eye fixations are influenced by the distribution of information within words

Acta Psychologica 72 (1989) 263-280 North-Holland

263

EYE FIXATIONS ARE INFLUENCED BY THE DISTRIBUTION OF INFORMATION WITHIN WORDS *

Geoffrey UNDERWOOD, Susan CLEWS and Howard WILKINSON University of Nottingham, UK

Accepted February 1989

Reader’s eye movements were monitored while they inspected isolated words in preparation for a synonym judgement task. The lo-letter words appeared on a screen near the point of fixation, with the first fixation being imposed near the beginning, or the centre, or the ending of the word. The words themselves had uneven distributions of information, in that the beginnings or the endings contained common sequences of letters in English. Three types of words were used: those with very redundant endings (e.g., yearningly), with moderately redundant endings (e.g., uarnish-

ing), and with moderately redundant beginnings (e.g., contravene). Redundancy was defined in terms of the total number of words in English which possess that particular sequence of five letters as the beginning or the ending. The experiments asked whether the convenient viewing location within a word varied according to the distribution of information, and whether the extent of redundancy in a word ending is reflected in the distribution of visual attention given to the word. The results were analysed separately for those cases where the reader made just two fixations upon the word before moving to the synonym task, and for those cases where the reader made exactly three fixations. These were the dominant fixation patterns. Evidence for the notion of a convenient viewing position consisted of long first fixations (when there were just two fixations), when this fixation was near the centre of the word. The distribution of information within the words did not influence the duration of the first fixation, although the duration of the gaze within each half-word did increase when more informative letter sequences were being inspected. The extent of redundancy was also seen to influence the inspection patterns, when a comparison was made between the two types of words with redundant endings. Words with very redundant endings received fewer fixations when the first fixation was at the beginning, and for words gaining exactly two fixations, the second fixation was shorter if the word had a very redundant ending.

* Jukka Hyonl helped with interpretation of the data during a visit to Nottingham sponsored by the Academy of Finland. The research was supported by Medical Research Council project grant G8602578N. Kevin G’Regan and a second reviewer provided valuable comments upon the first draft of this paper.

Requests for reprints should be sent to G. Underwood, Dept. of Psychology, University of Nottingham, Nottingham NG7 2RD, UK.

OoOl-6918/89/$3.50 0 1989, Elsevier Science Publishers B.V. (North-Holland)

264 G. Underwood et al. / Eye fixations and worak

Does the reader’s knowledge of the orthographic redundancy within words influence the pattern of eye fixations which is used to extract lexical information? Redundancy can be defined as that which increases as the probability of occurrence increases. For example, im- mediately after the letter Q, in common English words, the letter U is predictable and therefore redundant. The same can be said of whole sequences of letters in long words. For example, the word quarantine is the only common word with ten letters which begins with the letters quura-. We can therefore say that the final five letters are redundant in the context of this word, because it is unnecessary to gain visual information from the page in order to identify the word, provided that the reader knows that it is a ten-letter word beginning with quara-. On the same basis, there are words which have unique beginnings in the context of a given ending. For example, given the ending -plier in a ten letter word, there is only one common completion: the word multiplier is the only word with this ending although there are many words which start with multi-.

The most efficient reading strategy should be one which maximizes the use of the reader’s orthographic and lexical knowledge and which uses this knowledge to minimize the number of fixations in the extraction of redundant visual information. This strategy should therefore not provide fixations upon these redundant beginnings and endings, and if they do occur, then the duration of the fixation should be short. Extraction of the visual information from the non-redundant part of the word should be sufficient for word recognition to be completed. The present experiment investigated the inspection patterns upon long words with uneven distributions of information. A number of experiments have observed that readers tend to first

fixate words towards the left of the word’s centre (Dunn-Rankin 1978; Rayner 1979; Kliegl et al. 1983; Underwood et al. 1987: exp. 2; 1988). This is the typical inspection pattern when reading words presented in their natural sentence contexts, and the location of this first fixation may be described as the preferred landing position.

The preference for a fixation slightly to the left of centre also has a correlate in the inspection of isolated words which appear on a screen at the point of fixation. Optimal processing is observed for words presented with the point of fixation slightly to the left of centre, but the two positions do not coincide exactly. Optimal processing can be assessed by measuring the total amount of visual attention given to the

G. Underwood et al. / Eye fixations and words 265

word before moving away from it. When the imposed fixation is slightly to the left of centre the total amount of visual attention is less than when the imposed fixation is at the beginning or the end. The preferred landing position is nearer to the beginning of the word, and optimal processing is obtained when the first fixation is nearer to the centre of the word.

