Ординатура / Офтальмология / Английские материалы / Eye Movements A Window on Mind and Brain_Van Gompel_2007

Finnish. In the next section we briefly discuss these studies. In the subsequent section we summarize those Finnish compound studies in which the focus is on foveal processing, but in which the investigated variables are at the intersection of morphology and visual acuity.

1. Parafoveal processing of constituents in compound words

Hyönä and Pollatsek (1998) compared the processing of long 1st constituent compounds (e.g., MAAILMAN/SOTA ‘world war’) with that of short 1st constituent compounds (e.g., YDIN/REAKTORI ‘nuclear reactor’) by matching them on overall length and frequency. They argued that if the constituents are processed similarly to separate words, the location of initial fixation location would be located further in the word when the 1st constituent is long, since the preferred landing position on words is shown to be somewhat left of the word centre (O’Regan, 1992; Rayner, 1979). Because a saccade into a word is programmed when the eyes are fixated on a preceding word, a constituent length effect on first fixation location would imply morphological parafoveal preprocessing. Hyönä and Pollatsek found that first fixation location was not dependent on the length of the 1st constituent, but second fixation location was further in the word when the 1st constituent was long. These findings imply that morphological structure is recognized once the word is fixated, but not before that (i.e. parafoveally). However, it is possible that a morphological preview effect does not show itself in the initial fixation location (i.e., that morphological processing does not affect where the eyes are sent).

In principle, it is possible that a morphologically complex word appearing in the parafovea affects the processing of the preceding word, Word N –1 (a so-called parafoveal- on-foveal effect). Hyönä and Bertram (2004) reanalysed the data of five compound word experiments in order to find out whether such parafoveal-on-foveal effects exist. Even though they found some impact of the frequency of the 1st constituent of the parafoveally presented compound words, their results were inconsistent in that sometimes a positive, sometimes a negative, and sometimes no frequency effect was found; in addition, the generalizability was questionable in that often the item analyses remained non-significant. However, post-hoc regression analyses showed that the 1st constituent of parafoveally presented short compound words might affect the processing of Word N –1, with a low frequency constituent attracting a fixation towards it more readily than a high frequency 1st constituent, leading to a shorter gaze duration on Word N –1 in the former compared to the latter case.

In sum, both studies assessed the issue of morphological preprocessing to some extent. One of them did not reveal a morphological preview effect, the other hinted at such an effect for short compound words, but the paradigm and the dependent measures used may not have been the most optimal to reveal morphological preview effects. The eye contingent change paradigm as employed in the present study more directly tested morphological preprocessing either by presenting a morphological preview or not.

2. Foveal processing of compound words and the visual acuity principle

Two-constituent compounds can be (a) long (e.g., TULEVAISUUDEN/SUUNNITELMA ‘future plan’); (b) short (e.g., HÄÄ/YÖ ‘wedding night’); (c) relatively long but with a short 1st constituent (e.g., YDIN/REAKTORI ‘nuclear reactor’); or (d) relatively long with a long 1st constituent (e.g., MAAILMAN/SOTA ‘world war’). It is quite clear that the four types of compound words described here differ greatly as to which morpheme would be submitted to detailed visual analysis during the initial fixation on the word. For word type (a) it may be unlikely to have all characters of the initial constituent in foveal vision; word type (b) will have all characters of both constituents in foveal vision; word type (c) will have all characters of the 1st constituent in foveal vision in the great majority of initial fixations; and word type (d) will have the 1st constituent in foveal vision as long as the first fixation is around the centre of the 1st constituent. Even though it may seem obvious that the length of a complex word or a morphemic unit within a word may significantly modulate the role morphology plays in lexical processing, it is only recently taken into consideration when assessing the role of morphology in word recognition.

Hyönä and Pollatsek (1998), Bertram and Hyönä (2003), and Hyönä, Bertram, and Pollatsek (2004) have demonstrated solid effects of 1st constituent frequency on the identification of relatively long two-constituent compounds (12 or more characters). These 1st constituent frequency effects appeared from the first fixation onwards, implying that the 1st constituent is already involved in the early stages of the word recognition process. For short compounds (7–8 characters), the identification process is much more holistic with early and late effects of whole-word frequency, but only a later and statistically marginal effect for 1st constituent frequency (Bertram & Hyönä, 2003). On the basis of these results, Bertram and Hyönä argued that compound-word length indeed modulates the role of morphology due to visual acuity constraints of the eye. Thus, all letters of short words fall into foveal vision and can be identified in a single fixation. Since whole-word access is visually possible and in principle faster than decomposed access (for one thing, one does not need to parse out the individual constituents in case of whole-word access), the holistic access procedure dominates the identification process of short compounds. In contrast, when reading long compound words, a second fixation is needed (typically 80–90% of the time) to make the letters of the 2nd constituent available for foveal inspection. This, in combination with solid 1st constituent frequency effects from the first fixation onwards, suggests that recognition of long compound words is achieved by first accessing the initial constituent followed by the access of the 2nd constituent and that of the whole-word form.

