIP, the probability that the ROI was skipped on first‐pass reading).

The data were analysed using a linear mixed‐effects model (LMM) for continuous variables and a generalized mixed‐effects model for binary variables (Baayen et al., [2]; Jaeger, [18]). All statistical models included random slopes for participants (subject) and items (trial item); the random effect variances of these models are also reported in Appendix S2 as suggested by Meteyard and Davies ([36]).

To test the experimental hypotheses, we constructed the model with interaction for each eye movement measure. All model results (see Tables 4, 6, 8 and 10) are described in the main texts. Given that prior studies did not reach a consistent conclusion on the relationship between individuals' food‐word recognition and their BMI (Coricelli et al., [9]; Devoto et al., [10]; Soetens & Braet, [47]; Werthmann et al., [58]), one can conclude that BMI might be a potential factor that independently influenced individuals' food‐word recognition in this study. The average BMI of restrained eaters (M = 23.03, SE = 0.98) was significantly higher than that of unrestrained eaters (M = 20.16, SE = 0.39), t (62) = 2.823, p = .006. The Pearson correlation between participants' restraint scores and BMI was significant, r = .321, p < .01. Thus, we included BMI as a fixed factor in the data analysis when restrained eating was analysed and the variance inflation factor (VIF) was reported (see Tables 6 and 10). The BMI was excluded if the model could not converge with the BMI (i.e. models of the skipping probability on the target and pre‐target region for restrained and unrestrained eaters in the valid preview condition; see Table 10).

The Lme4 package (version 1.1‐12, Bates et al., [4]) was used for data analysis in the R environment (R Core Team, [41]). The fixation duration was log‐transformed to meet LMM assumptions, and the analyses of log‐transformed durations yielded similar results to the untransformed analyses. The p‐values were estimated using the lmerTest package (version 2.0‐33, Kuznetsova et al., [22]). All the data and code used in the present study are available from https://osf.io/9d3b4/.

Target region: Superiority of foveal food‐word processing

The detailed eye movement measurements for the target region are displayed in Tables 3 and 5. We reported all models with interactions for target region analyses (Fixed factors: preview condition, word category, interactions between the preview condition and the word category; Random factors: subject and trial item). However, all interactions between the preview and word conditions were not significant for all five eye movement measurements, ts < 1 (see Table 4). Nonetheless, preview benefits were observed for all eye movement measurements: the target word regions were fixated on for less time and skipped more often in the valid preview conditions (see Table 3). Therefore, H1 is supported, which means that the gaze‐contingent paradigm was successfully performed. More importantly, the superiority of food‐word processing superiority was supported by the fixation measurements in that food words were fixated on for a significantly shorter time than non‐food words, thus validating H2 (see Tables 3 and 4).

3 TABLE Eye movement measurements on the target region

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Valid preview</th><th align="left">Invalid preview</th></tr><tr><th align="left">Food</th><th align="left">Non&#8208;food</th><th align="left">Food</th><th align="left">Non&#8208;food</th></tr></thead><tbody valign="top"><tr><td>First fixation duration</td><td>223 (4.2)</td><td>233 (3.8)</td><td>261 (5.7)</td><td>269 (6.5)</td></tr><tr><td>Gaze duration</td><td>248 (5.6)</td><td>271 (6.4)</td><td>327 (10.3)</td><td>343 (11.1)</td></tr><tr><td>Total time</td><td>401 (12.7)</td><td>418 (11.2)</td><td>491 (13.8)</td><td>517 (14.9)</td></tr><tr><td>Fixation count</td><td>1.95 (0.06)</td><td>2.08 (0.06)</td><td>2.26 (0.07)</td><td>2.37 (0.08)</td></tr><tr><td>Skipping probability</td><td>0.08 (0.01)</td><td>0.07 (0.01)</td><td>0.03 (0.01)</td><td>0.02 (0.01)</td></tr></tbody></table>

