Chapter 1. Literature review
1.1 Eye Movement Studies in Reading Languages other than Chinese
1.1.1 The Influence of Low-level Features on Saccade Targeting
國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
1.1 Eye Movement Studies in Reading Languages other than Chinese
1.1.1 The Influence of Low-level Features on Saccade Targeting
Word Length
The decision about where to move the eyes next in alphabetic languages3 appears to be primarily determined by low-level visual information obtained from parafoveal vision;
in particular, two variables have been found to greatly influence the Initial Landing Position (ILP) within words, being those the main cues upon which the oculomotor system relies for sending the eyes forward: word length and interword spacing. The results provided by Rayner (1979) on the proportion of ILP in English words of different lengths are the cornerstone findings suggesting the influence of low-level features in oculomotor control;
he found that the ILP within a word is modulated by its length: short words (i.e. ≤ 3 letters) had a higher probability of being initially fixated at the right edge as well as a higher probability of being skipped, while long words (i.e. ≥ 7 letters) were likely to be initially fixated at the center and had higher probability of being refixated. That is to say, word length information is used by the English readers for moving their eyes along the text.
3 The category of alphabetic languages in this context includes those languages that make use of letters (i.e.
vowels and consonants) to conform words/linguistic units that are separated by blank spaces, resulting in words/linguistic units of different length. Such category is not restricted to Indo European languages (i.e.
English, German, French, Spanish etc.), but also includes Thai, Hebrew & Uighur, languages which have been also the focus of eye movement studies during the past decades.
‧ 國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Other crucial findings from Rayner’s study were that when making forward saccades the readers’ eyes tended to systematically land between the beginning and the middle of a word as suggested by the distribution of initial fixations, and when making regressive saccades the readers’ eyes tended to land half way between the end and the middle of the word. The distribution of initial fixations was interpreted as an attempt to send the eyes towards the center of the upcoming word, but due to a mixture of influences the ILP was slightly before the intended location4. Rayner (1979) interpreted the distribution of initial fixations as evidence about the existence of a Preferred Viewing Location (PVL) between the middle and the nearest edge of the word, observation that is regarded as one of the primary indicators that eye movement control in reading English is word-based.
The PVL phenomenon was later replicated in reading German by Nuthmann and Kliegl (2009), who showed that the average fixation positions were close to the word center with a leftwards shift with increasing word length for the set of forward single fixation cases. Moreover, a similar analysis in reading Hebrew, a language naturally written from left-to-right, provided qualitatively similar results than those in reading other alphabetic languages (Deutsch & Rayner, 1999). These set of results allowed scholars to validate the PVL phenomenon across alphabetic languages; consequently, interword spaces are regarded as salient visual cues readily available from parafoveal vision that are used in deciding where to send the eyes next (see Figure 1).
4 The influences that explain a shift from the intended target have multiple sources, errors in the oculomotor program or systematic errors as the saccade range effect (McConkie et al., 1988).
‧ 國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Figure 1. Cross language validation of the PVL phenomenon
Note: The red points represent the mean ILP of all interword forward saccades for different word lengths
Thus fixations within alphabetic words of different length is somehow systematic, as suggested by the observation that readers eyes tended to initially land about halfway between the beginning and the middle of words (see Table 1). Rayner (1979) labeled this location as the PVL which is different from the Optimal Viewing Position (OVP) described by O'Regan (1981), the latter being the position within a word in which recognition time is shorter. Thus the saccade target in reading alphabetic languages is assumed to be the word center which usually corresponds to the OVP, but the eyes normally land short of the OVP onto the PVL due to oculomotor noise in saccade programming and oculomotor execution.
In this way, the OVP makes reference to the location in a word where an optimal reading performance is expected and the PVL makes reference to the position within a word where the eyes actually land.
