Elsevier

Cognition

Volume 147, February 2016, Pages 1-20
Cognition

Universality in eye movements and reading: A trilingual investigation

https://doi.org/10.1016/j.cognition.2015.10.013Get rights and content

Abstract

Universality in language has been a core issue in the fields of linguistics and psycholinguistics for many years (e.g., Chomsky, 1965). Recently, Frost (2012) has argued that establishing universals of process is critical to the development of meaningful, theoretically motivated, cross-linguistic models of reading. In contrast, other researchers argue that there is no such thing as universals of reading (e.g., Coltheart & Crain, 2012). Reading is a complex, visually mediated psychological process, and eye movements are the behavioural means by which we encode the visual information required for linguistic processing. To investigate universality of representation and process across languages we examined eye movement behaviour during reading of very comparable stimuli in three languages, Chinese, English and Finnish. These languages differ in numerous respects (character based vs. alphabetic, visual density, informational density, word spacing, orthographic depth, agglutination, etc.). We used linear mixed modelling techniques to identify variables that captured common variance across languages. Despite fundamental visual and linguistic differences in the orthographies, statistical models of reading behaviour were strikingly similar in a number of respects, and thus, we argue that their composition might reflect universality of representation and process in reading.

Introduction

The issue of universality has been central to linguistics and psycholinguistics for decades. Chomsky (1965) argued that “…the main task of linguistic theory must be to develop an account of linguistic universals that, on the one hand, will not be falsified by the actual diversity of languages, and, on the other, will be sufficiently rich and explicit to account for the rapidity and uniformity of language learning, and the remarkable complexity and range of the generative grammars that are the product of language learning” (pp. 27–28). In a recent article, Frost (2012) makes a strong argument for a Universality Constraint in relation to reading, suggesting that psychological models of the process of reading should reflect cognitive operations that are common across languages with different writing systems or scripts. In his thesis, Frost goes beyond the traditional Chomskyan notions of universality, making the case for cross-linguistic commonality at the level of cognitive processing. Frost further argues that establishing reading universals is a prerequisite for the development of meaningful, theoretically motivated cross-linguistic models of reading. The responses to Frost’s target article are very interesting in that they reveal a broad spectrum of views pertaining to the issue of universality in relation to written language processing, ranging from broad agreement (e.g., Feldman & Moscoso del Prado Martin, 2012) through to the suggestion that there are no such things as universals of reading (Coltheart & Crain, 2012). The views delivered in the article along with the responses to them provide a very relevant context to the experimental project that we report here. We were keen to investigate whether it might be possible to identify factors that could account for common variance across very different written languages in an on-line measure known to reflect moment-to-moment cognitive processing during reading. Our objective in doing this was to first establish whether such variables did exist, and if so, try to evaluate whether those variables might represent universal aspects of reading. If such universals do exist, they represent common principles by which the written language processing system extracts information from print across different languages. Indeed, if this is the case, then one of the strongest predictions that can be made on the basis of Frost’s universality formulation is that whilst different writing systems may visually represent linguistic information in quite different ways, the extraction of meaning from comparable units of language should require a similar amount of time. That is, whilst the moment-to-moment machinations of meaning computation may differ across languages, overall, the time to compute meaning from comparable units of written language should be similar. Arguably, at a fundamental level, universality suggests that an assumption of temporal unity in relation to the attainment of comprehension (regardless of visual format) should hold, and this in turn strongly implies comparability in the time required to attain that state. Finally, our approach in this project also provided an opportunity to pursue a more general objective, namely, to provide comparable cross-linguistic descriptives of reading behaviour.

Before developing our claims in detail, it is necessary to be clear about two points. First, unlike the implicit position adopted by Frost (2012), we do not consider theories of written word identification to be the equivalent of theories of reading (see Liversedge, Blythe, & Drieghe, 2012). Instead, we consider comprehension of multi-word text to constitute reading, rather than simply the identification of isolated words. Furthermore, it is our view that word identification occurs differently for isolated words than during normal reading (see Rayner & Liversedge, 2011). Thus, whilst word identification is clearly a central and critical aspect of reading, numerous other cognitive processes are also required for successful text comprehension (e.g., syntactic, semantic, discourse processes, anaphor resolution, inferential processing, etc.). For these reasons, when we discuss reading in the present article we include consideration of processing beyond word identification. Our second qualification concerns exactly what we mean when we refer to universality. As should become clear, we do not restrict our use of the term to the notion of Formal and Substantive Universals as originally stipulated by Chomsky (1965). Instead, perhaps unsurprisingly, we will consider universality in relation to representations and cognitive processes that are common to reading across languages (with the exception of Braille). It is in this sense that our claims will be about aspects of written language processing that are universal.

