Reading Text Under Normal and Disappearing Presentation Conditions

AbstractIn this article I discuss data from a series of experiments in which readers＇ eye movements were recorded as they processed sentences in which each word disappeared or was masked 60ms after fixation onset. We used this paradigm to investigate whether we could induce a gap effect during reading, and how visual and linguistic factors affected eye movements under these conditions. The data showed that no gap effect occurred in our experiment. Overall reading times were the same under normal and disappearing presentation conditions.However, readers did adopt a strategy of making fewer but longer fixations when the text disappeared than when it did not. Additionally, clear frequency effects occurred regardless of whether the text was presented normally or disappeared. This finding indicates that while the visual uptake of information is important, cognitive processes associated with the lexical identification of words are a primary influence on when readers move their eyes during reading. The findings are taken to support the E-Z Reader model of eye movement control.

Key wordseye movements, disappearing text, reading.

In this paper I will summarise a series of experiments in which we[1,2]investigated whether it was possible to induce a "gap effect" during reading and to examine how well people can read when the text literally disappears from in front of their eyes. To carry out the work we recorded readers＇ eye movements and employed a novel saccade contingent change technique in which we made the word that the reader was fixating disappear 60ms after fixation onset. In addition to investigating whether a gap effect might occur during reading, this work also investigated whether the primary determinant of when we move our eyes during reading is visual or linguistic processing. To this extent, the experiments I will discuss in this paper represent work that is relevant to two sub-disciplines of research that exist within the community of researchers investigating eye movements during reading: Research investigating oculomotor control per se, and research that employs eye movements as a tool to study the psychological processes associated with written language comprehension. There is a logical dependence between visual and linguistic processing during reading(visual extraction of information necessarily must occur prior to the initiation of linguistic processing of that information). Note also that eye movements are the basic behavioural interface between the stimuli(the words of a sentence and processing of those stimuli). Thus, I will argue that the precise characterisation of eye movement behaviour and the relationship between it and underlying visual and linguistic processing is absolutely necessary if we are to develop a comprehensive understanding of the psychological process of written text comprehension.

The Gap Effect

The gap effect is a basic, robust oculomotor phenomenon whereby the time to make a saccade is reduced when the object currently under fixation disappears prior to saccade onset, compared with when it disappears after saccade onset[3]. For example, if a participant is required to fixate a central cross and as quickly as possible make a saccade to a target square that appears to the right of the cross, then they will be faster to make that saccade if the cross disappears prior to the target onset than if the cross remains on the screen until after the target appears. The basic phenomenon has been repeatedly demonstrated under different experimental conditions[4~6], and has been shown to occur for both pro-saccades to fixation onsets as well as cognitively controlled anti-saccades in the opposite direction to a target onset[7~9]. It has been argued that two factors contribute to the reduction in saccade onsets that is the gap effect. The first is the discontinuation of foveal stimulation which permits a speeded release from the currently fixated position. This may occur because fixation offset causes a reduction in fixate cells within the superior colliculus(cells that are, at least in part, responsible for the maintenance of a stable fixation)which in turn permits speeded saccade initiation. A second possible contributory factor is that the fixation offset acts as a preparatory signal as to saccade initiation and this in turn speeds saccade triggering.

The fact that the gap effect has been shown to occur for cognitively controlled anti-saccades as well as reflexive pro-saccades indicates that, at least in principle, it might be possible to influence saccades under cognitive control that occur during reading in a similar way. That is to say, if it were possible to make a word disappear shortly after it had been fixated, then would this cause the time to saccade onset(i.e. the fixation duration on the word)to be reduced. We set out to test this possibility in the first of the experiments I will discuss here. To do this, we used a variant of the saccade contingent change paradigm[10,11]. In our experiment we had two presentation conditions-a normal condition in which single sentences were presented in the usual way on the computer screen. In addition to this condition we had a disappearing text condition in which participants were again required to read the sentences normally, however, we manipulated the display such that 60ms after the participant first fixated a word it disappeared from the screen. When participants then made a saccade away from that word to fixate another word, the word from which they made the saccade immediately reappeared and the newly fixated word disappeared 60ms after fixation onset(see Figure 1 for a schematic of this procedure). Importantly, if participants made more than one fixation on a word, the word did not reappear at the onset of the second fixation on the word. To be clear, the reader had to make a saccade from one word to fixate a different word in order for the originally fixated word to reappear.