Bertram, Pollatsek, and Hyönä (2004) showed that the identification of a relatively long 1st constituent (7–9 characters) is problematic, when there are no cues to indicate the constituent boundary, but identification can proceed smoothly when a boundary cue is present – for instance, when two vowels of different quality appear at the constituent boundary. In Finnish, vowels of the same type (e.g., all vowels are front vowels) always appear in monomorphemic, inflected and derived words, but vowel quality can differ in

4. Method

4.1. Participants

Twenty-four university students took part in the experiment as part of a course requirement. All were native speakers of Finnish.

4.2. Apparatus

Data were collected by the EyeLink eyetracker manufactured by SR Research Ltd. The eyetracker is an infrared video-based tracking system combined with hyperacuity image processing. There are two cameras mounted on a headband (one for each eye) including two infrared LEDs for illuminating each eye. The headband weighs 450 g in total. The cameras sample pupil location and pupil size at the rate of 250 Hz. Registration is monocular and is performed for the selected eye by placing the camera and the two infrared light sources 4–6 cm away from the eye. The spatial accuracy is better than 0 5 . Head position with respect to the computer screen is tracked with the help of a head-tracking camera mounted on the centre of the headband at the level of the forehead. Four LEDs are attached to the corners of the computer screen, which are viewed by the head-tracking camera, once the subject sits directly facing the screen. Possible head motion is detected as movements of the four LEDs and is compensated for on-line from the eye position records.

4.3. Materials

The experiment was conducted in Finnish. The target words were embedded in single sentences that extended a maximum of three lines (the target word always appeared sentence-medially in the first text line). An eye movement contingent display change paradigm (Rayner, 1975) was employed. An invisible boundary was set three character spaces to the left from the target word (see also Balota, Pollatsek, & Rayner, 1985; Pollatsek, Rayner, & Balota, 1986), as readers typically do not fixate on the last two letters of a word. When the eyes crossed this boundary, the target word was changed into its intended form. It took an average of about 13 ms to implement the change once the eyes crossed the invisible boundary. Thus, when the target was foveally inspected, it always appeared in the correct form. Two parafoveal preview conditions were created for each short and long 1st constituent compound: (a) a full preview (when crossing the boundary, the word was replaced by itself), and (b) a partial preview (the initial 3–4 letters were preserved, but the rest was replaced with random letters). In the partial preview condition, the short 1st constituent was visible in its entirety, whereas only about half of the letters of the long 1st constituents were parafoveally available. In Table 1 an example of all four conditions is presented.

Fifty compounds with a short (3–4 letters) initial constituent were paired with fifty compounds with a long (8–11 letters) initial constituent. A sentence frame was prepared

‘John’s current physical capacity is lower than ’

for each pair so that the sentence was identical up to the target word; the length of Word N +1 was matched. Two stimulus lists were prepared so that in one list a given compound word appeared in the full preview condition and in the other list in the partial preview condition. Thus both lists contained 25 short and 25 long initial constituent compounds in each of the two preview conditions. The presentation of the stimulus lists was counterbalanced across participants. Using the newspaper corpus of Turun Sanomat comprising 22.7 million word forms (Laine & Virtanen, 1999), the two conditions were matched for whole word frequency, word length, average bigram frequency, and initial trigram frequency, but it was not possible to perfectly match for the frequency of the 1st and 2nd constituent (in Finnish, and probably in any language, short lexemes tend to be much more frequent than long ones). The lexical-statistical properties of the two conditions are found in Table 2. The sentences were presented in 12 point, Courier font.

Table 2

Lexical-statistical properties of the two compound conditions

aLong1c = compounds with a long 1st constituent; short1c = compounds with a short 1st constituent.

With a viewing distance of about 60 cm, one character space subtended approximately 0 3 of visual angle.

4.4. Procedure

Prior to the experiment, the eye-tracker was calibrated using a 9-point calibration grid that extended over the entire computer screen. Prior to each sentence, the calibration was checked by presenting a fixation point in an upper-left position of the screen; if needed, calibration was automatically corrected, after which a sentence was presented to the right of the fixation point. Participants were instructed to read the sentences for comprehension at their own pace. They were further told that periodically they would be asked to paraphrase the last sentence they had read to ensure that they attended to what they read. It was emphasized that the task was to comprehend, not to memorize the sentences. A short practice session preceded the actual experiment.