O’Regan and Levy-Schoen (1987) and Holmes and O’Regan (1987) have described the optimal position in a word from which the inspection should start, and they describe this as the convenient viewing position. Their data showed that when the first fixation upon a word is slightly to the left of centre then fewer fixations are required for completion of processing, and that the overall gaze time upon the word is lower, in comparison with trials when the first fixation is upon the first letter or towards the end of the word (see also O’Regan et al. (1984) for a preliminary report of this effect). For short words and for high frequency words the optimal viewing position was in the centre of the words, and for long words and low frequency words it was slightly left of centre. Optimal processing is consistent with a first fixation located three or four letters from the beginning, thereby providing fovea1 processing for the first half of the French words used in these experiments, and given the left-to-right eye movements which are observed even with isolated words. A fixation near to the beginning would not capture as many letters of the word, and a fixation towards the end would require a leftwards saccade so that the first few letters could be inspected. A similar pattern of results was reported in an investigation of the inspection of Finnish words (Underwood et al. 1987: exp. 1).

A reanalysis of the O’Regan et al. (1984) data, by O’Regan and Levy-Schoen (1987), suggested that the main effect of the convenient viewing position was upon the inspection strategy rather than upon fixation durations. If the first, imposed, fixation was slightly to the left of the word’s centre, then there was an increased probability of the reader making just one long fixation. When the first fixation was elsewhere, multiple fixations were generally made. The benefit gained by the first fixation being upon the convenient viewing position was that most of the time no other fixations needed to be made.

The distribution of information within long words was seen to influence the inspection patterns in the French and Finnish experiments. When the beginnings of the words were relatively redundant,

266 G. Underwood et al. / Eye fixations and words

the convenient viewing position was nearer to the end of the word, or was not apparent at all.

The present experiment asked three related questions about the convenient viewing position and its influence upon the inspection of words with uneven distributions of information: can the convenient viewing position be confirmed with long English words when the initial point of fixation is varied; does the convenient viewing position vary as the location of information within the word varies; and does the extent of the redundancy change the pattern of fixations? The first two questions represent attempts to replicate, for the case of English words, previous studies of the convenient viewing position (O’Regan and Levy-Schoen 1987; Underwood et al. 1987: exp. 1). For this purpose words with informative endings were compared with those with informative beginnings. A third type of word was also used to provide a comparison between two types of words, each of which have informative beginnings and redundant endings. One of these types of information-beginning word has a very common sequence of letters at the end (more than 200 words of that letter length share those terminal five letters), and the other type of information-beginning word has a less redundant ending. The second type of word also had common sequences of terminal letters (shared by more than 50 other words), but they are less common than those in the first type. The experiment tested the hypothesis that readers would be sensitive to this four-fold increase in redundancy by giving the more redundant letters fewer and shorter fixations.

Single words were presented to the point on a screen where the subjects were fixating. After inspecting this target word, the subjects looked at two further words to decide which was the best synonym for the target word. The experiment asked whether the redundancy of the beginning or ending of the target word would influence the pattern of fixations during its initial inspection.

Method

Subjects

Twelve subjects from the University of Nottingham completed the experiment. They were paid E2.00 for participating. All subjects had normal uncorrected eyesight, and had English as their first language.

G. Underwood et al. / Eye fixations and words 261

Materials and apparatus

Three groups of 27 words were selected. The words were all 10 letters long and were chosen so that the informative part of the word was restricted to the first few or the final few letters. The selection procedure was as follows: a dictionary search (Schwartz 1986; Simpson 1985) revealed a group of words which had unusual letter patterns in the first five or in the last five letters. For Example quarantine is the only lo-letter word beginning with quara- and underneath is the only lo-letter word ending in -neath. These areas are zones of high information since the identity of the word could be determined from these few letters alone. When the assessment was repeated for nine-letter and eleven-letter words, the classification remained unchanged.

The first group of words used in the experiment had informative beginnings and very redundant endings, with the word endings having an average of 210.7 beginnings. There were 27 words in this sample, and they can be described as the IR words. Words in this group included accumulate, relocation, and yearningly. The endings are very redundant because if a reader familiar with English orthography is presented with only the ending of one of these words (for example, -‘ingZy’ and ‘-ation’ and ‘date’) then the identity of the whole word is not available. When given the beginnings of these words (with these examples, ‘yearn-’ and ‘reloc-’ and ‘accum-‘) the words can be identified without knowledge of the endings. The average frequency of these IR words was 2.88, according to the Kucera and Francis (1967) corpus. There occur naturally in English many words with unusual beginnings and very common endings but the reverse does not apply, and so it is not possible to compare these words with a group with highly redundant beginnings.