As mentioned, reading is a visually mediated psychological process. Humans process visual information via the eyes. Visual information, and more specifically in the case of reading, text, is encoded and then represented in an abstract form after which it is linguistically processed by later cognitive systems. Written language comprehension results in the formation of a representation of the meaning of text, often referred to as a discourse representation. In this sense, the human visual processing system (including “the brain’s letter-box”, Dehaene, 2009) sub-serves the linguistic processing system, delivering the information that the language processor needs in order to carry out its computations. As already indicated, the eyes are the means by which visual information is encoded for subsequent processing, and the human eye has a particular physiological make up that has important implications for the eyes’ functional role in the uptake of visual linguistic information. At (approximately) the middle of the retina there is the fovea, a small circular area (roughly 2°), that provides high acuity visual information, and beyond which, in the parafovea and the periphery, vision is of much reduced visual acuity. Consequently, this causes humans to visually sample their environment by making a series of fixations, which are short periods where the eye is comparatively still (usually lasting about quarter of a second during reading), and saccades, which are fast, ballistic rotations of the eyeball. During fixations humans cognitively process the visual information that they have encoded, whereas during saccades, there is no useful visual input. All humans across all cultures who have an undisrupted visual system visually sample their environment in this way (Findlay & Gilchrist, 2003), and it has been argued that saccadic sampling has evolved due to its efficiency in relation to visual information processing (Gilchrist, Brown, & Findlay, 1997). Furthermore, eye movements are very largely under cognitive control, and measurement of temporal and spatial properties of saccades and fixations during reading provides an excellent on-line index of cognitive processing (Liversedge and Findlay, 2000, Rayner, 1998, Rayner, 2009). Thus, despite our perceptual experience during reading being one of a smooth, continuous flow of information, in fact, it takes place via a staccato succession of discrete snapshots, each providing detailed information from a small portion of the sentence (usually a word or two). In other words, detailed visual information necessary for linguistic processing beyond the centrally fixated (foveal) region is not available. It is important to understand, however, that readers do not exclusively process text directly at fixation. If this were the case, then linguistic processing would be extremely tightly yoked to specific fixations (c.f., the Eye-Mind Assumption, Just & Carpenter, 1980). Instead, there has been substantial work (see Rayner, 1998, Rayner, 2009) showing that readers partially pre-process upcoming text in the parafovea in the direction of reading (McConkie and Rayner, 1975, Rayner, 1975). In sum, saccadic eye movements during reading are common throughout the human species, regardless of culture or language, and saccadic sampling and the retinal make-up constrain the rate at which visual information is encoded and delivered by the visual system to the language processing system.

Not only does commonality exist in relation to human eye movements during reading, but also certain linguistic effects on eye movement behaviour occur across languages. For example, lexical frequency effects are known to occur robustly across most languages such that words that are more frequent are read more quickly than words that are less frequent (Ellis, 2002). Also, word length effects have been demonstrated across languages, whereby longer words take longer to read than shorter words (Bertram and Hyönä, 2003, Just and Carpenter, 1980, Rayner et al., 1996). Finally, words that are more predictable on the basis of preceding sentential context are read more quickly than words that are less predictable (Balota et al., 1985, Ehrlich and Rayner, 1981, Inhoff, 1984). The fact that word frequency, word length and word predictability effects (the “big three” in reading, Clifton et al., in press), are found across languages provides evidence for the more general suggestion of the importance of word based processing during reading across languages (see Li, Bicknell, Liu, Wei, & Rayner, 2014).