For our disappearing text manipulation, we decided to make the word disappear 60ms after fixation onset on the basis of existing studies that have demonstrated that masking text 50~60ms after fixation does not produce disruption to processing [12~14]. Thus, while we wished to remove the foveal word from the screen prior to saccade onset, we did not wish to disrupt processing such that normal reading could not occur. By presenting sentences under normal and disappearing conditions, we investigated whether it was possible to obtain a gap effect during reading. To be precise we hypothesised that if the discontinuation of the foveated stimulus caused a gap effect to occur during reading, then fixation times as well as overall reading times for the sentences should be shorter when the text disappeared compared with when it did not.

Visual and Linguistic Influences on When to Move the Eye

In addition to manipulating the manner in which the sentences were presented to participants we included two further important manipulations. We identified two critical words within each sentence. For the first of these critical words we manipulated its length such that it was either long (10 letters)or short(4 letters), while both were matched for frequency. For the second critical word we manipulated its frequency such that it was either high frequency(mean=105 counts per million) or low frequency(mean=1 count per million), but was always 6 letters long(see Table 1 for example stimuli). Word frequencies were calculated using the CELEX database[15].

Table 1. Example sentences. Short and long target words are shown in italics. Frequent and infrequent target words are underlined.

1. Yesterday the office boss/supervisor moaned about the broken/snazzy equipment upstairs.

2. Sam wore the horrid coat/spectacles though his pretty/demure girlfriend complained.

3. He found the secret swag/manuscript inside the little/sturdy farmhouse on the hill.

4. A proper gift/collection scheme boosted the annual/frugal donations to the charity.

These manipulations are important because they allowed us to investigate the influence of visual factors(such as word length), as well as linguistic factors(such as frequency)on eye movement behaviour even after a word has disappeared. This question is itself important because there has been substantial debate within the eye movement community concerning whether visual or linguistic processes play the primary role in determining when a saccade is initiated in order that a different portion of text be fixated. Broadly speaking, the debate can be characterised as being between two opposing positions, namely, those researchers who argue that fixation durations in reading are primarily influenced by higher order cognitive processes associated with language processing, and those who argue that fixation durations are primarily influenced by low level processes associated with the perception of the visual information from the page or screen. Proponents of the latter position[16,17]argue that basic visual scanning routines are adopted during reading and that these routines are very rarely influenced by the cognitive systems associated with linguistic processing of the fixated word. By contrast, those researchers who argue that eye movements reflect higher order cognitive processes[18] believe that such processes play a central role in the decisions concerning when a fixation will be terminated through saccade initiation. Perhaps the most explicitly developed theoretical account from the cognitive control position is the E-Z Reader model of eye movement control during reading[19,20]. Space limitations prevent extended discussion of this model, however, suffice it to say that within the E-Z Reader model completion of a stage of processing associated with word identification dictates when the decision to move the eye is made. Thus, according to this model, a cognitive process associated with language comprehension rather than a visual scanning procedure determines when the eyes are moved during reading.

Experimental Details

We tested sixteen naïve native English speaking participants who were all University of Durham undergraduate students. We constructed 40 experimental sentences, each including a long/short critical word and a high/low frequency critical word. We also included 10 filler items and we presented the sentences to the participants in two blocks. One of the blocks was presented normally and the other under disappearing text conditions. The order of blocks was counterbalanced across participants. During the experiment we used a Dual Purkinje eye tracker to monitor participants＇ eye position as they read the sentences.

In order to address the first of the two questions we set out to investigate, that is, whether we could induce a gap effect during reading, we undertook analyses of a number of measures. We computed total sentence reading time for the complete sentence. For each word of the sentence we also computed first pass reading times(the sum of all fixations on a word prior to a fixation on a different word)and total reading times(the sum of all fixations on a word). In addition, we calculated frequency distributions for first fixation durations, gaze durations and the lengths of the inter word saccades.

