"The-skipping" revisited in French : programming
saccades to skip the article "les".
Vincent Gautier (1)
J. Kevin O'Regan (2)
J.F. Le Gargasson (1)
(1) Laboratoire de Biophysique de la Vision, INSERM, Université Paris VII, Faculté de Médecine Lariboisière Saint-Louis,
10 Avenue de Verdun, 75010 Paris.
(2) Laboratoire de Psychologie Expérimentale,
CNRS, EPHE, EHESS, Université Paris V, 28 rue Serpente, 75006 Paris.
Three experiments were performed to verify O’Regan’s (1979) finding that in reading, the eye moves further forward when going towards the word "THE" than when going towards a three-letter verb. The experiments were performed in French instead of English, and compared the plural article "les" with different three-letter verbs. It was confirmed that the eye did indeed move about 1.5 letters further in the case of the article "les". Further investigation of the phenomenon suggested that the effect was present even when the prior fixation duration was short: Only when prior fixation was around 200 ms or less, was there a suggestion that the "les"-skipping effect disappeared.
More than fifteen years ago, brought forward evidence that, in reading, the eyes will tend to "skip" a frequent word like the word "THE" more often than more rarely occurring three-letter words like certain verbs.
Although referred to the phenomenon as "THE-skipping", this is in fact a misnomer. As shown in Figure 1 which reproduces O'Regan's result, actually both the three-letter verbs and the word "THE" tended to be skipped: however, the eyes moved about 0.8 - 1.5 letters further rightwards in the case of "THE" than in the case of the verbs.
Insert Figure 1 here
Despite the fact that the effect was small and tested only for verbs, it sparked the imagination of authors interested in linguistic processing and has often been quoted. Presumably this is because the result was an example of a case where the behavior of the eyes depended immediately on linguistic processing of material in parafoveal vision: the existence of such an immediate link between linguistic processing and eye movements raised the exciting possibility that eye movements might constitute a moment to moment indicator of the lexical and linguistic processing underlying reading.
In fact however the study does not guarantee of the existence of such an immediate control of eye movements: it may be that the "THE-skipping" phenomenon actually was present only in a particular subset of the data in which the duration of the fixation that occurred before the jump towards the critical "THE" or verb was particularly long, or in which more than a single fixation had occurred in the previous word, thereby allowing more time for processing of the parafoveal "THE" or verb.
An unpublished pilot experiment reported briefly in a french-language replication of the "THE-skipping" experiment using the french article "le" instead of the english "THE" had indeed suggested that skipping occurred only in a subset of the sentences used, and then only when prior duration was long.
Other experiments in the eye movement literature do not elucidate the issue (see review by ). , (second experiment) failed to find evidence of word skipping as a function of parafoveal word content. Certain studies did observe skipping , , . However as was the case in study, the authors did not consider whether the observed skipping behavior might be restricted to the subset of cases where the prior fixation duration was long. Furthermore, in some of these studies the length of the word preceding the to-be-skipped word was not the same for the two word types that were being compared. Skipping of one test word type might have been observed because for that test word type, the length of the preceding word happened to be systematically shorter ? The eye would then tend to have been launched from a position closer to the to-be-skipped word, and the probability of skipping will have been larger.
Evidence concerning variations in saccade size (instead of skipping probability) is again inconclusive. On the one hand, several studies have shown that the position where the eye lands in a word is not affected by linguistic characteristics of the to-be-fixated word such as the word's predictability from the linguistic context or such as the distribution of information in the word, as measured by the statistical dependency between sequences of letters , , . The impossibility of modulating saccades as a function of parafoveally acquired lexical information was also suggested by studies showing that the landing site in a word to be fixated is the same both in conditions when the word is fully readable in parafoveal vision and in conditions when it is initially masked and only becomes visible when the eye lands upon it , , , . On the other hand two studies report that the distribution of information within words does affect landing position , , . However, in , the position from which the eye was launched was not controlled, and in , the prior fixation duration was very long and the observed difference in saccade size was only a fraction of a letter.
Overall then, the evidence for the existence of an immediate influence of linguistic information available in parafovea on saccades is weak. This, and other arguments, led to postulate that most of the time during reading, the eye is not being directly controlled by ongoing linguistic processing, but rather by a global scanning strategy whose execution does not need detailed linguistic processing of the words being scanned (the "Strategy-Tactics" theory). Other authors, notably and nevertheless favor a theory in which eye movements are governed by the instantaneous movements of attention across the line of text. But certainly whatever theory we adhere to, it seems reasonable to expect that when prior fixation durations are so short that there is no time for linguistic processing, skipping of a parafoveal word cannot occur differentially as a function of its lexical properties. However, beyond some critical duration we expect that skipping may start to occur. The exact value of the critical duration may be influenced by the reader's prior knowledge or expections of the text, since this facilitates lexical processing. Also, in the Strategy-Tactics theory, where eye movements are not generally assumed to be strongly linked to lexical processing, the critical duration may be expected to be much greater than in attention-based theories.
The present experiments were designed to further investigate these issues in a french "remake" of the original 1979 "THE-skipping" experiment. The french plural article "les" was used instead of the english "THE", and eye movement behavior on this was compared to that for a number of three-letter verbs ("fût", "cru", "est",...). We used two extreme situations: one situation was the mixed reading condition: it was similar to the original experiment , and close to normal reading, where the subject does not know in advance what the words in peripheral vision are. We here expected to obtain a result like that in the original study , with a fairly weak "THE-skipping" effect, possibly corresponding essentially to the subset of cases where prior fixation duration is long.