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Estimate</th><th align="left"><italic>SE</italic></th><th align="left"><italic>t</italic>/<italic>z</italic></th><th align="left"><italic>p</italic></th></tr></thead><tbody valign="top"><tr><td align="left">First fixation duration</td></tr><tr><td>(Intercept)</td><td>5.361</td><td>0.021</td><td>263.64</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>0.137</td><td>0.018</td><td>7.37</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.037</td><td>0.018</td><td>2.01</td><td>.044</td></tr><tr><td>Preview&#160;: word</td><td>&#8722;0.007</td><td>0.026</td><td>&#8722;0.29</td><td>.767</td></tr><tr><td align="left">Gaze duration</td></tr><tr><td>(Intercept)</td><td>5.433</td><td>0.028</td><td>194.29</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>0.245</td><td>0.023</td><td>10.31</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.071</td><td>0.023</td><td>2.99</td><td>.002</td></tr><tr><td>Preview&#160;: word</td><td>&#8722;0.025</td><td>0.033</td><td>&#8722;0.77</td><td>.443</td></tr><tr><td align="left">Total time</td></tr><tr><td>(Intercept)</td><td>5.902</td><td>0.041</td><td>144.89</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>0.263</td><td>0.031</td><td>8.49</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.061</td><td>0.031</td><td>1.97</td><td>.049</td></tr><tr><td>Preview&#160;: word</td><td>0.011</td><td>0.043</td><td>0.24</td><td>.808</td></tr><tr><td align="left">Fixation count</td></tr><tr><td>(Intercept)</td><td>1.928</td><td>0.077</td><td>25.26</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>0.308</td><td>0.059</td><td>5.18</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.107</td><td>0.059</td><td>1.81</td><td>.071</td></tr><tr><td>Preview&#160;: word</td><td>&#8722;0.007</td><td>0.084</td><td>&#8722;0.09</td><td>.924</td></tr><tr><td align="left">Skipping probability</td></tr><tr><td>(Intercept)</td><td>&#8722;2.761</td><td>0.222</td><td>&#8722;12.39</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>&#8722;0.941</td><td>0.259</td><td>&#8722;3.63</td><td>&#60;.001</td></tr><tr><td>Word</td><td>&#8722;0.211</td><td>0.216</td><td>&#8722;0.97</td><td>.329</td></tr><tr><td>Preview&#160;: word</td><td>&#8722;0.343</td><td>0.408</td><td>&#8722;0.84</td><td>.399</td></tr></tbody></table>

We further analysed whether restrained eaters had advantages in food‐word processing in the valid preview condition (Fixed factors: word category, restrained eating, interactions between the word category and restrained eating, BMI; Random factors: subject, trial item; see Tables 5 and 6). The interactions between the word and restrained eating conditions were not significant for all five eye movement measures (see Table 6), ts <1.35, ps > .177, indicating that restrained eaters did not show significant differences in foveal word processing from unrestrained eaters. The main effects of the word condition approached significance for the first fixation duration (p = .051). Food words were fixated upon for less time than non‐food words in the valid preview condition. Therefore, H2 is supported, as the superiority of foveal food‐word processing generally existed for both restrained and unrestrained eaters mainly on the early eye movement measure (FFD). Additionally, BMI was included as a fixed factor in all models, but it did not reach significance in all of them, ts <1.39, ps > .171. The statistical results for the variance inflation factor indicated that none of the models with BMI were affected by multicollinearity (VIFs < 1.5).

5 TABLE Eye movement measures on the target region for restrained and unrestrained eaters in valid preview conditions.

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Restrained eaters</th><th align="left">Unrestrained eaters</th></tr><tr><th align="left">Food</th><th align="left">Non&#8208;food</th><th align="left">Food</th><th align="left">Non&#8208;food</th></tr></thead><tbody valign="top"><tr><td>First fixation duration</td><td>220 (5.3)</td><td>236 (5.4)</td><td>226 (6.5)</td><td>230 (5.4)</td></tr><tr><td>Gaze duration</td><td>246 (6.4)</td><td>268 (8.8)</td><td>251 (8.9)</td><td>272 (9.2)</td></tr><tr><td>Total time</td><td>384 (16.5)</td><td>401 (12.3)</td><td>416 (18.8)</td><td>433 (17.8)</td></tr><tr><td>Fixation count</td><td>1.88 (0.09)</td><td>2.01 (0.09)</td><td>2.02 (0.08)</td><td>2.14 (0.09)</td></tr><tr><td>Skipping probability</td><td>0.09 (0.02)</td><td>0.09 (0.02)</td><td>0.08 (0.02)</td><td>0.06 (0.01)</td></tr></tbody></table>