5. Results

A total of 11.4% of the data was discarded from the statistical analyses. Almost all the discarded trials were excluded because the display change occurred too late (i.e., a fixation was initiated on the target word region before the display change was completed). Analyses of variance (ANOVAs) for participants and items were computed on several eye fixation measures, with display change (change vs no change) as a within-subject and within-item factor and 1st constituent length (short vs long) as a within-subject factor but a between-item factor. The condition means for the analyzed eye movement parameters are shown in Table 3.

5.1. Gaze duration

Gaze duration is the summed duration of fixations made on a word before fixating away from it (either to the left or right). An effect of 1st constituent length was significant in both analyses, but the display change effect was only significant in the participant analysis: 1st constituent length, F1 1 23 = 39 96, MSE = 668, p < 0 001, F2 1 98 = 4 02, MSE = 10109, p < 0 05, and display change, F1 1 23 = 4 42, MSE = 890, p < 0 05; F2 1 98 = 1 86, MSE = 4385, p = 0 18. Gaze duration was 33 ms longer for compounds with a long 1st constituent. The display change effect amounted to 13 ms. Most importantly, the 1st Constituent Length×Display Change interaction was not significant, F1 1 23 = 1 30, MSE = 871, p > 0 2, F2 < 1. If anything, gaze duration showed a non-significant trend in the opposite direction than was expected. That is, the display change effect was 20 ms for the short 1st constituent compounds and 6 ms for the long 1st constituent compounds, whereas we expected to find a larger benefit (in the form of a reduced preview effect) of the first 3–4 letters when these letters would constitute a real constituent as was the case for the short 1st constituent compounds. However, it should

a in milliseconds; b in character spaces.

be noted that separate t tests to assess the change effect were non-significant for both long and short 1st constituent compounds.2

5.2. First fixation duration

The duration of first fixation made on the word was reliably affected by 1st constituent length, F1 1 23 = 32 45, MSE = 145, p < 0 001, F2 1 98 = 17 51, MSE = 690, p < 0 001. First fixation was 14 ms longer when the initial constituent was short. The display change effect did not reach significance, F1 1 23 = 2 69, p = 0 12, MSE = 130, F2 < 1 1. The interaction was clearly non-significant, F1 2 < 1.

2 We made a peculiar observation that the difference in effect size may be related to some extent to long 1st constituents having on average slightly higher frequency initial trigrams. If ten items of each length group were excluded in order to perfectly match on initial trigram frequency, while preserving the matching on word frequency and length, the change effect on gaze duration was identical (15 ms) for long and short 1st constituent compounds. However, for both type of compounds the initial trigrams were identical in the change and no change condition, so that a straightforward explanation for this observation cannot be given. One possibility is that low initial trigrams attract attention somewhat earlier, making the display change slightly more salient.

5.3. Second fixation duration

The duration of second fixation on the target was to some extent influenced by the length of the 1st constituent, although the item analysis remained non-significant, F1 1 23 = 8 93, MSE = 249, p < 0 01, F2 1 98 = 1 30, MSE = 1076, p > 0 2. The second fixation was 10 ms longer for the compounds with a long 1st constituent. The display change effect did not reach significance, F1 1 23 = 1 18, p > 0 2, MSE = 281, F2 = 2 23, MSE = 756, p = 0 14. The interaction was clearly non-significant, F1 2 < 1.

5.4. Third fixation duration

Items and participants with less than two observations per cell were excluded. Only 16 participants and 61 items contributed to the analyses. The duration of third fixation on the target was to some extent influenced by the length of the 1st constituent, although the item analysis remained non-significant, F1 1 15 = 10 50, MSE = 353, p < 0 01, F2 1 98 = 1 68, MSE = 1976, p > 0 2. The third fixation was 15 ms longer for the compounds with a long 1st constituent. The display change effect and the interaction were clearly non-significant, F1 2 < 1.

5.5. Number of fixations

In the number of first-pass fixations, both main effects proved significant in the participant analysis, but the display change effect was not significant in the item analysis: 1st constituent length – F1 1 23 = 43 03, MSE = 0 02, p < 0 001, F2 1 98 = 7 86, MSE = 24, p < 0 01; and display change – F1 1 23 = 5 60, MSE = 0 02, p = 0 03, F2 1 98 = 2 23, MSE = 0 13, p = 0 13. Readers made on average 0.21 fixations more on long than on short 1st constituent compounds. The display change effect amounted to 0.08 fixations. The 1st Constituent Length × Display Change interaction did not reach significance, F1 2 < 1. Analyses of the probability of a second and third fixation yielded exactly the same results.

5.6. Initial fixation location

There were no statistically significant effects, all F s < 1.