The other condition used in the experiment was the blocked reading condition, in which Subjects repeatedly saw the same type of sentence structure. Here Subjects have prior knowledge about the words that may appear in peripheral vision. This condition presumably diminishes the difficulty of lexical processing and makes it likely that processing will have time to catch up with the ongoing scanning strategy. By using blocked reading we hoped to create ideal conditions for any potential "THE-skipping": we hoped "THE-skipping" would be stronger, and that it would occur even in cases when the prior fixation duration is in the normal range.
Since a major problem with previous work has been the failure to control for the duration of the fixation preceding the saccade towards the critical words, a further purpose of the present experiment was to do a post hoc analysis of the data as a function of the durations of the fixations prior to the saccade leading to (or skipping over) the test word "les" or verb. It seems reasonable to expect that there should be a critical fixation duration tcrit for fixations prior to a saccade, before which lexical processing can have no influence on the saccade, whereas above this duration, we should start to see differential behavior as a function of the processing that is done on the words in peripheral vision, with "THE-skipping" beginning to emerge. Furthermore, the critical duration tcrit should be earlier in the case of the blocked reading condition than in the case of the mixed reading condition. These predictions are schematised in Figure 2.
Insert Figure 2 here
Though not essential to the purpose of the present paper, two additional test word types were also used in the experiment, both in the mixed and the blocked conditions: one where the "word" in peripheral vision was a string of three x’s ("xxx" word type), and one where a blank space of width three letter-spaces was used ("space" word-type). The motivation for using these conditions was related to the problem of the "center of gravity" effect. It is known when attempting to saccade towards a target, the eye tends to be deviated away from the target location by the presence of surrounding stimulus materials, landing near a kind of weighted center of gravity of the global visual configuration surrounding the target , . Applied to the problem of "THE-skipping", the implication of the center of gravity effect is that the amount of skipping that can actually occur is limited: even if the eye "wants" to skip a word, because of the visual "weight" that it represents, the eye may not be able to move completely over it and into the next word. The space and "xxx" test string types give us a measure of the strength of center of gravity effects in the present experiment.
The space type test string gives a measure where we expect the eyes are "trying" to go, and would go if it were not for the constraints imposed by the center of gravity effect : since no linguistic processing at all is needed to see that there is a space in parafoveal vision, and since the visibility of the space is very good, we expect that the desired target location is the word beyond the space. The observed landing location is presumably close to the position that the eye would like to attain in the case when the test word is to be skipped.
The "xxx" test string is presumably also very
easy to see, given its repetitive, characteristic pattern. We expect that
very little processing in parafoveal vision is necessary to ascertain that
this string can be skipped (particularly in the blocked condition, when
readers know in advance what to expect). Yet the "xxx's" provide a visual
"weight" in the center of gravity calculation which is presumably quite
similar to that for a to-be-skipped word. The position the eye attains
in the "xxx" case is therefore an indication of the furthest rightmost
position that it could be expected to attain in the case of skipping.
All the sentences used were short (less than 35 letters) and constructed following the same pattern: a five-letter word, a three-letter test word (or string), a nine-letter word, followed by one or more words to end the sentence. The test word could either be a three-letter verb (e.g. the word "fût" in "Roger fût proviseur de collège"), the article "les" (e.g. in "Range les vêtements proprement"), the string "xxx" (e.g. "Odile xxx économise beaucoup") or a space (e.g. "Icare pulvérise ses records"). Half of the sentences were semantically correct, and half were anomolous (e.g. "litre les escalader moelleux"). The Subject's task was to read the sentence and press a button to indicate whether the sentence was or was not anomolous.
Half of the correct sentences and half of the anomolous sentences were randomly attributed to the mixed condition, and the other half groups of correct and anomolous sentences were attributed to the blocked condition. This random attribution was done anew for each Subject. Each Subject read 80 sentences in each of the mixed and blocked conditions. Half the subjects read the mixed condition first, and half read the blocked condition first.
In the blocked group, Subjects read 20 sentences of a given type in succession (e.g. the type with the three-letter verb as test word), followed by 20 sentences in succession of each of the other types. In this way, the Subject came to know what kind of word to expect in the test word position.
The whole test part of the experiment was preceded
by a group of 15 practise sentences having structures identical to those
used in the test sentences. The experimenter explained that these four
types of structures were going to be used in the subsequent test.
Horizontal eye movements were measured monocularly by a photoelectric limbus tracking device mounted on a helmet. An infra-red image of the eye that was reading (the other eye being masked) was reflected through an infrared mirror and focussed by adjustable optics on a group of phototransistors. The analog signal was sampled by an Acorn Archimedes computer every 10 ms.
The Subject's head was stabilised in a chin and fore-head rest with the eyes at 45 cm from the computer screen. The sentences were displayed in the middle of the screen, in mixed case, using an 8 x 8 pixel character font, in white on dark, each letter subtending approximately 1/3 degree.
At the beginning of the experiment a calibration was done in which the Subject fixated three targets distributed horizontally on the screen over the area where the sentence was to appear and the Experimenter adjusted the apparatus to give an accurate signal. This calibration was verified every 10 sentences.
Before each trial, a vertical bar appeared on
the screen, corresponding to the position where the middle of the first
word of the sentence would subsequently appear. The subject looked at the
bar, and a moving cursor appeared which indicated to the subject and the
experimenter where the computer instantaneously estimated the subject’s
eye to be looking. If this position corresponded exactly to the position
of the bar (to within plus or minus 1/4 letter, the whole sentence was
displayed. If the moving cursor and the bar did not coincide, the subject
could slightly adjust his head position to make them correspond, and the
sentence appeared. If the experimenter judged that too large a correction
had to be made (offset between cursor and bar more than a few letters),
he interrupted the sequence of trials and a new calibration procedure was
performed before recommencing.