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Estimate</th><th align="left"><italic>SE</italic></th><th align="left"><italic>t</italic>/<italic>z</italic></th><th align="left"><italic>p</italic></th></tr></thead><tbody valign="top"><tr><td align="left">First fixation duration</td></tr><tr><td>(Intercept)</td><td>5.401</td><td>0.112</td><td>47.94</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.109</td><td>0.056</td><td>1.95</td><td>.051</td></tr><tr><td>Restrained eating</td><td>0.012</td><td>0.036</td><td>0.34</td><td>.737</td></tr><tr><td>BMI</td><td>&#8722;0.003</td><td>0.003</td><td>0.73</td><td>.471</td></tr><tr><td>Word&#160;: Restrained eating</td><td>&#8722;0.047</td><td>0.034</td><td>&#8722;1.35</td><td>.177</td></tr><tr><td align="left">Gaze duration</td></tr><tr><td>(Intercept)</td><td>5.568</td><td>0.137</td><td>40.48</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.098</td><td>0.073</td><td>1.33</td><td>.184</td></tr><tr><td>Restrained eating</td><td>0.002</td><td>0.044</td><td>0.04</td><td>.971</td></tr><tr><td>BMI</td><td>&#8722;0.006</td><td>0.004</td><td>&#8722;1.38</td><td>.172</td></tr><tr><td>Word&#160;: Restrained eating</td><td>&#8722;0.017</td><td>0.046</td><td>&#8722;0.38</td><td>.705</td></tr><tr><td align="left">Total time</td></tr><tr><td>(Intercept)</td><td>5.741</td><td>0.226</td><td>25.31</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.038</td><td>0.104</td><td>0.37</td><td>.708</td></tr><tr><td>Restrained eating</td><td>0.063</td><td>0.072</td><td>0.87</td><td>.383</td></tr><tr><td>BMI</td><td>0.003</td><td>0.007</td><td>0.39</td><td>.692</td></tr><tr><td>Word&#160;: Restrained eating</td><td>0.016</td><td>0.064</td><td>0.26</td><td>.793</td></tr><tr><td align="left">Fixation count</td></tr><tr><td>(Intercept)</td><td>1.491</td><td>0.426</td><td>3.495</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.079</td><td>0.191</td><td>0.42</td><td>.678</td></tr><tr><td>Restrained eating</td><td>0.161</td><td>0.136</td><td>1.18</td><td>.239</td></tr><tr><td>BMI</td><td>0.011</td><td>0.014</td><td>0.72</td><td>.473</td></tr><tr><td>Word&#160;: Restrained eating</td><td>0.026</td><td>0.119</td><td>0.22</td><td>.827</td></tr><tr><td align="left">Skipping probability</td></tr><tr><td>(Intercept)</td><td>&#8722;0.236</td><td>1.465</td><td>&#8722;10.16</td><td>.872</td></tr><tr><td>Word</td><td>0.001</td><td>0.681</td><td>0.01</td><td>.98</td></tr><tr><td>Restrained eating</td><td>&#8722;0.478</td><td>0.419</td><td>&#8722;1.14</td><td>.255</td></tr><tr><td>BMI</td><td>&#8722;0.088</td><td>0.054</td><td>&#8722;1.62</td><td>.106</td></tr><tr><td>Word&#160;: Restrained eating</td><td>&#8722;0.158</td><td>0.441</td><td>&#8722;0.36</td><td>.719</td></tr></tbody></table>

Pre‐target region: Superiority of parafoveal food‐related word processing for restrained eate...

The detailed measurements of eye movements in the pre‐target region are displayed in Tables 7 and 9. Models with interactions are reported in Table 8 (Fixed factors: preview condition, word category, interactions between the preview condition and the word category; Random factors: subject, trial item). The results of the first fixation duration and the gaze duration were the critical indexes for the POF effects as they only included fixations that occurred before the eyes made a saccade into the later region (i.e. the target region). However, there were no significant differences between the valid and invalid preview conditions for the first fixation duration, the gaze duration or the skipping probability, ts < 1. These results reveal that the preview conditions did not affect the processing of pre‐target regions. Furthermore, the main effect of word category showed a significant difference between the food and non‐food word conditions in the first fixation duration and a marginally significant difference in the gaze duration (see Table 8). These findings support hypothesis H3, which concerns individuals' early‐stage attentional bias towards food words.