The next point that we will consider concerns the script, or the physical form of a written language. As Perfetti and Harris (2013) make clear, reading depends on the writing system that encodes the language. We strongly concur with this view. Here, we will consider the writing system of the language in relation to two issues relevant to reading: (1) the visual and informational density of the written language and (2) the intricacies of the orthography (notational system) that capture and represent linguistic characteristics.

Scripts vary across languages to a very significant degree. Some written languages are extremely, visually dense (e.g., Chinese), whilst others are less dense and (usually) horizontally spatially extended (e.g., English, or even more so, Finnish). To be clear, by visual density, we mean the amount of visual information that is available per unit of text. This definition, in itself, raises complexities in relation to what actually constitutes a unit of text. For the moment, however, let us sidestep this question and consider visual density in relation to one of the three languages we have chosen to examine. In written Chinese, visual density can be indexed in terms of the stroke complexity of characters and words (e.g., Liversedge et al., 2014) since all characters occupy the same unit of space, and some characters are comprised of many strokes, whilst others are comprised of comparatively few; the more strokes that comprise a character or word, the greater the visual density. Note, though, that not all strokes carry equivalent weight within a character (Wang et al., 2013, Yan et al., 2012), and for this reason, stroke count in relation to the relative importance of those strokes may offer a more veridical index of informational density. In English, Finnish and other alphabetic languages, the letters that comprise words are directly comparable, and to this extent, words of equal length can be considered to be comparably visually dense. Furthermore, it is generally accepted that in alphabetic languages words comprised of more characters are more visually complex than words comprised of fewer characters, and in this sense, alphabetic word length is a proxy for visual complexity (Liversedge et al., 2014). However, on average, words are longer in Finnish than in English (and most often, there will be fewer words in the Finnish than the English version of a directly translated sentence), and therefore, in relation to the amount of information a character in a word conveys, then the informational density of words in English is greater than in Finnish. That is to say, on average, in English more information is packed into a smaller word unit than is the case in Finnish. This is partly due to prepositions being expressed as inflectional suffixes in Finnish. From that perspective, Finnish words may be argued to contain more information than English words. Yet, information density per character is still greater in English due to written Finnish marking all phonemes with separate graphemes which lengthens the written words (see below, for more details concerning grapheme–phoneme correspondence in the studied languages).

A key point is that the visual density of Chinese, and the informational density of English and Finnish directly impact on how readily the written form of the language can be encoded during a fixation, or across multiple fixations. Recall that detailed visual information is only encoded from a limited portion of the retina. When a script is visually or informationally dense, a greater amount of information is foveally available to be processed on any particular fixation, and therefore fewer fixations are required to encode that information. However, at least to some extent, due to the increased visual or informational density of that information, the duration of the fixations required to encode it will be increased. In contrast, for less dense scripts, readers need to make a greater number of fixations in order to encode a comparable amount of information (since less of the script is foveally available during any single fixation). Consequently, for less dense scripts, additional fixations are required for successful encoding. Note, however, fixations during reading of such languages will be shorter in duration because the amount of information encoded is reduced. In this basic respect we can start out by formulating a strong experimental prediction in relation to eye movement behaviour during reading across languages. Specifically, we can anticipate a trade-off between fixation durations and numbers of fixations that is related to the visual or informational density of the written language. For more dense scripts, fewer fixations of longer duration will occur. In contrast, for less dense scripts, there will be more fixations but these will be of shorter duration.