7 native french speakers with normal or corrected-to-normal
vision, aged 22-52 years, took part in the Experiment.
The Subjects’ accuracy in making the semantic correctness judgements was good. Percentage of errors ranged from 4% to 8%, depending on Subject. All records were retained for analysis since the portion of the eye’s behaviour that interested us was presumably unaffected by whether or not the Subject correctly interpreted the sentence.
We were interested in the position that the eye attained in (or beyond) the test word after leaving the initial fixation point in the first word. However in looking at the data it appeared that the Subjects tended not to move directly out of the first word. Instead, at the moment that the sentence occurred, they made a small "repositioning" movement before making a saccade out of the first word. Since such repositioning movements were not of interest to us, we chose to ignore them in the data analysis. We selected only data with a first saccade leaving the first word (63.2 % of the data).
Figure 4 shows histograms of the eye’s landing position in those cases where after the sentence presentation the first saccade leave the first word, for the four test word types and in the blocked and mixed conditions. Figure 5 gives the associated mean values.
Insert Figure 4 and Figure 5 here
The main question posed by this experiment is, first, whether the "THE-skipping" effect found in the original O’Regan (1979) experiment can be replicated, and second whether this becomes stronger in the blocked condition.
It is clear from the figures that in the mixed condition there is a THE-skipping effect, with the eye landing 2.1 letters further to the right for the "les" test word as compared to the "verb" case. Consideration of the blocked condition however, does not show a stronger THE-skipping effect. On the contrary, it is actually weaker (1.8 letters).
Indeed, an analysis of variance with factors les/verb and mixed/blocked showed only the global les/verb comparison to be significant (F(1,6)=9.3, p < 0.05), and no effect of Mixed/Blocked factor (F(1,6)=1.0, ns) or the interaction of the two (F(1,6) < 1, ns).
The behaviour for the test words "xxx" and "space"
are as expected. For the "space" test string, the eye goes significantly
further to the right than for the other word types. It seems reasonable
to suppose that (certainly in the blocked condition) this position is approximately
where the eye is aiming when it attempts to skip a word. The landing position
in the case of the "xxx" string, which the eye is presumably also trying
to skip, is not as far to the right, no doubt owing to visual "weight"
of the xxx's. It is indeed interesting to note that the landing positions
for "les" and "xxx" strings are very similar, suggesting that "les" is
being skipped as much as is possible given the oculomotor constraints imposed
by the center of gravity effect.
Landing position as a function of prior fixation duration.
The previous analyses looked at landing position independently of the duration of the fixation that precedes the saccade towards the test word. We expect that a more fine grained analysis should show that the difference between 'les' and verb conditions should only appear when the prior duration is longer than a critical duration. This critical duration should be longer in the mixed case than in the blocked case.
Figure 6 shows the results of an analysis of the eye's landing position as a function of the durations of the fixations prior to the saccades leading into the test words. Since the prior fixation durations were classified according to which 50-ms bin they occurred, each data point represents comparatively few measurements (from 1 to 29), and these measurements are not equally distributed over the subjects or the sentences: differences in the curves may reflect subject or sentence differences, so must be considered with caution. Nevertheless several aspects of the data are suggestive.
Insert Figure 6 here
First, compared to durations of about 250 ms normally found in reading, Subjects spent an abnormally long time fixating in the word preceding the test word (mean in blocked condition: 566 ms; mixed: 537 ms).
It is possible that these rather long fixation durations were caused by the particular procedure used in the experiment. In this, at the moment of appearance of the first word to be read at the initial fixation mark, Subjects were possibly still involved in positioning the cursor so as to align it with the mark. Whereas this should not have differentially affected the eye's behaviour going toward the different word types, it may have caused an overall longer fixation duration prior to leaving the fixation point.
Second, there appears to be a rising trend in the curves: the eye goes further when the prior fixation duration is long. Similar results had been observed in aiming for a target letter in a string of letters .
Finally, the most important feature of the curves
as concerns our predictions is the question of whether there is a critical
time beyond which the curves for the "les" and the "verb" type test words
separate, and whether this critical time is later in the mixed condition
as compared to the blocked condition: It is clear that there is no evidence
for such a trend.
At first sight the results of Experiment 1 seem to contradict our prediction that there should be an interaction between the THE-skipping effect and the mixed/blocked conditions. However we now see that actually the results of the experiment are not incompatible with our predictions: because of the very long prior fixation durations, both in the mixed and the blocked conditions, Subjects will have had sufficient time to analyse the word in peripheral vision and to define different desired fixation locations depending on whether the test word is "les" or "verb". Under these conditions it is natural therefore to observe THE-skipping behaviour in both mixed and blocked conditions. Furthermore, it is not surprising to observe that over the range of prior fixation durations of 400-700 ms plotted in Figure 6 there should be no evidence of a critical time at which curves for "les" and "verb" types separate.
In Experiment 2 we use a different method of stimulus
presentation which ensures more normal prior fixation durations.
Materials and Procedure.
