The effect of visual complexity and word frequency on eye movements during Chinese reading

Abstract

Eye movements of native Chinese readers were monitored when they read sentences containing single-character target words orthogonally manipulated for frequency and visual complexity (number of strokes). Both factors yielded strong main effects on skipping probability but no interaction, with readers skipping visually simple and high frequency words more often. However, an interaction between frequency and complexity was observed on the fixation times on the target words with longer fixations for the low frequency, visually complex words. The results demonstrate that visual complexity and frequency have independent influences on saccadic targeting behaviour during Chinese reading but jointly influence fixation durations and that these two factors differently impact fixation durations and saccade targeting during reading.

Keywords:

• Visual complexity
• word frequency
• skipping
• eye movements
• Chinese reading

The work described in this article was supported by a Grant from the Natural Science Foundation of China [31100729], a Postgraduate Scholarship from the China Scholarship Council, Economic and Social Research Council Grants [RES-000-22-4128 & ES/I032398/1] and a Leverhulme Trust Grant [F/00 180/AN]. The authors would like to thank Elizabeth Schotter, Jinmian Yang and an anonymous reviewer for their helpful comments on a previous version of this paper.

The work described in this article was supported by a Grant from the Natural Science Foundation of China [31100729], a Postgraduate Scholarship from the China Scholarship Council, Economic and Social Research Council Grants [RES-000-22-4128 & ES/I032398/1] and a Leverhulme Trust Grant [F/00 180/AN]. The authors would like to thank Elizabeth Schotter, Jinmian Yang and an anonymous reviewer for their helpful comments on a previous version of this paper.

Notes

¹ Visual complexity of a Chinese character is also correlated with sub-lexical character complexity, for example, the number of constituent radicals that comprise a character. Consideration of radicals in relation to visual complexity complicates matters since the literature contains inconsistent definitions of the radical (e.g., see Chen, Allport, & Marshall, 1996; Su & Samuels, 2010; Taft & Zhu, 1997), and this confusion even occurs within the same Chinese dictionary. For example, in the dictionary of Chinese Character Information (1988), the character “横” (meaning horizontal) is defined as having two radicals ( and ), while the character “黄” (meaning yellow) is defined as having three radicals (, and ). The key point in relation to the current study is that neither stroke complexity, nor the number of constituent radicals that comprise a character, is a perfect index of the character's visual complexity. Instead, both are approximations of it, and both are correlated with each other. For this reason, regardless of whether the number of strokes or the number of radicals is responsible for the effects discussed in this paper, the influence of visual complexity is clear. Future research is required to dissociate effects of stroke complexity from effects of the complexity of orthographic structure (number of radicals).

² In order to maximize the possibility of skipping occuring, single character words were selected as target words. These words were selected such that they never combined with the next character to form a two-character word (as per Cai & Brysbaert, 2010). However, when we requested 60 participants (15 in each of the four counterbalancing conditions, none of whom took part in the eye movement experiment) to actually mark where they thought the word boundaries were within the sentences, we found that readers did sometimes take the single character to be the first character of a two character word. To check whether this factor could have influenced the eye movement results, comprehensive Linear Mixed Model comparisons were conducted, revealing that agreement did not explain additional variance in the models in any of the measures examined for either skipping or fixation times (all ps > .16).

³ The character frequency (regardless of whether the character appeared as a single character word, or as part of a multi-character word) was also analyzed. The data showed that, similar to word frequency, the character frequency between high- and low-frequency words was also reliably different, F(1, 39) = 28.39, p < .001. To this extent, word frequency and character frequency were perfectly confounded, and it is therefore quite possible that character frequency per se may also contribute to the effects we report here. Most importantly, however, our analyses showed that character frequency was not significantly different between high- and low complexity words (all Fs < 1.73, ps > 0.5), and therefore, effects of this variable could not be attributed to extraneous effects of character frequency.

⁴ The mean reading times and skipping rates across conditions may initially appear to be comparable, and arguably, reflect a similar interactive pattern. Note, however, the interactive pattern for the reading times is such that first fixations, single fixations and gaze durations were longest when the target word was low frequency and complex relative to the other three conditions. It is also true that a comparable counterpart pattern occurred in word skipping. Numerically, the lowest skipping rates occurred for the low frequency complex target words relative to the other conditions. However, quite unlike the reading time measures, this effect was completely non-significant (SP p = .789), and therefore, we interpret the reading time and skipping effects as being qualitatively different in nature.

⁵ Following the suggestion of two reviewers we also ran LMM models with a “maximal” random-effect structure including random slopes for all fixed effects and their interactions (Barr, Levy, Scheepers, & Tily, 2013) on the fixation duration data. When the model did not converge we removed the correlations between the random slopes. All the resulting models gave qualitatively identical patterns for the data. As requested by one of the reviewers, we also ran additional contrasts directly testing the main effects, for skipping the main effects of frequency (b = −.38, SE = .09, p < .001) and complexity (b = −.46, SE = .10, p <. 001) were significant but none of the main effects for the fixation durations were (all ts < 1.80).