5.7. Location of second fixation

The second fixation was located 0.80 character spaces further into the word when the initial constituent was long, F1 1 23 = 26 14, MSE = 59, p < 0 001, F2 1 98 = 18 88, MSE = 2 20, p < 0 001. The main effect of display change was marginal in the participant analysis, but significant in the item analysis, F1 1 23 = 3 13, MSE = 75, p = 0 09, F2 1 98 = 6 20, MSE = 1 36, p = 0 01. The second fixation was positioned 0.31 characters further into the target word when there was no display change. The interaction remained non-significant.

5.8. Duration of fixation prior to fixating the target

The duration of the final fixation prior to fixating the target word showed no reliable effects. There was a non-significant 4 ms difference between the short and long 1st constituent compounds, F1 (1, 23) = 2.99, MSE = 169, p = 0 10, F2 < 1. The previous fixation was slightly longer before fixating a long 1st constituent compound. Other effects were non-significant, all F s < 1.2.

6. Additional analyses

6.1. Launch site analyses

We tested the possibility of finding evidence for morphological preprocessing when the target word is processed parafoveally from a close distance prior to inspecting it foveally. We did this by conducting a 3-way ANOVA including launch site as a within factor. Trials were categorized into near (within six letters of the target word, 42% of data) and far (more than six letters from the target word, 58% of data) launch site. Apart from main effects of launch site in several measures (e.g., 18 ms longer gaze durations and initial fixation 2.8 characters closer to the word beginning for trials launched from far away), no significant interactions between launch site and constituent length were found (all ps >0.13). This was also true for other categorizations (e.g., defining near as within three or four letters to the left from the invisible boundary). In other words, similarly to Kambe (2004), there was no sign of a morphological preview benefit even when the eyes were fixated close to the target compound prior to its foveal inspection. The only interaction approaching significance in the item analysis was a Preview × Launch Site interaction, indicating that the preview effect was larger for near launch site (24 ms) than for far launch site (8 ms), F1 1 22 = 1 38, MSE = 5440, p = 0 25, F2 1 55 = 2 83, MSE = 7431, p < 0 10.

6.2. Factors that predict gaze duration on the target compound

Using the regression analyses technique proposed by Lorch and Myers (1990), we set out to determine what factors predict gaze durations on short and long 1st constituent compounds. Separate analyses were conducted for the two types of compounds since 1st and 2nd constituent frequency distributions were different for short and long 1st constituent compounds. The regression analysis was performed separately for each participant, after which one-sample t tests on the unstandardized regression coefficients were conducted for four predictor variables. The four predictor variables were initial trigram frequency of 1st constituent, 1st constituent frequency, 2nd constituent frequency, and whole word frequency (log-transformed values were used for lexical frequencies). For long 1st constituent compounds, both the 1st constituent frequency and initial trigram frequency turned out to be significant predictors, whereas whole word frequency was

a marginally significant predictor (1st constituent frequency, t 23 = −5 35, p < 0 001; initial trigram frequency, t 23 = 2 35, p < 0 05; whole word frequency, t 23 = −2 03, p < 0 06). However, 2nd constituent frequency did not predict gaze duration for long 1st constituent compounds, t < 1. In contrast, for short 1st constituents compounds, 2nd constituent frequency and whole word frequency, but not initial trigram frequency or 1st constituent frequency, turned out to be significant predictors of gaze duration (1st constituent frequency, t < 1; initial trigram frequency, t < 1; 2nd constituent frequency, t 23 = −4 10, p < 0 001; whole word frequency, t 23 = −2 89, p < 0 01). The mean coefficients of all variables and their standard errors in the one-sample t tests can be found in Table 4. It appears from Table 4 that, for instance, for long 1st constituent compounds an increase of 1 log unit in 1st constituent frequency results in a decrease of 54.84 ms in gaze duration.

7. Summary of results

To sum up, preserving the first 3–4 letters of relatively long compound words while replacing the remaining letters with random letters brought about a 13 ms preview effect in gaze duration, which is in line with other studies in which the first few letters were preserved and the remaining letters were replaced with visually dissimilar letters (e.g., Inhoff, 1989; Kambe, 2004). However, the above analyses revealed no evidence supporting parafoveal morphological preprocessing. As morphological preprocessing seemed possible only for the short initial constituent compounds, a partial preview was expected to be more beneficial for short than for long 1st constituent compounds (e.g., the difference between full and partial preview was predicted to be smaller for short 1st constituent compounds). Our data did not provide support for this prediction. In all measures the change effect was moderate and not modulated by the length of the 1st constituent. Even

Table 4

Estimates of mean unstandardized regression coefficients and their standard errors for gaze duration of long 1st constituent compounds (e.g., KONSERTTI/SALI ‘concert hall’) and short 1st constituent compounds (HÄÄ/SEREMONIA ‘wedding ceremony’)