The materials and procedure in this experiment were identical to Experiment 1, with one difference. So as to remove the problem of refixations in the first word and to ensure that the fixation duration prior to the saccade into the test word would be of normal duration, we modified the way the sentences appeared. Instead of having the eye initially fixating in the first word, we had the eye start at a fixation mark situated three character-spaces to the left of this word. The Subject fixated the mark, using slight head movements to adjust the moving cursor so it was on the mark. When the computer detected accurate fixation (to within half a letter, as before), the mark disappeared and was replaced by an "x", with the sentence to the right. The Subject then moved his or her eye off the "x" and read the sentence. By placing the "x" three character-spaces away from the first word of the sentence, we ensured that any small saccades that occurred would occur outside the first word.
Another slight change in procedure in Experiment
2 was the fact that in order to encourage Subjects to make use of their
prior knowledge of the sentence structures in the blocked conditions, at
the beginning of each new block, the experimenter informed the Subjects
of what the type would be. The order of the block types was random for
Eight new subjects participated in the experiment.
They were all students in the 20-35 age range.
The change in procedure was successful in removing most of the "repositioning" saccades in the first word, since when these occurred, they occurred in the space prior to the first word. However a small proportion of the records contained repositioning movements within the first word, and these records were removed, leaving only records where the first saccade made from the first word of the sentence led into (or near) the test word. Additionally, for technical reasons, some records could not be analysed because only a total of 3 saccades could be preserved in the computer memory. When there were two repositioning movements prior to jumping into the first word, the final, third saccade into the test word was lost. 28.8 % of the records had to be rejected for one of the above two reasons.
The proportion of incorrect responses to the semantic correctness judgement was 8.2 %. As in experiment. 1, all these records were nevertheless retained.
An analysis of the (single) fixation durations in the first word gave the histograms in Figure 7. In the mixed and blocked conditions the mean durations were 260 ms and 275 ms respectively. These durations are perfectly in the range found in "normal" reading, and much shorter than the values of more than 500 ms observed in Experiment 1. More detailed discussion of fixation durations prior to the saccade into the test word will be presented later.
Insert Figure 7 and Figure 8 here
Before analysing the landing positions near the four types of test words, we need to check that the eye is starting off from a comparable position for each type. The mean launch positions of the eye preceding the saccade into the test words are shown as the left group of symbols in Figure 8. The mean was very close to the center of the word preceding the test word, but there was a small difference as a function of the different test words "les", verb, "xxx" and space (F(3,21)=4,391; p<0.05), and no difference as a function of the blocked versus mixed conditions (F(1,7)=1.466; ns) or interaction of the two (F(3,21)<1, ns). We shall ignore the small differences between the word types since these are negligible in size compared to the effects we are considering in the landing positions (Figure 8).
Insert Figure 9 here
Figure 9 gives the histograms of landing positions near the different test words, in the mixed and blocked conditions. The symbols on the right in Figure 8 give the associated mean landing positions. As for Experiment 1, the question posed is, first, whether there is a THE-skipping effect, and second, whether this is stronger in the blocked condition than in the mixed condition.
It is clear that there is a THE-skipping effect. In the mixed condition the eye lands 1.2 letters further to the right in the "les" case than in the verb case. In the blocked condition this difference is now 1.5 letters. An analysis of variance with factors verb/les and blocked/mixed confirms that there is a significant global effect of the verb/les comparison (F(1,7)=7.35, p<0.05) and blocked/mixed comparison (F(1,7) = 6.88, p<0.05).
The results for the "xxx" and the "space" word
types are fairly similar to those for Experiment 1. For the space type,
the landing position is further to the right than for all the other types.
In the mixed condition, the landing position for the "xxx" word type is
close to the landing position for the "les" test word. This is similar
to the situation in Experiment 1. A difference with Experiment 1 concerns
the blocked condition, however, where the "xxx" type string behaves more
like the space type than like the "les" type. It would appear that in the
blocked condition of Experiment 2, where Subjects were informed by the
experimenter about the upcoming structure of the sentences, the skipping
of the "xxx" string was more effective and approached the location where
the eye landed in the case of the space type string. This shows that the
previous estimate of the maximum distance the eye can move, given oculomotor
constraints, may have been an underestimation: when Subjects are familiar
with the sentence structure, they can be more efficient in skipping the
Analysis as a function of prior fixation duration
Figure 10 plots the landing positions as a function of the duration of the prior fixation. One might imagine in Figure 10 that there is a critical moment around 180 ms where "les" and verb conditions separate in the blocked condition, but then in the mixed condition, this separation should occur at a later stage. If there is a separation however, it occurs at around 150 ms in the mixed condition, which it does not appear to do. Thus the evidence for the existence of a critical moment when the curves for "les" and "verb" types separate is weak.
Insert Figure 10 here
Experiment 2 has failed to confirm the predictions according to which the strength of the "THE skipping" effect should be modulated by the Mixed/Blocked manipulation. There is also no strong evidence that the effect appears only when the duration of the fixation prior to the saccade into the test word is longer than a critical duration.
Contrary to the case of Experiment 1, the explanation of these failures cannot lie in a ceiling effect caused by long durations of the fixations preceding the saccade into the test word, since here in Experiment 2 the durations are similar to those found normally in reading, around 250 ms.
A peculiar fact about Experiment 2 concerns the
global effect of the Blocked/Mixed manipulation, with the eye going less
far in the blocked condition: this is contrary to expectation, since it
is in the blocked condition that linguistic processing will have been more
facilitated. The fact that the difference is present even for the space
type string suggests that the effect is not related to improved linguistic
processing, but rather to a difference in overall eye movement strategy
adopted by readers in the blocked condition. Nevertheless it is curious
that Subjects seemed to have chosen to move by smaller amounts precisely
in the case when they knew in advance what the sentence structure was.