7 TABLE Eye movement measurements on the pre‐target region.

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Valid preview</th><th align="left">Invalid preview</th></tr><tr><th align="left">Food</th><th align="left">Non&#8208;food</th><th align="left">Food</th><th align="left">Non&#8208;food</th></tr></thead><tbody valign="top"><tr><td>First fixation duration</td><td>231 (4.6)</td><td>222 (4.4)</td><td>232 (4.4)</td><td>232 (4.3)</td></tr><tr><td>Gaze duration</td><td>268 (7.5)</td><td>255 (7.1)</td><td>267 (7.6)</td><td>271 (7.9)</td></tr><tr><td>Total time</td><td>407 (13.0)</td><td>407 (15.0)</td><td>460 (15.1)</td><td>449 (13.4)</td></tr><tr><td>Fixation count</td><td>1.97 (0.06)</td><td>2.04 (0.07)</td><td>2.21 (0.08)</td><td>2.17 (0.08)</td></tr><tr><td>Skipping probability</td><td>0.24 (0.02)</td><td>0.26 (0.02)</td><td>0.24 (0.02)</td><td>0.27 (0.02)</td></tr></tbody></table>

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Estimate</th><th align="left"><italic>SE</italic></th><th align="left"><italic>t</italic>/<italic>z</italic></th><th align="left"><italic>p</italic></th></tr></thead><tbody valign="top"><tr><td align="left">First fixation duration</td></tr><tr><td>(Intercept)</td><td>5.389</td><td>0.019</td><td>271.43</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>0.001</td><td>0.017</td><td>0.05</td><td>.963</td></tr><tr><td>Word</td><td>&#8722;0.035</td><td>0.018</td><td>&#8722;1.96</td><td>.049</td></tr><tr><td>Preview&#160;: word</td><td>0.036</td><td>0.025</td><td>1.45</td><td>.145</td></tr><tr><td align="left">Gaze duration</td></tr><tr><td>(Intercept)</td><td>5.491</td><td>0.029</td><td>188.72</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>&#8722;0.008</td><td>0.023</td><td>&#8722;0.37</td><td>.709</td></tr><tr><td>Word</td><td>&#8722;0.044</td><td>0.023</td><td>&#8722;1.89</td><td>.059</td></tr><tr><td>Preview&#160;: word</td><td>0.057</td><td>0.033</td><td>1.72</td><td>.085</td></tr><tr><td align="left">Total time</td></tr><tr><td>(Intercept)</td><td>5.917</td><td>0.047</td><td>125.76</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>0.145</td><td>0.032</td><td>4.56</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.024</td><td>0.032</td><td>0.76</td><td>.466</td></tr><tr><td>Preview&#160;: word</td><td>&#8722;0.053</td><td>0.045</td><td>&#8722;1.18</td><td>.237</td></tr><tr><td align="left">Fixation count</td></tr><tr><td>(Intercept)</td><td>1.959</td><td>0.083</td><td>23.66</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>0.256</td><td>0.060</td><td>4.29</td><td>&#60;.001</td></tr><tr><td>Word</td><td>0.069</td><td>0.059</td><td>1.16</td><td>.247</td></tr><tr><td>Preview&#160;: word</td><td>&#8722;0.111</td><td>0.084</td><td>&#8722;1.32</td><td>.188</td></tr><tr><td align="left">Skipping probability</td></tr><tr><td>(Intercept)</td><td>&#8722;1.349</td><td>0.157</td><td>&#8722;8.59</td><td>&#60;.001</td></tr><tr><td>Preview</td><td>0.021</td><td>0.138</td><td>0.15</td><td>.881</td></tr><tr><td>Word</td><td>0.163</td><td>0.136</td><td>1.21</td><td>.231</td></tr><tr><td>Preview&#160;: word</td><td>&#8722;0.012</td><td>0.192</td><td>&#8722;0.06</td><td>.947</td></tr></tbody></table>