The visual and informational density of scripts is not the only variable that is important in relation to differences in eye movements during reading in different languages. Different scripts also vary in the way linguistic information is conveyed in their annotated form, that is, there are differences in their orthography. It has been argued that there are three broad categories of written language: logographic (or morpho-syllabic), syllabic and alphabetic languages, with each category being characterised according to the predominant correspondence between the unit of spoken form and the unit of the written language (e.g., Gelb, 1952). A good example of a logographic language is Chinese; a syllabic language is Japanese, and English and Finnish are both alphabetic languages. There are, however, further relevant and important distinctions between different orthographies of written languages that relate to the systematicity of the mapping relationship between written form and sound. In logographic languages like Chinese, in which syllables rather than individual phonemes comprise the basic units of sound, many visually different characters can represent the same syllable. This means that the relationship between the visual and spoken form of a character is arbitrary to a significant degree. Orthographic depth refers to the degree of consistency between the orthographic and phonological forms: the less consistent the relationship, the deeper the orthography. In relation to our examples, Chinese has a very deep orthography, with little by way of consistency in the mapping between characters and syllables. In English, a less deep orthography, whilst there is significant consistency, there is also a degree of inconsistency (e.g., mint, pint). Finally, in Finnish, a language with a very shallow orthography, the relationship is so consistent and transparent that the manner in which words in the language are pronounced is fully specified by their written form. Thus, phonological ambiguity in relation to orthographic form is very prevalent in Chinese, present but less abundant in English and virtually absent in Finnish. A final noteworthy point is that it has been consistently demonstrated (Durgunoglu, 2006, Ellis and Hooper, 2001, Hanley et al., 2004, Spencer and Hanley, 2003, Seymour et al., 2003 Ziegler et al., 2010) that children’s reading, as indexed by their ability to read out loud, appears to develop more rapidly in languages with shallower orthographies (though whether there are corresponding differences in comprehension remains an open question, Seidenberg, 2011). The point to take here is that scripts of languages not only differ in their visual and informational densities, but also in how the orthography conveys linguistic information. To this extent, written languages also vary in their linguistic specificity. We need to now consider this fact in relation to the arguments we have developed concerning commonality in visual processing and eye movement behaviour during reading across cultures, and further, how this constrains the delivery of information to the language processing system. It should be very clear that because visual encoding (both in relation to retinal acuity and the nature of visual sampling) is constant across readers of different languages, but that the visual and linguistic characteristics of different written languages are themselves very different, then the rate and nature of cognitive processing during reading will be necessarily constrained to differing degrees for those different languages.

Given this argument, the reader may be confused by our intention to investigate universality of processing during reading. Indeed, thus far, the issues we have raised seem to point very directly to the suggestion that, if anything, there should be a significant degree of language specificity in relation to the rate and nature of linguistic processing during reading. However, if there were universal representations associated with reading, and those representations mediated the nature of such processing across different languages, then a strong hypothesis would be that even in the light of marked linguistic differences between the written forms of languages, it should still be possible to observe commonality in relation to representation and process. In order to show this, however, it would be necessary to construct maximally comparable translations of text stimuli across three quite different writing systems (e.g., Chinese, English & Finnish). This would ensure that any differences that did occur would not be due to language specific content differences. Furthermore, it would be necessary to record eye movements using identical methods from expert native readers of comparable linguistic proficiency, thereby avoiding the possibility that cross-linguistic effects could be attributable to measurement or participant differences. If these requirements could be met then we might argue that universality of process (and representation in relation to such process) should be associated with variables that capture common cross-linguistic statistical variance in the eye movement data. To reiterate our claim, given the combination of visual encoding that is operationally the same across languages, and cross-linguistic specificity in relation to visual and linguistic characteristics of text, one might assume that linguistic processing during reading, and more precisely, the construction of a phonological representation on the basis of orthography, proceeds in a language-specific, idiosyncratic fashion. Alternatively, despite cross-language differences in relation to visual and linguistic form, and the fact that these are encoded via a visual system that is operationally the same regardless of the particular characteristics of a script, at a very fundamental level, the linguistic representations and processes that exist across languages might share commonality.

To investigate this issue, we conducted an eye movement experiment across three languages. The languages that we selected were Chinese, English and Finnish. We selected these languages very purposefully due to their particular linguistic characteristics. As described earlier, Chinese is logographic, whereas English and Finnish are alphabetic. Chinese has the deepest orthography of the three languages. English and Finnish also differ in terms of their orthographic depth, even though they are both alphabetic. English is a comparatively deep orthography, whereas Finnish is one of the most shallow of all orthographies. Another important difference between the languages concerns the fact that it is not the case that there is perfect word-to-word correspondence across the three languages – some words that occur in one language do not occur in the other languages. For example, most prepositions in Finnish are coded morphologically as part of a noun. Thus, there are a significant number of characters in Chinese, and short function words that appear in English that do not appear in Finnish. Similarly, articles appear in English, but not in Chinese or Finnish, and some characters that feature in Chinese do not have corresponding words in English or Finnish (e.g., operators of quantity; characters marking possessives, characters marking words as an adjective or a noun, etc.).