The second set of 27 words contained the information in their endings. The average number of endings for each beginning was 51.9, and so these words can be described as having informative endings and moderately redundant beginnings (r1 words). Items in this group of r1 words included contravene, orthopedic and supervisor. These words started with relatively common sequences of letters, and ended uniquely. The average frequency of the r1 words was 2.74 (Kucera and Francis 1967).

The r1 words were matched with a third set of 27 words which had informative beginnings and moderately redundant endings (the Ir words). Words in the Ir group included lavishness, monetarist and varnishing. The Ir words had an average number of 51.6 beginnings for each five-letter ending, and can be considered to be less redundant versions of the IR words. The average frequency of the Ir words was 2.63 (Kucera and Francis 1967).

For each of the 3 X 27 target words a pair of response alternatives was chosen. One of the pair had a similar meaning to the target and the other was an unrelated distractor. For example, the alternatives for the target stimulus unquotable were unrepeatable and unsolveable. When the target word was removed from display the two response alternatives appeared along a line below and equidistant from the target.

The words were displayed on a high resolution monochrome monitor (Taxan KX-12) controlled by a BBC Master microcomputer, which was also interfaced with the eye-movement recorder. The eye recorder sampled the position of the left eye every 10 msec (Bray et al. 1983: 429), detecting the horizontal and vertical position of the eye


by monitoring the corneal reflex from an infra-red point source on a 64 x 64 photoelectric matrix. This procedure gives an accuracy of plus or minus one character space. At the viewing distance of 48 cm used in the experiment, a ten-letter word subtended a horizontal angle of 3 O, and the maximum letter height was 0.6 O. The recorder has been described in detail by Wilkinson (1979). The infra-red source and photoelectric matrix were mounted on a dental plate which the subject held between the teeth. The dental plate was coated in a thermoplastic dental composition which moulded and then hardened to the shape of the subject’s teeth. Displays were terminated by a button press which was hand-held by the subject.

Design

There were two within-subjects factors: word type and point of initial fixation. The three types of words presented those with information at the beginning and highly redundant endings (IR words), those with information at the beginning and moderately redundant endings (Ir words), and finally those with information at the end and moderately redundant beginnings (r1 words). There were 27 words of each of the three types.

Words were presented in isolation to the region of fixation but in one of three positions on the monitor. They were shown centrally to fixation on a third of trials but on the remaining trials the words were displaced slightly so that the initial fixation fell upon either the 3rd or the 8th letter rather than centrally between the 5th and 6th characters. Subjects saw all of the words with counterbalancing over whether the initial fixation on a word would be centrally or onto the 3rd or onto the 8th character.

Procedure

After the dental plate had been coated and an impression of the subject’s teeth made, the eye movement recorder was attached to the bit bar which was held in place between the subject’s teeth. A chin rest and frame restricted head movements. The apparatus was then calibrated first by roughly adjusting the position of the recorder in order to obtain a strong reading and then more finely to check the horizontal and vertical alignment of responses.

In all there were 81 experimental trials and 9 practice trials. A trial began with a fixation mark appearing in the middle of the screen. The fixation mark disappeared and was replaced by a target word only when the recorder detected at the subject had fixated the mark for 100 msec. As soon as the subject read the word he pressed a button and the target word was replaced by the two choice alternatives. The subject read the two alternatives and then pressed a button as he looked at the response alternative which meant the same as the target word. This second button press removed the alternatives from the screen and initiated presentation of the fixation mark signalling the next trial appeared.

G. Undenoood et al. / Eye fixations and words 269

Table 1 Distribution of fixations expressed as a percentage of all fixations upon each type of word.

IR type words Ir type words r1 type words

1 fixation exactly 11.14% 8.35% 6.65% 2 fixations exactly 28.30% 31.82% 37.11% 3 fixations exactly 31.07% 33.03% 26.15% More than 3 fixations 29.49% 25.06% 30.08%

Note: IR words have informative beginnings and highly redundant endings; Ir words have informative beginnings and moderately redundant endings; rI words have moderately redundant beginnings and informative endings.

Results

Fixations made to response alternatives were examined to ensure that the correct response had been made. Subjects had no difficulty in selecting the correct word as a synonym for the target: no data were excluded on these grounds. Fixation durations

and locations on target words were recorded and subjected to a number of analyses. Data from individual target words were accepted only if the observed position of the

Table 2 Gaze durations and total inspection times (msec).

IR type words Ir type words rI type words

Half word: Beginning End Beginning End Beginning End

Gaze durations Point of initial fixation

Beginning 608 403 546 384 551 450 Centre 586 338 559 369 471 486 End 540 365 472 455 368 369

Total inspection times Point of initial fixation

Beginning 682 468 735 446 786 582 Centre 755 532 746 591 600 600 End 631 703 597 764 462 672

Note: Gazes are defined here as the total aecummulated fixations upon a half word, and are terminated by a movement which results in fixation in the other half of the word or away from the word.