The POF effect was used to determine whether Chinese readers could perceive food‐word information using parafoveal vision. Since the pseudoword previews were the same in both of the invalid preview conditions, we further compared the differences between the food and non‐food conditions in the two valid preview conditions. The results showed significant POF effects in the word category. The first fixation duration on the pre‐target region (see Table 7) was significantly longer in the food than the non‐food conditions, b = −0.037, SE = 0.017, t = −2.134, p = .033, and the gaze duration was also significantly longer in the food than non‐food conditions, b = −0.047, SE = 0.022, t = −2.066, p = .039. There were no significant differences in total time, fixation count and skipping probability between the food and non‐food conditions, ts < 1.247, ps > .213. These results suggest that Chinese readers can perceive the differences between the food and non‐food preview conditions using parafoveal vision. Parafoveal food‐word processing superiority may promote more competition between words in the target and pre‐target regions, leading to longer fixation durations (both first fixation durations and gaze durations) in the pre‐target region.

We further analysed whether restrained eating modulated parafoveal food‐word processing superiority in the pre‐target region (see Tables 9 and 10). BMI was included as a fixed effect, but no main effect or significant improvement was found in any model (Fixed factors: word category, restrained eating, interactions between the word category and restrained eating, BMI; Random factors: subject, trial item). The variance inflation factor results indicated that none of the models with BMI were affected by multicollinearity (VIFs < 1.5). The interaction between word category and restrained eating was marginally significant for the first fixation duration, but not for gaze duration. Because we were primarily focused on the potential differences between restrained and unrestrained eaters, further analyses were performed. The data (see Table 9) showed that the first fixation duration was significantly longer in the food than the non‐food preview conditions (b = −0.068, SE = 0.025, t = −2.693, p = .007) and the gaze duration was also significantly longer in the food than the non‐food preview conditions (b = −0.081, SE = 0.033, t = −2.416, p = .016) for the restrained eaters. However, there were no significant differences between the food and non‐food conditions for unrestrained eaters, ts <1 (see Figure 2). These results suggest that restrained eaters are more sensitive to parafoveal food words. Therefore, H3 is supported by the findings that the food‐word processing superiority can explain food‐related attentional bias in the early stages.

9 TABLE Eye movement measures on the pre‐target region for restrained and unrestrained eaters in valid preview conditions.

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Restrained eaters</th><th align="left">Unrestrained eaters</th></tr><tr><th align="left">Food</th><th align="left">Non&#8208;food</th><th align="left">Food</th><th align="left">Non&#8208;food</th></tr></thead><tbody valign="top"><tr><td>First fixation duration</td><td>232 (6.6)</td><td>217 (5.2)</td><td>229 (6.4)</td><td>228 (6.8)</td></tr><tr><td>Gaze duration</td><td>271 (11.8)</td><td>246 (8.8)</td><td>265 (9.6)</td><td>264 (10.7)</td></tr><tr><td>Total time</td><td>409 (17.6)</td><td>392 (17.0)</td><td>404 (19.1)</td><td>422 (24.0)</td></tr><tr><td>Fixation count</td><td>1.93 (0.08)</td><td>1.94 (0.09)</td><td>2.01 (0.09)</td><td>2.13 (0.11)</td></tr><tr><td>Skipping probability</td><td>0.23 (0.03)</td><td>0.23 (0.03)</td><td>0.25 (0.03)</td><td>0.29 (0.03)</td></tr></tbody></table>