At present we have no explanation for this aspect of our data.
It is surprising that in Experiment 2, even in
the case when the saccade leading to the test word has an extremely short
latency, the amplitude of the saccade that will take place is still dependent
on whether the upcoming word is a "les" or a verb. We decided to replicate
the experiment with new Subjects, and using eye movement measuring equipment
having more precise spatial resolution.
Materials were identical to those in Experiment
Eye movement were recorded monocularly with an infrared Scanning Laser Ophtalmoscope (SLO). Eye fundus images were recorded with an SVHS video recorder. Tests were projected onto the retina with an LCD display (ACE Euroview 600) piloted by an Apple 9500 computer. Optical pathways for projection of the LCD display and for eye fundus observation were mixed together on the same optical axis with a dichroic mirror.
The Subject's head was stabilised by a dental imprint with the eyes at an optical distance of 40 cm from the LCD screen. The sentences were displayed in the middle of the screen, in upper and lower case, using an 11 x 11 pixel character font, in white on dark, each letter subtending 1/2 degree.
At the beginning of the experiment a calibration was done in which the Subject fixated 8 targets distributed horizontally on the screen over the area where the sentence was to appear. The fundus images recorded during this calibration phase allowed eye positions during the reading phase to be accurately calculated by interpolation. Eye position detection had an accuracy of about 10 min of visual angle (1/3 letter) , and the temporal resolution was 40 ms (corresponding to the video frame rate of 25 frames per second).
Reading was monocular with the dominant eye as ascertained by a test before the experiment. The non-dominant eye was covered.
To trigger the display of each sentence to be
read, the Subject fixated a fixation line on the left of the screen, and,
using two buttons, horizontally moved a second, vernier, line so as to
coincide with the first. The vernier line appeared randomly with an offset
of one or two pixels to the left or right of the fixation line. When the
two lines coincided, the Subject pressed a third button and this caused
the sentence to appear. This mode of triggering the sentence ensured that
the Subject’s eye was directly fixating the initial fixation point when
the sentence appeared. After reading the sentence, the Subject made a response
indicating whether the sentence was semantically correct or not. The response
was given by using the buttons to move a cursor that appeared on the right
of the screen to the left or right, to indicate "correct" or "incorrect".
9 native French speakers with normal or corrected-to-normal
vision, aged 23-54 years, took part in the Experiment.
The data analysis in Experiment 3 was done off-line by an automatic computer program . The program calculated the shift in the video fundus images corresponding to each 40 ms video sample, and then used the previously obtained calibration images to deduce the absolute eye position. Owing to noise in the fundus images, 8% of the records had to be rejected. As in Experiment 2 we also rejected a further 12.9 % of the data because the Subject refixated the word preceding the target word, so that in total 27.2% of the data had to be excluded. An additional analysis was done with the refixation cases included, but this did not modify the results.
The results of Experiment 3 are in every way compatible with those of Experiment 2.
The proportion of incorrect responses to the semantic correctness judgment was 3.9%, but these were nevertheless included in the analysis, as was done for Experiment 2.
Insert Figure 11 here
Figure 11 is a histogram of the latencies of the saccades leading to the test word. These are in the normal range for fixations during reading, as for Experiment 2. In the mixed and blocked conditions the mean durations were 280 ms and 287 ms respectively. An analysis of variance with the four factors test word and two presentation conditions showed no significant effect of test word (F(3,24) = 2.579, ns), or Blocked/Mixed conditions (F(1,8) = 4.497, ns) or interaction of the two (F(3,24) < 1, ns).
Left symbols in Figure 13 gives the mean starting positions of the eye in the five-letter word preceding the test word. Most of the fixations started near the center of the word, as before. There was a difference as a function of the different test word, "les", verb, "xxx" and space (F(3,24) = 3.717; p < 0.05), and no difference as a function of Blocked/Mixed conditions (F(1,8) = 1.141; ns), or interaction of the two (F(3,24) = 1.348; ns). As in Experiment 2, we shall ignore the small differences between the word types since these are negligible in size compared to the effects we are considering in the landing positions (Figure 13).
Figure 12 gives the histograms of landing positions near the different test words, in the mixed and blocked conditions and Figure 13 gives the associated mean landing positions.
Insert Figure 12 and Figure 13 here
The results are virtually identical to those for Experiment 2. In the mixed condition the eye lands 1.5 letters further to the right in the "les" case than in the verb case. In the blocked condition this difference is now 1.2 letters. An analysis of variance with factors verb/les and blocked/mixed confirms that there is a significant global effect of the verb/les comparison (F(1,8)=22.729, p < 0.05). However, contrary to expectations, there is no interaction with the blocked/mixed factor (F(1,8) < 1, ns). There is no effect of the Blocked/Mixed condition (F(1,8) < 1, ns).
The results for the "xxx" and the space word types are also very similar to those for Experiment 2.
Figure 14 plots the landing positions as a function of the duration of the prior fixation. Again the results are virtually identical to those for Experiment 2. There seems to be no strong evidence for the existence of a critical time when the curves begin to separate.
Insert Figure 14 here
The compatibility between the results for Experiments 2 and 3 is very good, and we have replicated the surprising finding that there is no strong evidence for a critical fixation duration after which the curves of landing position in the test word begin to separate.