In addition to cross-linguistic differences, visual differences between the three selected scripts are also plentiful. Chinese is very dense with over 90% of the words in the language comprising one or two characters. Furthermore, the characters themselves have constituent radicals that are comprised of individual strokes, and importantly, there is structure to the arrangement of these strokes. In fact, the stroke structure of a character has implications for processing during Chinese character identification and reading (Wang et al., 2013, Yan et al., 2012). In stark contrast to Chinese, because Finnish and English are alphabetic, the written form is horizontally spatially extended to a greater degree, with the constituent letters of words being presented horizontally adjacent to each other. Alphabetic letters are less visually dense than Chinese characters. Note, also, that since Finnish is an agglutinative language, being comprised of a high proportion of multi-morphemic words, many of the words in Finnish are very long (words of 12 or more letters are quite common). Thus, the horizontal spatial extent of words and sentences is greater in Finnish than English, and far greater than in Chinese. A final characteristic of Chinese that makes it visually distinct from Finnish and English, is that Chinese is an unspaced language, that is, there are no spaces between the words in Chinese sentences. The lack of word spacing in character-based languages has been shown to be very important in relation to eye movements, saccadic targeting and reading (see Bai et al., 2008, Blythe et al., 2012, Li et al., 2011, Sainio et al., 2007, Shen et al., 2012, Yan et al., 2010, Zang et al., 2013). The lack of spaces between words in Chinese contributes further to its reduced horizontal extent, and this also means that a process of word segmentation is required for word identification to occur that is unnecessary in English and Finnish (with the exception of long, multimorphemic compound words, Bertram, Pollatsek, & Hyönä, 2004).

Eye movement recordings have been used to investigate aspects of alphabetic reading and aspects of non-alphabetic reading separately for many decades (Huey, 1900, Javal, 1878, 1879, Shen, 1927, Tinker, 1936a, Tinker, 1936b). However, to date, there have only been two experiments that have directly compared eye movements during reading across alphabetic and character-based languages (Feng et al., 2009, Sun et al., 1985), and both of these examined Chinese and English readers. To date, no study has investigated Chinese reading in relation to an alphabetic language with a shallow orthography such as Finnish (and indeed, no study has investigated differences in relation to reading of alphabetic languages with shallow and deep orthographies such as Finnish and English respectively).

In the study by Sun et al. (1985) participants whose native language was Chinese were required to read paragraphs of text (taken from Scientific American) that had been translated into Chinese, as their eye movements were recorded. Their eye movement recordings were compared with eye movement data obtained from native English speaking participants with a similar level of scientific training reading the original English versions of the texts. Sun et al.’s analyses of Chinese and English reading behaviour were largely descriptive with few formal statistical comparisons, and overall, they focused on the similarities in eye movement behaviour that existed between the two languages. They showed that the general pattern of saccades and fixations was broadly similar in the two languages, with readers making a succession of left to right saccades and fixations as the text was read. They also argued that fixation durations were comparable in the two languages (though note that the average fixation duration was longer for English [270 ms] than for Chinese [260 ms]), and that there were similar numbers of regressive eye movements. Furthermore, when they applied a scaling factor of 1.5 (number of Chinese words to English), then further eye movement metrics also became quite comparable. For example, saccades were 1.7 Chinese units relative to 1.8 words in English, and under this conversion, reading rates were 390 units per minute in Chinese compared to 380 words per minute in English. Indeed, as Sun et al. argued, these values are quite comparable, however, as we will see later, the method of scaling employed by Sun et al. is quite arbitrary (see also Yang & McConkie, 1994). Quite what it means to say that 1.5 words in Chinese is the equivalent of 1 word in English is not at all clear. Furthermore, such scaling can mask real, important differences that exist in eye movement behaviour during reading of Chinese and English. We will return to these issues below.