The total sentence reading times were the same for text presented under disappearing conditions(3327ms)as for text presented normally(3286ms). Note also that comprehension questions were also responded to with equal accuracy under the different display conditions(85% correct). Additionally, there were no differences between first pass reading times and for total reading times for each word of the sentence. Together these measures indicate that, consistent with previous research[12~14], participants found sentences no more difficult to read when the text disappeared after 60ms than when it did not. Furthermore, these data provide no evidence that reading speeded up when text disappeared compared with when it was presented normally. This in turn indicates that at least under our experimental conditions, a gap effect did not occur during reading. While we obtained no evidence of a gap effect in the present experiment, the disappearing text manipulation did influence eye movement behaviour in an important respect.There were differences in the patterns of eye movements under the two conditions such that readers made fewer fixations when the text disappeared(12.7 per sentence)than when it did not(13.6 per sentence), but the mean duration of fixations under disappearing text conditions was longer(264ms)than when the text was presented normally(248ms).

Figure 2 Panel A shows the frequency distributions for gaze durations and Panel B shows the frequency distributions for total reading time for normal and disappearing text.

This difference is also reflected in the frequency distributions(See Figure 2)for first fixation duration(Panel A)and gaze duration for each word of the sentence(Panel B). The distributions of reading times are clearly shifted to the right when the disappearing compared with normal text is read. Furthermore, for the gaze duration data it is consistently the case that gaze durations were longer under normal conditions than under disappearing conditions for bins of data greater than 375ms. This is probably due to readers making fewer refixations on words under disappearing conditions(7% of trials)than under normal conditions(14% of trials). Thus, while there was no overall difference in comprehension times for the normally presented and disappearing sentences, readers did appear to adapt their reading strategy such that they made fewer but longer fixations when the text disappeared compared with when it did not. Furthermore, readers were less inclined to make a refixation on a word in the disappearing text condition compared with the normal condition presumably because they realised that the refixation would provide no further information for them to process. Finally, as one might expect, given the similarity between overall reading times, yet differences in the mean number of fixations and their duration, mean saccade lengths were also longer when the text disappeared(9.1 characters)than when it did not(8 characters).

To summarise, overall the sentences were no more difficult to read under disappearing text conditions than under normal conditions. However, participants did adopt a strategy whereby they made fewer fixations of slightly longer duration when the text disappeared compared with when it did not.

Let us now turn to the local analyses that we conducted to examine reading times for the long and short and the high and low frequency critical words that were embedded in the sentences. For these regions we compared first fixation durations and gaze durations(see Figures 3 & 4).

Figure 4 Panel A shows the first fixation durations and Panel B shows the gaze durations(in ms)for the high and low frequency critical words.

For the long and short critical words, for the first fixation duration data under normal conditions, showed clear length effects such that initial fixations were longer for long than for short words. However, under the disappearing text conditions, there was no such difference with equally long first fixations regardless of critical word length. This result is consistent with the strategy that participants adopted under the disappearing text conditions whereby they were more likely to make single, long fixations when words disappeared than when they did not. A similar pattern of effects was obtained for the gaze duration data with a very pronounced length effect under normal conditions, but an effect of far smaller magnitude under disappearing text conditions.

Perhaps the most interesting data that the present experiment produced were the first fixation and gaze duration data for the critical high and low frequency words. Here we obtained reliable frequency effects for both measures regardless of whether the text disappeared, or was presented normally. That is to say, we found that first fixations and gaze durations were shorter for high frequency words than for low frequency words under both presentation conditions. Thus, our results demonstrated that under the disappearing text conditions when the fixated word disappeared after 60ms, readers continued to fixate a blank portion of the sentence for a period of time that was inversely proportional to the frequency of the word that was no longer there. The results strongly suggest that processes associated with the lexical identification of the word, rather than those associated with the perception of the visual information were primarily responsible for determining when the eyes were moved during reading.