Total inspections are defined as the total accumulated fixations upon the whole word, regardless of the number of location of fixations, and are terminated by a movement which results in a fixation away from the word.


first fixation was within two character spaces of the intended position. The two main factors involved in all these analyses were position of initial fixation and word type.

The dependent measures all concerned the inspection of the target words. We first calculated gaze durations (defined here as the duration of visual attention to a half-word prior to crossing the mid-point boundary a second time). Secondly, we calculated the total inspection durations on each word prior to the saccade resulting in a fixation away from the target, and thirdly we calculated the total number of fixations upon each word prior to leaving the word boundary. At a finer level, we also calculated the duration of each individual fixation upon the target, for those cases where the target word was fixated just once, where the target was fixated exactly twice, and where it was fixated exactly three times. The distribution of the number of fixations gained by each word, for each initial point of fixation, is presented in table 1.

Gaze durations

A gaze is defined here as the uninterrupted visual attention given to a half-word without the eyes leaving that half of the word. This is the average time spent inspecting the beginning or the end of each word before either fixating outside the word or making a saccade across the midline boundary of a word for the second time. This is not the definition of ‘gaze’ used universally, but we were particularly interested in the accumulated visual attention given to half words because the experiment specifically varied the information distribution between the two halves of each word. Gaze time here is taken to be the total time spent fixating an area of a word without fixating outside that area. For example, if a reader had the beginning of the word presented to the first fixation, and secondly fixated the end of the word, before finally fixating the beginning of the word again, we would count three fixations upon the target. The first two of these fixations would be considered as contributing to the gaze data, but the third fixation, which is a refixation of the beginning after leaving that half word, would not be added to the cumulative total gaze duration. (This third fixation would, however, be a contribution to the ‘total inspection duration’ which is discussed below.) If the reader did not fixate a half word, then the gaze duration would be zero for that half word. A gaze duration of zero was entered if the reader made only one fixation before leaving the word, for example, or made two or more fixations upon one half of the word without inspecting the other half at all.

The gaze means are presented in table 2. An analysis of variance was performed on these data, with repeated measures over all three factors (word type, point of initial fixation, and gazes upon the beginning versus the end of the word). Only reliable effects will be described, here and elsewhere.

The main effect of gaze durations at the beginnings versus the ends of words was reliable (F(1, 11) = 5.6, p < 0.05), with the durations at the beginnings of words being longer. This difference interacted with word type (F(2, 22) = 5.8, p < 0.01). An analysis of simple main effects showed that gaze times were significantly longer at the beginnings than at the ends of words only for the two types of information beginning words (IR and Ir both p < 0.05).

G. Underwood et al. / Eyejxations and words 271

Total inspection durations

The total amount of time spent inspecting each half of each target word was calculated, and these means are also presented in table 2. Fixations ceased to be added to these inspection durations when a saccade was made which took the reader’s eyes away from the target word. Multiple crossings of the word’s midline boundary were ignored by this measure, which is an indication of the total visual attention given to each half of the word.

Inspection durations were submitted to an analysis of variance with the three factors of word type, point of initial fixation, and word half. None of the main effects approached reliability, and the one interaction which was reliable involved the point of initial fixation and word half (F(2, 22) = 7.40, p < 0.005). The means in table 2 suggest that this is a clear cross-over interaction, and this was confirmed with an analysis of simple main effects. For initial fixations at the beginning there was more time spent inspecting the first half of the word rather than the second half (p < 0.005), and for initial fixations at the end there was more time spent inspecting the second half of the word rather than the first half (p < 0.05). This appears to be a consequence of ensuring at least one fixation in a given location, but it is not clear why it did not emerge in the analysis of the gaze durations. When the initial fixation was upon the centre of the word, there was no difference in the total amount of visual attention given the word’s beginning or ending.

Total number of fixations

The average number of fixations made to each half of the target words was calculated and these data are presented in table 3. An analysis of variance was performed with repeated measures on all three factors. These factors were word type (IR, Ir and rI), point of initial fixation (at the beginning, centre, or end of the word) and half word (beginning vs. end of the word).

None of the main effects were reliable. The interaction between word type and point of initial fixation was reliable (F(4, 44) = 2.89, p < 0.05). An analysis of simple main effects showed that the point of initial fixation influenced the total number of fixations required only for type IR words (p c 0.05). The number of fixations required was least when initial fixation was upon the information zone at the beginning of the word.