Surely no reasonable model of eye movement guidance would claim that lexical analysis of parafoveal information would occur instantaneously. It seems that there must therefore be a critical time before which lexical information is available, and before which saccades cannot be affected by the lexical content of the parafoveal stimulus. Could it be that in the case of "les" (and of the "xxx" string) this lexical content is available so early that it is able to affect the target of saccades with latencies as short as 150 ms? And could it be that this information is so easy to extract that it does not matter whether the reader knows in advance what to expect?
Before accepting this conclusion, which seems surprising, we can ask whether some other mechanism might explain the existence of the "THE-skipping" effect at such short prior fixation durations. One possibility is that the actual weight of the visual form of the word "les" in the center of gravity calculation is less than for the verbs used in the experiment, thereby explaining why the eye tends to move further rightwards in the case of the "les". We do not know exactly what center of gravity calculation is performed by the visual system. However, in a task of saccading towards a letter string, has looked at various possibilities, and has found that the data are fairly well fit by a function that weights each pixel by a factor proportional to (1 + 1.7e ), where e is the eccentricity of the pixel, in degrees. To check on the possibility that this might explain the "les"-skipping effect observed here, we calculated the center of gravity of the "les" and the verb configurations. The differences amounted to only a fraction of a letter. It seems that this cannot explain the observed effects.
Another possibility to account for the "les" skipping effect here is that what allows readers to go further rightwards in the case of the "les" string is, not full-blown lexical processing, but something like the familiarity of the particular letters. As suggested by , high bigram frequency of the word in parafoveal vision may increase saccade size. Because of the very high frequency of the word "les" in french, as compared to other words, it is clear that the bigram token counts for the constituent bigrams of "les" will be very high. But note that the familiarity of the string "xxx" and its constituent bigrams will, on the contrary, be very low. And yet for this string skipping behaviour was very similar to skipping for the word "les". This weakens a possible explanation of the "les"-skipping in terms of bigram familiarity, although it is still possible that even though the "xxx" string had rarely-occurring bigrams, some other feature, for example its repetitive structure, nevertheless allowed it to be recognised very quickly.
Yet another possibility to account for the fact that the "les" skipping effect is present at such short latencies might be the fact that information about the test word had already become available before the eye fixated in the first word of the sentence, that is, while the eye was still fixating on the initial fixation point. This would be somewhat surprising, since the test word is situated 9.5 letters to the right of the initial fixation point. Nevertheless, to test for this eventuality, we computed the total time the eye had spent before saccading into the test word and plotted the eye's arrival position in the test word as a function of this. The total time spent before saccading into the test word was taken as the time from the moment the sentence appeared (while the eye was at the intial fixation point) until the moment the eye began its saccade into the test word. This total time usually included two fixations: the final part of the fixation on the initial fixation point, and the fixation in the first word, prior to saccading into the test word.
The data, plotted in Figure 15 show a suggestion that at very short values of the cumulated fixation duration (near 400 ms), the curves for "les" and verb are close together, and become separate for larger cumulated fixation durations. However the effect is not very clear, and furthermore, it is clearer in the mixed than in the blocked conditions, which is contrary to what would be expected, since it should be easier for the visual system to analyze information in peripheral vision in the blocked case.
Yet another possibility to elucidate our failure to find a critical moment before which skipping cannot occur is the following. It might be the case that the effect of prior fixation duration was somehow masked in our experiment by the presence of excessive variability in the data deriving from some other source. In particular, have shown that the landing position in a word is systematically influenced by the position that the eye starts out from: as the eye starts out closer to the word, the mean landing position moves rightwards too. plot a graph plotting landing position as a function of starting position which shows a reliable linear relation, with a slope of about 0.5 (or 0.7 in ) -- implying that for every letter moved closer to the word, the mean landing position moves rightwards by half a letter.
Figure 16 shows graphs similar to McConkie's plotted for the data of our Experiment 3, separately for each of the different conditions (space, "les", verb and "xxx"). The first thing to notice in these graphs is the fact that the data points fall on straight lines, with slopes around 0.8, that is, compatible with McConkie et al.'s work, but slightly steeper. The second point to notice is that the heights of the curves are different, corresponding to the previously noted effects of linguistic difficulty: the eye goes much farther rightwards in the "space" condition, somewhat further rightwards in the "les" condition, and the least far in the verb condition.
Now let us consider the effect of the duration of the fixation duration prior to the saccade on these curves. Figure 17 shows the same curves as before but now divided into four separate sets, corresponding to the case when prior fixation duration lay was in the ranges [160,200], ]200,240], ]240,280] and ]280,360] ms (the results for the xxx condition have been left out for clarity; Mixed and blocked groups have been cumulated so that enough data would be available). These categories contained respectively 15.5, 24.8, 21.2 and 22.9 percent of the data. We see that in the category of short fixation durations, the curves for "les" and verb are not distinctly separate, whereas they are for all the longer duration categories.
Here we have therefore perhaps found evidence coherent with the hypothesis of a critical duration before which linguistic processing cannot influence saccade behavior. Of course, because the data are noisy owing to the fact that they have been subdivided into three temporal categories, it is difficult to be confident of this conclusion.
Insert Figure 15 here
The results of our study are coherent, but somewhat surprising. Our interpretation of the results of Experiment 1 is that there was a ceiling effect: fixation duration prior to jumping towards the critical "les" or verb string were rather longer (500 ms) than in normal reading, so even in the mixed condition, where Subjects did not know in advance what the structure of the upcoming sentence was, there will have been enough time to use the information available in parafoveal vision to modify the saccade as a function of the lexical status of the parafoveal word. But the procedure used in Experiment 2 guaranteed more normal, 250 ms, fixation durations prior to jumping into the critical word. Here we expected that the blocked/mixed manipulation would affect the strength of the "THE-skipping" phenomenon. We also expected that analysis of skipping as a function of the duration of the prior fixations should reveal a critical duration before which no skipping occurred. However neither of these expectations were confirmed. There was a slightly stronger skipping effect in the blocked condition than in the mixed condition, but the interaction was not significant. There was certainly no strong suggestion of a change in the strength of the skipping as a function of prior fixation duration.