Feng et al. (2009) undertook the second study that attempted to quantify differences in reading behaviour between alphabetic and non-alphabetic languages. In their study, as with the study by Sun et al. (1985), they compared reading behaviour for Chinese and English. However, Feng et al. were primarily concerned with cross-linguistic changes in eye movement behaviour during reading across development, focusing both on the unique characteristics of the different orthographies, along with psychological changes associated with maturation and reading development. Feng et al. focused on three aspects of the languages for which there were pronounced differences; the fundamental linguistic units of each language, its orthographic depth, and the presence of boundary demarcations between linguistic units (word spacing). They tested Chinese and English readers in three age groups; 9 years, 11 years and adults. For their stimuli they used a combination of texts that were direct translations from English to Chinese (16% of stimuli), and texts that differed in content between the age groups and languages (84% of stimuli), and in their analyses, they reported results from data collapsed across each type of stimulus. This is a critical aspect of this study, in that this design, at least in principle, allows for the possibility that differences in content between stimuli could have contributed to any effects (or lack of them) between languages (or ages). Nevertheless, the main findings from the adult participants reported by Feng et al. are relevant to some of the issues that we investigated in the current study, and we will, therefore, consider these briefly. To summarise, for reading times per word, number of fixations per word, mean fixation durations per word, refixations on words and number of progressive fixations, Feng et al. found no reliable differences between Chinese and English readers. Saccades were longer on average in Chinese than in English readers, though note that the physical size of the text was made approximately equal, and given that word units rather than text size is a primary influence on saccade targeting (Morrison & Rayner, 1981) this could have artificially inflated the length of saccades in Chinese relative to English. Finally, Feng et al. found that, on average, Chinese readers made more regressions than English readers, although they tentatively suggested this effect might be an artefact associated with the software they used to establish word boundaries in their Chinese stimuli.

As should be clear, both the study by Sun et al. (1985) and the study by Feng et al. (2009) suggest commonality in the characteristics of the eye movements of Chinese and English readers. However, in both studies it is not unambiguously clear that the failure to find differences in eye movements between Chinese and English readers, or that those differences that were found, occurred due to extraneous factors. In the present study, therefore, we took several steps to avoid potential confounds in our experimental procedures in order that we might more unequivocally assess the extent of commonality in eye movement behaviour during Chinese, Finnish and English reading. To do this, we constructed a series of short expository texts that we translated across Chinese, English and Finnish. In constructing the texts across languages we were very careful to ensure that the content and constructions were as similar as possible, thereby permitting maximal comparability of eye movement behaviour across the stimuli in each language (see Fig. 1).

Our rationale in this respect was to use texts that caused readers to engage in linguistic processing that was as similar as possible across the languages. Recall that this approach is quite different, and arguably more powerful, in comparison to the study of Feng et al. (2009), in that our use of minimally different texts across languages minimised variability due to content differences.

To ensure comparability of eye movement measures during reading, it was necessary to decide upon regions of text that adequately corresponded across languages (and within limits, regardless of text size). This was a non-trivial decision since the nature of the written form of the three languages differs to such a degree. To allow for comparison of equivalents, it was necessary to compute measures of eye movement behaviour over portions of text that unambiguously convey the same information in each language. For this reason, the region of analysis that we adopted in this investigation was the sentence. For sentences, in all of the languages it is the case that all the constituents from the first word in the sentence until the following period comprise the text that conveys comparable information. To adopt a more granular level of analysis would bring into question the level of homogeneity of linguistic information contained within corresponding regions over which fixations were made and eye movement measures computed. Under such circumstances, comparison of eye movement measures between languages may not represent a comparison of like with like (in the course of this project we have come to refer to this issue as the “apples and pears” issue). Our general hypothesis was that, in the absence of cross-linguistic differences in comprehension, any commonality across languages in relation to linguistic representation and process should be revealed as commonality in statistical models accounting for variance in the same measures of eye movement behaviour during reading of comparable regions of text in the different languages. If such commonality did exist, then given similar constraints on human visual encoding, but marked visual and orthographic differences between the written forms of Chinese, English and Finnish, commonality may be taken to represent universality in aspects of representation and process during reading of these languages.