Table 3 Average number of fixations made upon each type of word.

Half word:


Beginning End Beginning End Beginning End - - Point of initial fixation

Beginning 2.24 1.31 2.57 1.54 2.63 2.03 Centre 2.23 1.91 2.21 2.06 1.86 1.97 End 1.97 2.65 1.77 2.48 1.75 2.37


Table 4 Average fixation durations (msec) when only one fixation was made upon the word.


Point of initial fixation Beginning 835 1127 636 Centre 717 785 550 End 363 366 230

The interaction between point of initial fixation and the number of fixations made upon the two halves of the words was highly reliable (F(2, 22) = 17.28, p < 0.001). This is to be expected since the manipulation of point of fixation determined the location of at least one fixation. Inspection of table 3 indicates a larger number of fixations at the ends of words when the initial fixation is at the end.

Targets gaining only one fixation

On 8.7% of trials the subject made a saccade away from the target word directly from the imposed first fixation. The means for these fixation durations are presented in table 4, but these should be interpreted with great caution as some of the cells contain means from very small numbers of trials (cf. table 1). There were too many empty cells to undertake any formal analysis. The pattern for the three types of words is very similar, however. In each case, the nearer the fixation was to the word’s beginning, the longer was the duration of that fixation.

Targets gaining exactly two fixations

The most common fixation pattern was for the subject to make two fixations upon the target word before leaving the word boundary (32.6% of trials). The fixation durations of these first two fixations, for those occasions when the subject made exactly two fixations, were submitted to a separate analysis of variance with three factors - point of initial fixation, word type, and first/second fixation. The means of these fixation durations are presented in table 5.

Point of initial fixation emerged as a main effect (F(2, 22) = 5.30, p < 0.05). Simple paired comparisons showed that initial fixations at the word’s beginning resulted in shorter average fixations than when the initial fixation was either in the word’s centre or the words ending (both p c 0.05). Word type also appeared as a main effect (F(2, 22) = 4.25, p < 0.05). Simple paired comparisons showed that fixations were shorter upon IR type words than upon Ir type words (p < 0.05).

The analysis of variance revealed two interactions. Point of initial fixation interacted with first/second fixation (F(2, 22) = 21.74, p < O.OOl), and analyses of simple main effects indicated shorter first fixations, in comparison with second fixations, with an initial fixation at the word’s beginning and at the word’s ending (both p c 0.05). There were longer first fixations when this was upon the word’s centre (p < 0.01).

G. Underwood et al. / Eye fixations and worak 273

Table 5 Average fixation durations (msec) when two fixations were made upon the word.

IR type words

Fixation: 1st 2nd

Point of initial fixation Beginning 231 370 Centre 541 288 End 435 387

Ir type words

1st 2nd

280 579 591 329 388 577

rI type words

1st 2nd

251 404 502 387 331 519

Simple comparisons indicated shortest first fixations for initial fixations at the beginning, longer fixations at the ending, and longest fixations at the word’s centre (all p < 0.05). The second fixation was shortest when the initial fixation was at the centre, in comparison with initial fixations at the beginning and at the ending (both p < 0.05).

The second interaction was between word type and first/second fixation (F(2, 22) = 3.59, p < 0.05). An analysis of simple main effects indicated an effect of word type upon the duration of the second fixation (p < 0.01) but not upon the duration of the first fixation: IR words gained shorter fixations than Ir words (p < 0.05) only. In each of these cases the second fixation was, on average, located 0.84 characters to the left of an IR word’s centre and located 0.56 characters to the left of an Ir word’s centre. In both cases, then, the second fixation tended to be upon the more informative part of the word. The second fixations upon r1 words also tended to be to the left of centre, by 0.78 characters, but in this case the fixation is upon the less informative part of the word.

Targets gaining exactly three fixations

On 30.2% of trials the target word was fixated exactly three times before the readers’ eyes left the word boundary. This was the second most common fixation pattern. The mean fixation durations for these three fixations are contained in table 6. These fixation data were submitted to an analysis of variance with point of initial fixation, word type, and the sequence of fixation as factors.

Table 6 Average fixation durations (msec) when three fixations were made upon the word.

Fixation:

IR trpe words Ir type words r1 type words

1st 2nd 3rd 1st 2nd 3rd 1st 2nd 3rd

Point of initial fixation Be&ming 194 259 408 236 231 467 190 242 293 Centre 284 293 303 218 455 270 283 265 395 End 289 307 402 305 304 345 246 266 465


Only the sequence of fixation emerged as a main effect (F(2, 22) = 10.57, p c O.OOl), and simple paired comparisons indicated that the third fixation lasted longer than the first two fixations (p < O.Ol), which in turn did not differ from each other.