A third experiment replicating Experiment 2 confirmed all these findings using more reliable equipment. The existence of the THE-skipping effect is therefore definitely confirmed, but a surprising fact remains to be explained: counter to any reasonable model of eye movements in reading, the effect seems to be much more resistant to variation of saccadic latency than would be expected. It seems to be present even for cases like the mixed condition, and for very short saccade latencies, for which lexical processing could surely not yet have proceeded far enough to allow differentiation of the article and the verb. Even an analysis where all fixation durations prior to saccading into the test word are cumulated yields no clear indication of a critical moment. There is some indication from an analysis taking into account the eye's starting position in the word prior to the test word, that for fixation durations less than or equal to about 200 ms, the "les"-skipping effect disappears, but, because of lack of data, it is hard to be confident in the effect.
In conclusion, we have been able, in some sense, to confirm globally the "THE-skipping" effect in a French version of the original experiment. But it's apparent insensitivity to the duration of the prior fixation and to the presence of prior knowledge about sentence structure, suggests the need to do further work to determine if the effect is due, not to normal lexical processing, but to some lower-level mechanism.
Balota, D.A., Pollatsek, A. & Rayner, K. (1985). The interaction of contextual constraints and parafoveal visual information in reading. Cognit Psychol, 17(3): 364-390.
Beauvillain, C., Dore, K. & Baudouin, V. (1996). The 'center of gravity' of words: evidence for an effect of the word-initial letters. Vision Research, 36(4): 589-603.
Brysbaert, M., Vitu, F. & Schroyens, W. (1996). The right visual field advantage and the optimal viewing position effect: on the relation between foveal and parafovel recognition. Neuropsycology, (10): 385-395.
Coeffe, C. & O'Regan, J.K. (1987). Reducing the influence of non-target stimuli on saccade accuracy: predictability and latency effects. Vision Res, 27(2): 227-240.
Doré, K. & Beauvillain, C. (1997). Latency dependence of word-initial letter integration by the saccadic system. Perception & Psychophysics, 59(4): 523-533.
Ehrlich, K. & Rayner, K. (1981). Concextual effects on word perception and eye movements during reading. J. Verbal Learn. Verbal Behav., 22: 75-87.
Findlay, J.M. (1982). Global visual processing for saccadic eye movements. Vision Res, 22(8): 1033-1045.
Henderson, J.M. & Ferreira, F. (1990). Effects of foveal processing difficulty on the perceptual span in reading: implications for attention and eye movement control. J Exp Psychol Learn Mem Cogn, 16(3): 417-429.
Hyönä, J., Niemi, P. & Underwood, G. (1989). Reading long words embedded in sentences: informativeness of word halves affects eye movements. J Exp Psychol Hum Percept Perform, 15(1): 142-152.
Inhoff, A.W. (1989). Parafoveal processing of words and saccade computation during eye fixations in reading. J Exp Psychol Hum Percept Perform, 15(3): 544-555.
McConkie, G.W., Kerr, P.W. & Dyre, B.P. (1994). What are normal eye movement during reading: toward a mathematical description. In: Ygge, J. & Lennerstrand, G. (Eds.). Eye movements in reading. (pp. 315-327). Oxford: Elsevier.
McConkie, G.W., Kerr, P.W., Reddix, M.D. & Zola, D. (1988). Eye movement control during reading: I. The location of initial eye fixations on words. Vision Res, 28(10): 1107-1118.
Morrison, R.E. (1984). Manipulation of stimulus onset delay in reading: evidence for parallel programming of saccades. J Exp Psychol Hum Percept Perform, 10(5): 667-682.
O'Regan, J.K. (1980). The control of saccade size and fixation duration in reading: the limits of linguistic control. Perception & Psychophysics, 28(2): 112-117.
O'Regan, J.K. (1990). Eye movements and reading. Rev Oculomot Res, 4: 395-453.
O'Regan, K. (1979). Saccade size control in reading: evidence for the linguistic control hypothesis. Percept Psychophys, 25(6): 501-509.
Pollatsek, A., Rayner, K. & Balota, D.A. (1986). Inferences about eye movement control from the perceptual span in reading. Percept Psychophys, 40(2): 123-130.
Rayner, K. & Morris, R. (1992). Eye movement control in reading: evidence against semantic preprocessing. J Exp Psychol Hum Percept Perform, 18(1): 163-172.
Schustack, M.W., Ehrlich, S.F. & Rayner, K. (1987). Local and global sources of contextual facilitation in reading. J. Mem., 26: 322-240.
Toro, R. & Gautier, V. (1998). Frame tracking software and stabilization of eye fundus images. ARVO, Fort Lauderdale, Florida.
Underwood, G., Clews, S. & Everatt, J. (1990). How do readers know where to look next? Local information distributions influence eye fixations. Q J Exp Psychol [A], 42(1): 39-65.
Underwood, G., R., B. & Clews, S. (1988). Information influences the pattern of eye fixations during sentence comprehension. Perception, 17(2): 267-278.