Section snippets

Participants

Twenty-five Chinese undergraduate students from Tianjin Normal University, twenty-one English undergraduate students from the University of Southampton and twenty Finnish students from the University of Turku took part in the experiment. The Chinese participants read texts in Chinese, the English participants texts in English and the Finnish participants read texts in Finnish. All participants were undergraduate psychology students, and all had normal or corrected to normal vision. All of the

Results and discussion

Trials where there was tracker loss were removed from the data prior to the analyses. Fixations shorter than 80 ms that were within one character of the previous or following fixation were merged, all other fixations shorter than 60 ms that were within one character of the previous or following fixation were merged and all other fixations shorter than 60 ms or longer than 800 ms were removed. Also, when calculating the eye-movement measures, for each participant, data more than 2.5 standard

General discussion

Let us start by considering the basic descriptive data from the texts that were translated from English into Chinese and Finnish. Recall that the translation process was carried out quite painstakingly to ensure that sentential constituents were maximally comparable across languages. Even so, due to quite marked differences across languages, it was still the case that there were far from word-to-word correspondences. Most strikingly, the number of words in the Finnish stimuli was substantially

Acknowledgements

Research visits of Liversedge, Drieghe and Li to Turku were financially supported by University of Turku; Li’s visit was also supported by the CIMO (Centre for International Mobility, Finland), and the National Social Science Fund for Education (CBA120106). The research was also supported by the Recruitment Program of Global Experts (1000 Talents Award from Tianjin) and NSFC grant number (Natural Science Foundation of China Grant: 31571122). The authors would like to thank Chuanli Zang and Lei

References (63)

  • M. Sainio et al.

    The role of interword spacing in reading Japanese: An eye movement study

    Vision Research

    (2007)
  • X. Bai et al.

    Reading spaced and unspaced Chinese text: Evidence from eye movements

    Journal of Experimental Psychology: Human Perception and Performance

    (2008)
  • N. Chomsky

    Aspects of the theory of syntax

    (1965)
  • Clifton, C., Ferreira, F., Henderson, J. M., Inhoff, A., Liversedge, S. P., Reichle, E. D., & Schotter, E. R. (2015)....
  • M. Coltheart et al.

    Are there universals of reading? We don’t believe so. Invited commentary on “Towards a universal model of reading”

    Behavioral and Brain Sciences

    (2012)
  • Contemporary Chinese Dictionary (2005). Beijing: Commercial Publishing...
  • L. Cui et al.

    Parallel processing of compound word characters in reading Chinese: An eye movement contingent display change study

    Quarterly Journal of Experimental Psychology

    (2013)
  • S. Dehaene

    Reading in the brain: The science and evolution of a human invention

    (2009)
  • A.Y. Durgunoglu

    How the language’s characteristics influence Turkish literacy development

  • N.C. Ellis

    Frequency effects in language processing

    Studies in Second Language Acquisition

    (2002)
  • N.C. Ellis et al.

    Why learning to read is easier in Welsh than in English: Orthographic transparency effects evinced with frequency-matched tests

    Applied Psycholinguistics

    (2001)
  • L.B. Feldman et al.

    Does a focus on universals represent a new trend in word recognition?

    Behavioral and Brain Science

    (2012)
  • G. Feng et al.

    Orthography and the development of reading processes: An eye movement study of Chinese and English

    Child Development

    (2009)
  • J.M. Findlay et al.

    Active vision: The psychology of looking and seeing

    (2003)
  • J.M. Findlay et al.

    A model of saccade generation based on parallel processing competitive inhibition

    Behavioral and Brain Sciences

    (1999)
  • R. Frost

    Towards a universal model of reading

    Behavioral and Brain Sciences

    (2012)
  • I.J. Gelb

    A study of writing: The foundations of grammatology

    (1952)
  • I.D. Gilchrist et al.

    Saccades without eye movements

    Nature

    (1997)
  • J.R. Hanley et al.

    How long do the advantages of learning to read a transparent orthography last? An investigation of the reading skills and reading impairment of Welsh children at 10-years of age

    The Quarterly Journal of Experimental Psychology

    (2004)
  • E.B. Huey

    On the psychology and physiology of reading. I

    American Journal of Psychology

    (1900)
  • J. Hyönä et al.

    The role of component morphemes on eye fixations when reading Finnish compound words

    Journal of Experimental Psychology: Human Perception and Performance

    (1998)
  • Cited by (86)

    View all citing articles on Scopus
    View full text