The sequence of fixation also interacted with the point of initial fixation (F(4, 44) = 2.81, p < 0.05). An analysis of simple main effects revealed that the trend of the third fixation being the longest held good when the initial fixation was either at the beginning or at the end of the word (both p < 0.01) but that there was no difference in the lengths of the three fixations when the first fixation was upon the word’s centre.

A three way interaction involving word type, point of initial fixation, and sequence of fixations was also evident (F(8, 88) = 2.30, p < 0.05). Paired comparisons indicated a number of contrasts, and they all involved extended second or third fixations. The durations of the first fixations did not vary as a function of word type or point of initial fixation. These comparisons can be described as variations in the patterns found as a main effect of the third fixation being longer in duration than the first two fixations.

For IR words, the third fixation was the longest only when the first fixation was at the beginning of the word (p < 0.05). There were no differences in durations when the first fixation was at the centre or at the end. For Ir words, the third fixation was longest when the first fixation was at the beginning (p < 0.05), and the second fixation was longest when the first was upon the word’s centre (p < 0.05). When the first fixation was at the end, then the three fixation durations did not differ. For r1 words, the pattern of a long third fixation was only observed when the first fixation was upon the word’s ending (p < 0.05). The other two starting positions produced relatively similar fixation durations. The second fixations tended to be located to the left of the word’s centre: IR words were 0.87 characters to the left of centre, Ir words were 0.81 characters to the left, and r1 words were 0.52 characters to the left.

For the third fixation, an initial imposed fixation upon the word’s beginning resulted in longer fixations for Ir words in comparison with r1 words (p < 0.05). Both of these third fixations tended to be located to the right of the word’s centre. Ir words gained 467 msec fixations located 0.65 characters to the right; and r1 words gained 293 msec fixations located 0.51 characters to the right. These third fixations therefore tended to be towards the end of the word, and a short fixation was observed when it was upon the informative part of the word. The opposite effect is indicated for initial fixations upon the word’s ending, but in this case the effect is unreliable. Ir words gained 345 msec fixations 1.33 characters to the left of centre (in the informative half-word), whereas r1 words gained 465 msec fixation 0.74 characters to the left (in the less informative half-word). The tendency is similar in the case of both of these comparisons: longer fixations were made when the fixation was upon the less informative part of the word.

The experiment was designed to investigate three questions con- cerning the inspection of isolated words which had uneven distributions of information.

G. Underwood ef al. / Eye fixations and words 215

The first question asked whether the convenient viewing position can be indicated by the initial point of regard. Evidence was found to support this suggestion from those words receiving exactly two fixations. The first fixations were of longer duration when the word was presented so that readers first saw the centre of the word. This supports O’Regan and Levy-Schoen’s (1987) suggestion that the optimal recognition of a word proceeds when the fixation is near to or slightly to the left of the centre. A long fixation is indicative of optimal processing in their analysis, whereas a short fixation indicates a decision to move to a more informative location.

The second question addressed by the experiment was whether the effect of the convenient viewing position is modified by the distribution of information within the word. The words were selected so that they could be more easily predicted by knowledge of either the first five letters or the last five letters, and it was predicted that the slightly-left- of-centre location of the convenient viewing position would be most apparent for words with information to the left of centre (IR and Ir words), and that its location would be towards the ends of words if the information was located towards the ends (r1 words). The amount of visual attention given to a word did vary with the distribution of information: words with information to the left of centre received longer gazes at their beginnings. This does not provide support for the notion of a convenient viewing position, however, for O’Regan and Levy-Schoen (1987) report that the total amount of visual attention decreases when the informative part of the word is inspected first. When we take the measure of total inspection duration into account, a measure close to the one used by O’Regan and Levy-Schoen, then there is no sensitivity to the distribution of information. These analyses therefore fail to provide unequivocal support to the notion of a convenient viewing position which is associated with optimal word recognition.

Differences between words with their information at the beginnings and at their ends were also found when readers made exactly three fixations, although these data do not speak directly to the prediction of a variable convenient viewing position. There was a tendency for the third fixation to be longer than the first two, in these cases, but this was dependent upon the position of the initial fixation and upon the type of word being inspected. The tendency was not observed when the first fixation was at the word’s centre, and was clearest when the third


fixation was upon the least informative part of the word (comparing Ir and r1 words). This result indicates that the visual attention given to a half word is determined in part by the information in that half word.