<table><thead valign="bottom"><tr><th align="left" /><th align="left">Estimate</th><th align="left"><italic>SE</italic></th><th align="left"><italic>t</italic>/<italic>z</italic></th><th align="left"><italic>p</italic></th></tr></thead><tbody valign="top"><tr><td align="left">First fixation duration</td></tr><tr><td>(Intercept)</td><td>5.481</td><td>0.126</td><td>43.45</td><td>&#60;.001</td></tr><tr><td>Word</td><td>&#8722;0.128</td><td>0.055</td><td>&#8722;2.32</td><td>.021</td></tr><tr><td>Restrained eating</td><td>&#8722;0.021</td><td>0.041</td><td>&#8722;0.54</td><td>.589</td></tr><tr><td>BMI</td><td>&#8722;0.002</td><td>0.004</td><td>&#8722;0.62</td><td>.535</td></tr><tr><td>Word&#160;: Restrained eating</td><td>0.061</td><td>0.034</td><td>1.74</td><td>.081</td></tr><tr><td align="left">Gaze duration</td></tr><tr><td>(Intercept)</td><td>5.609</td><td>0.165</td><td>33.82</td><td>&#60;.001</td></tr><tr><td>Word</td><td>&#8722;0.145</td><td>0.073</td><td>&#8722;1.98</td><td>.047</td></tr><tr><td>Restrained eating</td><td>&#8722;0.013</td><td>0.053</td><td>&#8722;0.24</td><td>.812</td></tr><tr><td>BMI</td><td>&#8722;0.004</td><td>0.005</td><td>&#8722;0.81</td><td>.423</td></tr><tr><td>Word&#160;: Restrained eating</td><td>0.064</td><td>0.046</td><td>1.41</td><td>.158</td></tr><tr><td align="left">Total time</td></tr><tr><td>(Intercept)</td><td>5.779</td><td>0.276</td><td>20.91</td><td>&#60;.001</td></tr><tr><td>Word</td><td>&#8722;0.119</td><td>0.101</td><td>&#8722;1.18</td><td>.235</td></tr><tr><td>Restrained eating</td><td>0.023</td><td>0.085</td><td>0.27</td><td>.787</td></tr><tr><td>BMI</td><td>0.005</td><td>0.009</td><td>0.52</td><td>.603</td></tr><tr><td>Word&#160;: Restrained eating</td><td>0.095</td><td>0.063</td><td>1.51</td><td>.131</td></tr><tr><td align="left">Fixation count</td></tr><tr><td>(Intercept)</td><td>1.571</td><td>0.47</td><td>3.34</td><td>.001</td></tr><tr><td>Word</td><td>&#8722;0.075</td><td>0.186</td><td>&#8722;0.41</td><td>.686</td></tr><tr><td>Restrained eating</td><td>0.127</td><td>0.147</td><td>0.86</td><td>.391</td></tr><tr><td>BMI</td><td>0.009</td><td>0.015</td><td>0.58</td><td>.559</td></tr><tr><td>Word&#160;: Restrained eating</td><td>0.098</td><td>0.117</td><td>0.84</td><td>.402</td></tr><tr><td align="left">Skipping probability</td></tr><tr><td>(Intercept)</td><td>&#8722;1.572</td><td>0.448</td><td>&#8722;3.51</td><td>&#60;.001</td></tr><tr><td>Word</td><td>&#8722;0.154</td><td>0.444</td><td>&#8722;0.35</td><td>.729</td></tr><tr><td>Restrained eating</td><td>0.173</td><td>0.274</td><td>0.63</td><td>.528</td></tr><tr><td>Word&#160;: Restrained eating</td><td>0.199</td><td>0.271</td><td>0.73</td><td>.463</td></tr></tbody></table>

DISCUSSION

The present study used a gaze‐contingent boundary paradigm to explore foveal and parafoveal food‐word processing in Chinese reading. For foveal processing, we found empirical evidence of preview benefits, lending to the validation of H1. Additionally, food words were processed faster than non‐food words and this foveal processing superiority of food words was not modulated by restrained eating levels, thus validating H2. For parafoveal processing, the POF effect was found in restrained eaters, and thus, H3 is valid. The validation of H2 and H3 demonstrated the existence of food‐word processing superiority, which improves the Goal Conflict Model in explaining the attentional bias for food words. Therefore, the results of this study have important implications for our understanding of eye movement control in Chinese reading, as well as for food‐related attentional bias in the field of eating behaviours.