Vitu, F. (1991). The existence of a center of gravity effect during reading. Vision Research, 31(7-8): 1289-1313.
Vitu, F. (1991). The influence of parafoveal preprocessing and linguistic context on the optimal landing position effect. Perception & Psychophysics, 50(1): 58-75.
Zola, D. (1984). Redundancy and word perception during reading. Perception & Psychophysics, 36: 277-284.
"THE- skipping" effect : the mean positions where the eye landed when leaving a starting region and going towards a test word which could either be the article "THE", or a three-letter verb like "HAD" or "WAS" or "ATE".
Predictions for the eye's landing position when going towards a test word (shown as white rectangle on the ordinate) as function of the duration of the fixations prior to the saccades leading to the test word.
Experiment 1: Histograms of latency when the eye
leaves the five-letter word and after a single fixation.
Experiment 1: Histograms of the eye's landing position when leaving the five-letter region on the left of the test word, and moving towards the test word. The white box on the abscissa indicates the particular test word concerned.
Experiment 1: Means and one standard error on
each side of the mean of the landing position of the eye when leaving the
five-letter region and going towards the test word (white rectangle on
Experiment 1: Means and one standard error on each side of the mean of the landing position of the eye when leaving the five-letter word and going towards the test word (white rectangle on ordinate), as a function of the latency of the fixation in the five-letter word prior to the test word.
Experiment 2 : Histograms of the latency of the
saccade leaving the five letter word and going towards the test word (white
rectangle on abscissa) after only one fixation inside this five-letter
Experiment 2: Symbols on the left: Means and one standard error on each side of the mean, of the eye's starting position inside the five-letter word, before going towards the test word (white rectangle on abscissa). Symbols on the right: landing positions near the test word after leaving the starting positions, and having made only one fixation in the five-letter word.
Experiment 2: Histograms of the eye's landing
position when leaving the five-letter word, and moving towards the test
word ) after only one fixation inside this five-letter word. The white
box on the abscissa indicates the position of the test word.
Experiment 2 : Upper pairs of curves: means and one standard error on each side of the mean of the landing position of the eye when leaving the five-letter word and going towards the test word (white rectangle on ordinate) ) after only one fixation inside this five-letter word, as a function of the latency of the fixation in the word prior to the test word. The lower pairs of curves show the means of the corresponding starting positions in the five-letter word preceding the test word.
Experiment 3: Histograms of the latency of the saccade leaving the five letter word and going towards the test word after only one fixation inside this five-letter region.
Experiment 3 : Histograms of the eye's landing
position when leaving the five-letter region, and moving towards the test
word after only one fixation inside this five-letter region. The white
box on the abscissa indicates the particular test word concerned.
Experiment 3 : Symbols on the left: Means and one standard error on each side of the mean, of the eye's starting position inside the five-letter word, before going towards the test word (white rectangle on abscissa). Symbols on the right: landing positions near the test word after leaving the starting positions, and having made only one fixation in the five-letter word.
Experiment 3 : Means and one standard error on each side of the mean of the landing position of the eye when leaving the five-letter word and going towards the test word (white rectangle on ordinate) after only one fixation inside this five-letter word, as a function of the latency of the fixation in the word prior to the test word.
Experiment 3: Means and one standard error on
each side of the mean of the landing position of the eye near the test
word (upper curves) as a function of the cumulated time between the presentation
of the sentence and the arrival of the saccade near the test word. Also
shown in these graphs (lower pair of curves in each diagram) are the corresponding
positions where the eye started from in the departure zone (i.e. in the
Experiment 3: Landing site in the test word (white rectangle on the ordinate) as a function of departure position in the five-letter word preceding the test word (abscissa). Data for each departure position have been pooled over mixed and blocked conditions. The regression lines and r2 values are:
Verb : Y = 7.29 + 0.76 * X; r2 = 0.882
Les : Y = 8.49+ 0.81 * X; r2 = 0.925
Space : Y = 10.89+ 0.748 * X; r2 = 0.936
xxx= Y = 8.26+ 0.80 * X; r2 = 0.692
Experiment 3: Landing site in the test word (white rectangle on the ordinate) as a function of departure position in the five-letter word preceding the test word (abscissa). Data for each departure position have been pooled over mixed and blocked conditions. Different diagrams correspond to subsets of the data in which the latency of the fixation in the five-letter word preceding the test word was in one of four ranges.
Y : Landing site
X : Launch site
Verb : Y = 7.49 + 0.90 * X; r2 = 0.746
Les : Y = 8.25+ 1.62 * X; r2 = 0.83
Space : Y = 10.878+ 0.55 * X; r2 =
Experiment 3: Landing site
Latency ]200-240] ms
Y : Landing site
X : Launch site
Verb : Y = 7.85 + 0.57 * X; r2 = 0.442
Les : Y = 8.86+ 0.62 * X; r2 = 0.82
Space : Y = 10.733+ 0.95 * X; r2 =
Experiment 3: Landing site
Latency ]240-280] ms
Y : Landing site
X : Launch site
Verb : Y = 7.73 + 0.55 * X; r2 = 0.547
Les : Y = 8.82+ 0.95 * X; r2 = 0.78
Space : Y = 10.86+ 0.94 * X; r2 = 0.90
Experiment 3: Landing site
Latency ]280-360] ms
Y : Landing site
X : Launch site
Verb : Y = 7.26 + 1.00 * X; r2 = 0.847
Les : Y = 8.31+ 0.57 * X; r2 = 0.819
Space : Y = 10.79+ 0.53 * X; r2 = 0.9