‧ 國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Table 1. Mean ILP by word length and language
Word length (letters)
Language Author Frequency 3 4 5 6 7 8 9 10
English
Rayner (1979) – 2.1 2.4 2.8 3.2 3.5 3.8 – –
Rayner et al (1996) LF – – 2.8 3.0 3.7 4.1 4.4 4.7
Rayner et al (1996) HF – – 2.7 3.2 3.6 4.1 4.4 4.5
Hebrew Deutsch & Rayner (1999) – 2.3 2.7 2.8 3.2 3.5 3.8 – – German Kliegl & Underwood (2009) – 1.7 2.1 2.5 2.9 3.0 3.2 3.6 3.5 Uighur Yan et al. (2014) – 2.0 2.4 2.7 3.2 3.3 3.8 – –
Note: English and German are languages naturally written from left to right; Hebrew and Uighur instead, are language that are written from right to left. The column Frequency makes reference to the word frequency;
in Rayner et al. (1996) Low Frequency (LF) words had frequencies of 10 per million or less, and High Frequency (HF) words had frequencies of 50 per million or larger.
Finally, it is worth to mention the correlation between word length and word frequency, two word properties that are often used as independent variables in reading studies. In order to assess whether word frequency influences the decision of where to move the eyes next, Rayner, Sereno, and Raney (1996) analyzed the ILP in words of different lengths with two levels of frequency (i.e. words with frequencies of 10 per million or less, Low Frequency words, and words with frequencies of 50 per million or greater, High Frequency words). As can be seen in Table 1, the mean ILP for words of the same length varying in frequency was not significantly different from each other, dismissing the role of word frequency on saccade targeting mechanism.
‧
The selection of a target location within a word can not only be discussed in terms of ILP but also as consequence of the take-off location distance of the immediately previous fixation, referred as the launch site. McConkie et al. (1988) confirmed that the eyes are usually send to a location between the beginning and the middle of a word, they also realized that this position systematically varied as a function of the distance from the previous fixated location up to the target word’s beginning, in such way that for larger launch site distances the mean ILP shifted leftwards while for shorter launch site distances the mean ILP shifted rightwards (see Figure 2).
Figure 2. Mean ILP by word length distinguishing incoming saccades from different launch sites, data from McConkie et al. (1988)
Figure 2 shows how the mean ILP shifts rightwards as the launch site gets closer to the target word for words of different length, observations that make of the launch site a
0
‧ 國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
variable of relative importance when analyzing the distribution of initial fixations and making inferences about the saccade targeting mechanism. Accordingly, saccades send from larger launch sites will result in initial fixations closer to the beginning of words, conversely shorter launch sites will result in initial fixations further into the word. Hence launch site could be arguably considered to be the correlate of an oculomotor constraint in reading, thus indicating that the eyes might have a tendency of making saccades of certain amplitude, distance that matches the perceptual span or the window from which detailed information can be successfully obtained.
As demonstrated by the moving window paradigm devised by McConkie and Rayner (1975), the amount of information that can be successfully retrieved in a given fixation is reduced exponentially with eccentricity; in that study the sentence display varied following the reader’s oculomotor pattern in such a way that some of the letters within a sentence were masked based on the current position of the eyes. Thus and for a given fixation, only certain parts of the sentence were unmasked or available for the reader.
Through systematic variation in the amount of available text to the left and right of the currently fixated point, it was possible to establish that word length patterns can be acquired further into the peripheral vision up to approximately 3° of Visual Angle (VA), and that oculomotor guidance was not influenced by information beyond 12 letter positions.
The previous observations redefined the PVL as the maximum point in the distribution of all the initial fixations within a word, being a composite distribution in nature.
McConkie et al. (1988) noticed that given a constant launch site and word length the ILP distribution in a word was normally distributed and from this acknowledgment, scholars redefined the PVL as a launch site contingent distribution. Accordingly, the PVL curve is
‧ 國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
a weighted average across several launch site contingent distributions, with word length only having a small effect on the means and variability of the ILP distribution. These findings imply a tendency to send the eyes towards a center of gravity influenced by the luminance patterns and constrained to a normal distribution within a word (Tsai, 2014;
Yang & McConkie, 1999).
Lack of Interword Spaces
One implication of a word-based saccade strategy is that readers use the word boundary information retrieved from parafoveal vision and then use it for selecting the next saccade target (Rayner et al., 1996). In English as well as in other alphabetic languages, the knowledge about the beginning and end of words is provided by interword blank spaces, but to what extent is this cue necessary for efficient oculomotor guidance? In order to assess the effect of omitting word boundary information in the selection of saccade targets Rayner et al. (1998) presented sentence in which the blank spaces were either filled with x or removed (see Figure 3). Under these unusual reading situations, the reading rate decreased nearly 50% due to an increase in fixation durations, a shortening in the length of forward saccades and an increase in the number of regressions; additionally, the PVL curve shifted towards the word beginning and become a negative linear slope.