Although the purpose of this study was not to differentiate between the parallel and serial attention allocation hypotheses, the preview benefits of foveal word recognition and the POF effect in the present study are of great importance in understanding attention allocation during reading (see Schotter et al., [46] for a review). It is well known that there are no explicit visual cues such as spaces between Chinese words, but words still have a psychological reality in Chinese reading (i.e. word properties significantly influence eye movement control in Chinese reading) (Bai et al., [3]; Li & Pollatsek, [25]; Li et al., [24]; Zang et al., [65]). Prior studies have revealed that word recognition occurs at the same time as word segmentation during the reading of unspaced Chinese text (Li et al., [26]; Li & Pollatsek, [25]; Ma et al., [31]). There is no doubt that semantic information can be extracted from the parafovea in any reading model, since skipping frequent and shorter words is a common phenomenon in both Chinese and English reading (Li et al., [24]; Liversedge et al., [28]). The recognition of English words drives readers to skip those words, while the spaces between English words give the benefit of accurate landing positions. However, in unspaced Chinese reading, a recognized parafoveal word may possibly give rise to an unnecessary landing position on the recognized word, which is one potential reason why semantic preview benefits are frequently observed in Chinese reading (Yan et al., [61]; Yan et al., [63]; Yang et al., [64]).

The general processing superiority of food words in the fovea validated H2. According to the spreading‐activation theory of semantic processing (Collins & Loftus, [8]), food information can be represented as a short loop in the mental lexicon given that it is strongly related to human survival. Therefore, food words should be activated faster and recognized in a shorter time. Extra variables, such as word frequency, word predictability, plausibility and character complexity, would not have influenced the findings of our study, since all these variables were well controlled. The superiority of food‐word recognition might be contributed by the personalized mental lexicon. The personalized mental lexicon can lead to many recognition superiorities, for example the superiority of recognizing our names, which is also known as the cocktail‐party effect (Cherry, [5]). In addition to the superiority of food‐word processing, other stimuli related to human survival, such as threatening words and terrifying animals, usually lead to more attentional bias (Koster et al., [21]; Lipp & Derakshan, [27]), suggesting that survival‐related information might be better represented in human brains. However, this kind of word processing superiority has not been included in any reading models (Engbert et al., [12]; Reichle et al., [44]), since these models were designed for common eye movement control and did not account for special individual characteristics. Therefore, future models may consider more individual differences to improve the performance of reading models in explaining human reading behaviour.

The validation of H3 suggests that the lexical POF effects can be modulated by individual characteristics, such as restrained eating levels. In the pre‐target regions, there seemed to be no difference between the time that restrained eaters spent reading food words and the time that unrestrained eaters took to read either food or non‐food words. These interesting results may relate to the basic levels of parafoveal word processing in restrained and unrestrained eaters. However, the present study did not specifically measure the basic level of parafoveal word processing for each participant, and future studies could further clarify whether the basic levels of parafoveal word processing caused these results. Nevertheless, there was no doubt that the restrained eaters, not the unrestrained eaters, contributed to the differences in word processing between food words and non‐foods in the pre‐target regions. Therefore, H3 was partially confirmed in this study as food words were processed easier in the parafovea for restrained eaters, delaying the recognition of pre‐target words. However, this phenomenon is not easily explained by the Goal Conflict Model, which typically assumes that restrained eaters will have shorter fixation durations on pre‐target regions, and thus will approach food words (i.e. target regions) faster. This assumption cannot be supported by our results. Instead of the Goal Conflict Model, the competition hypothesis in Chinese reading could interpret these findings well.

The observed POF effects reflect the fact that readers with restrained eating can obtain food‐word information through parafoveal vision. The POF effect represents high‐level lexical processing of information from the parafovea, which has been traditionally considered to be evidence for parallel attention allocation. However, in recent years, high‐level parafoveal information extraction has been explained by both the parallel and serial reading models (Schotter, [45]; Zhu et al., [66]). Moreover, Ma et al. ([33]) reported that Chinese readers can perceive a word even when it is composed of non‐contiguous characters, and a right‐hand activated word lengthens fixation on the left‐hand word. According to the word competition hypothesis proposed by Ma et al. ([33]), word processing does not strictly happen from left to right in Chinese reading. The occasionally faster activation of parafoveal words (e.g. the words with high availability such as high‐frequency words) may compete with words that are already fixated upon, leading to lengthened fixation durations (see also Wang et al., [55] for a similar argument).