The effects of initial viewing position and of the distribution of information within the word were more apparent with analyses which used data from individual fixations rather than overall measures of the visual attention given to a word. Separating the data according to the number of fixations upon the word revealed different effects. If we had only used an overall measure such as whole word gaze duration or total inspection time, then effects of initial fixation and information distribution would have been obscured. Half-word gazes were longer for beginnings than ends of words, but only for the two types of words with informative beginnings. The total time spent on a word showed only that more visual attention was allocated to the part of the part initially fixated. O’Regan and Levy-Schoen (1987) suggested that an initial fixation away from the convenient viewing location has the effect of increasing the probability that a two- or three-fixation strategy will be used. An initial fixation upon the convenient viewing location would allow the possibility of fewer fixations. The present data do not support this suggestion. The number of fixations did not vary as a function of the point of initial fixation, and neither did the total inspection time upon the whole word.

The third question was whether the degree of informational redundancy influenced the patterns of fixations upon the two halves of a word. The IR type words used in the experiment had very redundant endings, whereas the Ir words had less redundant endings. The IR words are the -date, -ine.ss, -ement, -ative and -ction words which are common in English, with an average of more than 200 beginnings for each of these endings. The Ir words also had redundant endings - they are the -ering, -cture, -isher, -ional, -shing, and -arist words - and had approximately one quarter of the available beginnings that the IR words had. Differences were observed in the fixations allocated to these two word types. When the first fixation was upon the informative half, there were fewer fixations upon IR words; and secondly, when there were two fixations upon the word, then the second was shorter for IR words; and finally, for words gaining three fixations, when the first fixation was at the word’s centre the second fixation (which was upon the informative half) was shorter for IR words. All three measures point to less visual attention being given to the words containing the


more redundant sequences of letters. These are not the most direct measures that were anticipated, however, and the extent of the orthographic redundancy does not extend to influence gaze durations or the durations of each individual fixation with a word. Some fixations show sensitivity, but not all of them. In particular, the first fixation upon the word did not show any influence of the location of information in the word, whereas location of that fixation was effective. This same result was found in the case of our investigation of the inspection of isolated Finnish words (Underwood et al. 1987) - information influenced the duration of the second but not the first fixation, whereas the duration of the first fixation did vary according to whether it was at the word’s beginning or ending. The effects of information processing appear to take time to manifest themselves in the duration of fixations, and are not apparent in the first fixation.

The present results differ from those summarised by O’Regan and Levy-Schoen in a number of respects, and the experiment cannot be considered to be a convincing replication of the work involving the inspection of French words. Our experiments agree in finding that when the initial fixation is towards a word’s beginning, there are fewer fixations, at least for IR words. We disagree about the gaze duration - English gazes on words with informative beginnings are longer than French gazes when they start from the convenient viewing position - but this may be a consequence of our definition of gaze. O’Regan and Levy-Schoen used the more conventional definition of all visual attention prior to leaving the word boundary, whereas we were interested in the visual attention given to each half word. When we did use the total amount of visual attention as our measure - the ‘total inspection duration’ - we found little evidence of sensitivity to lexical processing constraints.

It must also be pointed out that the experimental procedure in the present experiment is not exactly the same as that used in the French studies, and these procedural differences may have contributed to the different pattern of results. The principal difference is that our subjects pressed a button to remove the test word and bring up the comparison words, whereas in the French experiment fixation was terminated by a movement towards masked comparison words. As fixation on the test word ends, the mask lifted from the comparison words, and the test word itself became masked. Differences in methodology may induce different inspection strategies.

278 G. Underwood et al. / Eyefixations and words

A second problem concerns the inspection strategy following an initial fixation at or away from the convenient viewing position. O’Re- gan and Levy-Schoen report that an initial fixation to the left of centre results in the reader being more likely to make just one fixation, and that this fixation will be slightly longer in duration than is normal (about 400 msec). We found too few single fixations to be confident about our conclusions (O’Regan and Levy-Schoen (1987: 370) also report insufficient data for a formal analysis), but these fixations were very long in comparison single fixations made to other locations (see table 4). An imposed fixation to the beginning of the word appears to give rise to a very extended fixation, which, for IR and Ir words, is longer than the summed fixations when two fixations are made (see table 5). Deciding to make one and only fixation, when fixating left of centre, appears to involve costs to processing rather than the benefits seen in O’Regan and Levy-Schoen’s analysis. It is important to point out that in both of these experiments a small number of cases are giving rise to tentative conclusions.