Therefore semantic boundaries indicated by interword spaces were found to play an important role in deciding where to send the eyes next in reading alphabetic languages. The omission of blank spaces from the written script greatly disrupted the reading performance, increasing word identification times and shortening saccade lengths; in other words, both temporal and spatial characteristics of the eyes’ behavior were found to be affected by the
‧ 國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
omission of interword spaces (Rayner et al., 1998). Therefore, interword spaces have a special status in reading alphabetic languages, being salient cues that are used to isolate words parafoveally, information that is used for the readers to move their eyes along the text with a role in the specification of saccade targets.
The PVL curves for words in unspaced conditions in Rayner et al. (1998) no longer followed a Gaussian shape but turned into a negative linear slope decreasing from the word beginning, finding that was later replicated by Kanjii, Nazir, and Osaka (2001) in reading Japanese. Japanese is a language naturally written without interword spaces that includes the simultaneous use of two syllabic scripts or Kana (i.e. Hiragana and Katakana), used to write particles, auxiliary verbs and grammatical elements, and loaned Chinese characters or Kanji, used to write nouns, verbs, adjectives and some adverbs (Trigger, 2004).
The study of Kanjii et al. (2001) suggested that in reading languages without clear interword spaces, readers might use a different set of visual cues to aid the ongoing language processing and the saccade targeting mechanism; in order to explain this negative linear slope without disregarding the role of semantic units in oculomotor control, they suggested that the simultaneous use of Kana and Kanji characters provide word segmentation cues that can be effectively used by Japanese readers to segment words parafoveally. This is so because the beginning of Japanese words is usually occupied by Kanji characters, which are visually more salient than other grammatical components written in Kana characters, low-level feature that was argued to add attractiveness to the word beginning to be selected as the saccade target.
A subsequent study testing the effect of inserting blank spaces in Japanese gave support to the previous remarks. Sainio, Hyönä, Bingushi, and Bertram (2007) asked native
‧
Japanese speakers to read Pure Hiragana or mixed Kanji – Hiragana sentences with and without interword spaces (see Figure 3). When reading Pure Hiragana text eye guidance was facilitated by the insertion of spaces, but when reading mixed Kanji – Hiragana text the PVL at the word beginning was not significantly affected by the insertion of spaces.
These findings suggest that interword spacing in reading natural Japanese provide redundant information because the visually salient Kanji characters, which predominantly appear at the beginning of words are effective segmentation cues per se.
Normal spacing
Figure 3. Spacing conditions in Rayner et al. (1998) and Sainio et al. (2007)
Note: The English translation for the Japanese sentence is “The management of Narita airport can by no means said to be excellent”
On a different account the analyses on ILP in reading spaced and unspaced Thai, an language that is naturally written without inter word boundaries, evidenced that spacing
‧ 國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
neither facilitated nor disrupted saccade targeting or early lexical segmentation and the ILP distribution for both conditions peaked at the left from the word center (Winskel, Radach,
& Luksaneeyanawin, 2009). The similar patterns of ILP distribution for both spacing conditions indicated that parafoveal word segmentation in reading Thai might rely on information other than spaces.
One answer that supports this suggestion, came from diagnosing the probability that letters had of occupying the word’s beginning or end within the Thai lexicon; according to the analysis performed by Reilly, Radach, Corbic and Luksaneeyanawin (2005), 10 letters accounted for a 77% of all word ending letter positions and was equally high for a set of 10 letters to occupy the word beginnings (54%). Given the probability of certain letters to appear either at the beginning or at the end of words, it is likely that native Thai readers become aware of it and use this information as word boundary cues in the absence of clear spaces between words.
Thus, the PVL in languages naturally written without spaces, such as Japanese and Thai, was found to be at the word’s beginning, indicating that the mechanism underlying parafoveal specification of saccade targets might rely on traits of diverse nature in reading different writing systems. Given these observations a word-based strategy might not universally rely on the presence of blank spaces, instead is likely to be constrained by the visual layout and the familiarity of native readers with specific word segmentation cues of their language, leaving open to research the spectrum of traits that can be effectively used to guide the eye movements in reading different writing systems.