In addition, we assumed that BMI might independently influence food‐related attentional bias. However, BMI was not significant in any model in the present study, which does not support this hypothesis. Given the inconsistent results of previous studies regarding whether BMI could influence food‐related attentional bias (Nummenmaa et al., [37]; Pimpini et al., [39]), two possible explanations for these results can be proposed. First, we did not intend to include BMI as an independent variable in the initial experimental design. There was a significant difference between the BMI of the restrained and unrestrained eaters. However, this difference was probably not sufficiently representative of actual groups with a high vs. a low BMI. Second, even though the possibility of multicollinearity had been excluded (VIFs < 1.5), the slight potential effect of BMI might be incorporated into other fixed factors (restrained eating). This is because higher BMI levels are a common characteristic of restrained eaters as they are usually unsuccessful in their dietary restraint (Herman & Polivy, [16]).

In conclusion, the present study confirmed a new hypothesis regarding the processing superiority of food words, which could explain the attentional bias towards food words. Food‐word processing superiority in the fovea was found for both restrained and unrestrained eaters, indicating a general advantage in individuals' ability to recognize food words. This suggests that the observed approach or avoidance attentional bias for food words should consider the contribution of individuals' faster lexical acquisition of food words. Furthermore, the POF effects showed a difference between restrained and unrestrained eaters, indicating that restrained eaters were capable of extracting parafoveal food‐word information in the early stages of the attentional process. However, we could not conclude that the lengthened fixations on the pre‐target region only indicated restrained eaters' avoidance of upcoming food words, since no difference between restrained and unrestrained eaters was found in foveal processing. Restrained eaters' early‐stage attentional bias for food words might be a combination of the conflict of eating goals and the superiority of food‐word processing, which should be considered in other attentional‐bias paradigms. Notice that although five eye movement indicators were recorded in this study, the POF effect was mainly observed on early eye movement measures (i.e. the first fixation duration and gaze duration). As we mentioned in the Results section, FFD and GD were used to reflect participants' preview of target words, while other indicators reflected gazing information of target words. Moreover, it also demonstrates the importance of eye‐tracking techniques in future studies on food word processing.

In addition to individuals' restrained eating behaviour, other person‐related factors, such as gender (Doolan et al., [11]) and age (Werthmann et al., [60]), were not further examined in this study. These person‐related factors have been demonstrated to influence food‐related attentional bias. However, it is still unclear whether (or how) they influence food‐word processing. Furthermore, according to the experimental design, the DEBQ questionnaire which was administered before the formal experiment might have led both restrained and unrestrained groups to pay more attention to the food words than non‐food words. The experimental sequence needs to be modified in future studies to avoid such situations. Moreover, food cues were not further divided into different categories, while food‐related attributes, as well as food‐related environmental factors, may influence consumers' attentional process (Ma & Zhuang, [34]). Future studies could focus on more person‐, food‐, and environment‐related factors to further investigate the mechanism of the food‐word processing and the attentional bias for food words.

CONCLUSION

This study revealed a foveal processing superiority of food words in general groups, and a parafoveal processing superiority of food words in restrained eaters. These findings suggested that the food‐word processing superiority should not be neglected when interpreting the attentional bias for food words, especially for groups with special eating behaviours, such as restrained eating.

AUTHOR CONTRIBUTIONS

<bold>Changlin Luo:</bold> Conceptualization; data curation; formal analysis; investigation; methodology; validation; visualization; writing – original draft. <bold>Mengyan Zhu:</bold> Data curation; formal analysis; investigation; methodology; software; visualization; writing – original draft. <bold>Xiangling Zhuang:</bold> Conceptualization; funding acquisition; methodology; project administration; resources; validation; writing – review and editing. <bold>Guojie Ma:</bold> Conceptualization; funding acquisition; methodology; project administration; resources; software; validation; writing – review and editing.

ACKNOWLEDGEMENTS

We are extremely grateful to the participants for giving up their time to take part in this study. This study was funded by the Fundamental Research Funds for the Central Universities grant number (GK202103133) and Natural Science Foundation of Shaanxi Province, China grant number (2022JQ‐183).

CONFLICT OF INTEREST STATEMENT

None.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from https://osf.io/9d3b4/.

GRAPH: Appendix S1

GRAPH: Appendix S2

By Changlin Luo; Mengyan Zhu; Xiangling Zhuang and Guojie Ma

Reported by Author; Author; Author; Author