When O’Regan and Levy-Schoen (1987) considered those words gaining exactly two fixations, then a trade-off was noted between the durations of the two fixations. First fixations near the convenient viewing position had long durations, followed by short second fixations (see their fig. 16.6, p. 374). First fixations far from the convenient viewing position were short, and were followed by longer fixations. We also found that if the first fixation was near the beginning of the word then it was the shorter fixation. Similar conclusions are suggested by the nine-letter and eleven-letter words receiving exactly two fixations in O’Regan and Levy-Schoen’s analysis. The longest initial fixations were observed when the first fixation was upon a point near to the centre of the word in both of these experiments. A similar picture also emerges from the analysis of our three fixation cases. For initial fixations at the beginning or the end, the first two fixations were shorter than the third fixation. This relationship did not hold for initial fixations in the word’s centre. If O’Regan and Levy-Schoen’s inference is correct, then we must also conclude that the convenient viewing position for our ten-letter words is near to the centre of the words.

While the sequence of inspection did show effects of the distribution of information, confirming the use of current lexical information by the mechanism which guides the reader’s eyes, it is important to note that we found no effects upon the first fixation upon the word. In the


analyses of the two fixation and three fixation cases, in which the distribution of information influenced the durations of fixation, the initial fixation did not vary according to the informativeness of what was imposed during this fixation. The location of the first fixation within the word was influential, but the information value was not. The effects of informative sequences of letters were to be seen only in the durations of subsequent fixations, suggesting that, at least for these conditions of imposed viewing, lexical processing does not determine the duration of the first fixation.

This is not to say that there is no lexical processing during the first fixation, or that under normal reading conditions the first fixation would be unaffected by lexical processing. However, the evidence from the present experiment suggests that the consequences of processing are not apparent until at least the second fixation. These conclusions are consistent with O’Regan and Levy-Schoen’s ideas about the first fixation being under visuomotor control, and second fixations showing additional sensitivity to lexical frequency and informativeness. The limitation upon these conclusions is that they are drawn from the inspection of displays which have little in common with natural reading, and it remains to be seen whether under more natural conditions we will find the same constraints upon eye movement control. Pre- liminary investigations with sentence comprehension tasks suggest that the first fixation does vary in duration according to the distribution of information within the word, with initial fixations at the ends of words increasing in duration for those words with informative endings (Un- derwood et al. 1988). If this pattern has any generality, then we must conclude that the insensitivity of the present experiment to any effect of lexical processing during the first fixation is a consequence of the imposed first fixation rather than being indicative of the interaction between lexical processing and eye guidance during reading.

References

Bray, A.C., A.C. Dickens and M.A. Holmes, 1983. The advanced user guide for the BBC microcomputer. Cambridge: Cambridge Microcomputer Centre.

Dunn-Rankin, P., 1978. The visual characteristics of words. Scientific American 238, 122-130. Holmes, V.M. and J.K. O’Regan, 1987. ‘Decomposing French words’. In: J.K. O’Regan and A.

Levy-Schoen (eds.), Eye movements: From physiology to cognition. Amsterdam: North-Hol- land.

280 G. Underwood et al. / Eye fixations and worak

Kliegl, R., R.K. Olson and B.J. Davidson, 1983. ‘On problems of unconfounding perceptual and

language processes’. In: K. Rayner (ed.), Eye movements in reading: Perceptual and language

processes. New York: Academic Press.

Kucera, H. and W.N. Francis, 1967. Computational analysis of present-day American English.

Providence, RI: Brown University Press.

O’Regan, J.K. and A. L&y-Schoen, 1987. ‘Eye-movement strategy and tactics in word recognition

and reading’. In: M. Coltheart (ed.), Attention and performance XII: The psychology of

reading. London: Erlbaum.

GRegan, J.K., A. L&y-Schoen, J. Pynte and B. Brugaillbre, 1984. Convenient fixation location

within isolated words of different length and structure. Journal of Experimental Psychology:

Human Perception and Performance 10, 250-257.

Rayner, K., 1979. Eye-guidance in reading: Fixation locations within words. Perception 8, 21-30.

Schwartz, J.C.P., 1986. Backwords for crosswords: A reverse-sorted word list from Chambers 20th

Century Dictionary. Edinburgh: Chambers.

Simpson, J., 1985. Chambers words for crosswords, Scrabble and all other word games. Edin-

burgh: Chambers.

Underwood, G., R. Bloomfield and S. Clews, 1988. Information influences the pattern of eye

fixations during sentence comprehension. Perception 17, 267-278.

Underwood, G., J. Hyena and P. Niemi, 1987. ‘Scanning patterns on individual words during the

comprehension of sentences’, In: J.K. G’Regan and A. Levy-Schoen (eds.), Eye movements:

From physiology to cognition. Amsterdam: North-Holland.

Wilkinson, H.P., 1979. Eye movements measured with a solid state photodiode matrix. Behavior

Research Methods and Instnmrentation 11, 429-432.

eye fixations are influenced by the distribution of information within words

Documents