Experimental Controls

The data from the present experiment are relevant to two important theoretical questions. The first of these was whether it is possible to induce a gap effect during reading. The data from the first experiment appear to indicate that this was not the case(at least with the disappearing text manipulation that we adopted). We felt that one possible reason why we may not have obtained a gap effect might have been because participants were required to not only perform a simple visual task involving only saccadic orienting, but instead in our experiment we required to both perform saccadic orienting as well as complex concurrent linguistic processing. To investigate this possibility, we conducted an additional experiment in which we replaced each letter of each word in our sentences with an X, and then required participants to scan these arrays of X＇s under normal and disappearing conditions. To reiterate, this manipulation allowed us to investigate whether our failure to obtain a gap effects was a consequence of participants having to concurrently conduct linguistic processing and saccadic orienting. Clearly, in the experiment using X strings rather than words, participants were still required to perform saccadic orienting, but did not perform linguistic processing. In fact, the results from this control experiment showed that even when participants were simply scanning(rather than reading), no gap effect occurred. This suggests that the failure to obtain a gap effect was not due to concurrent linguistic processing and saccadic orienting. Interestingly, however, we did find that participants adopted a similar strategy to that they employed in the first experiment whereby they made fewer but longer fixations when scanning disappearing X＇s than when scanning arrays of X＇s presented normally.

The second theoretical question to which our data are relevant concerns whether visual, or alternatively, cognitive processes are the primary determinant of when we move our eyes during reading. The frequency effects that we observed for normal and disappearing text appear to demonstrate quite compellingly that cognitive processes associated with language processing rather than processes associated with uptake of visual information are primarily responsible for when we move our eyes during reading.However, it could be argued that this interpretation of the frequency effects should be treated with caution. For example, an alternative explanation of these effects is that after each of the words of the sentence disappeared, readers may have access to a visual iconic trace of the word stored in iconic memory. If readers were able to make use of iconic memories of briefly presented stimuli in this way, then clearly the frequency effects that we observed in our first experiment might indeed be due to visual processing of an iconic trace, rather than, as we have suggested, lexical identification of a word. To ensure that this was not the case, we conducted another control experiment in which we presented the sentences in the same way as we did in the first experiment, other than for one important difference. In this control experiment, instead of making the text disappear after 60ms, we replaced the word with a string of X＇s that served to backward mask the word after it had disappeared, and therefore overwrite any iconic memory of the word. We anticipated that if we still obtained a frequency effect under these conditions, then the effects could not be due to visual processing of an iconic memory and that our former conclusion must be correct. Our results showed very clearly that even when we used this masking procedure we still obtained a reliable frequency effect supporting our claim that cognitive processes underlying language processing, and more specifically, lexical identification are the primary influence on when we decide to move our eyes during reading.

Conclusion

In this paper I have discussed a series of experiments to investigate the possibility of inducing a gap effect during reading and to determine how readers process disappearing text. The data from these experiments indicate that making a fixated word disappear 60 msec after fixation onset, does not induce a gap effect during reading. Furthermore, the failure to obtain a gap effect is not a consequence of concurrent visual and linguistic processing. Additionally, our data show that readers continue to fixate a word＇s position after it has disappeared for a period of time that is inversely proportional to the frequency of that word. This is true even when any iconic memory for the word is overwritten by a backward visual mask. These results strongly suggest that the primary determinant of when to move the eye during reading is processing associated with the lexical identification of that word rather than processes associated with the uptake of visual information.

References

1 Liversedge S P, Rayner K, White S J, et al. Eye movements when reading disappearing text: Is there a gap effect in reading? Vision Research, 2004, 44: 1013~1024

2 Rayner K, Liversedge S P, White S J, et al. Reading disappearing text: Cognitive control of eye movements. Psychological Science, 2003, 14: 385~388

3 Saslow M G. Effects of components of displacement-step stimuli upon latency for saccadic eye movement. Journal of the Optical Society of America, 1967, 57: 1024~1029

4 Kingstone A, Klein R M. Visual offset facilitates saccade latency: Does pre-disengagement of attention mediate the gap effect?Journal of Experimental Psychology: Human Perception and Performance, 1993a, 19: 251~265

5 Findlay J M, Walker R. A model of saccade generation based on parallel processing and competitive inhibition. Behavioral and Brain Sciences, 1999, 22: 661~674

6 Pratt J, Bekkering H, Leung. Estimating the components of the gap effect. Experimental Brain Research, 2000, 130: 264~268

7 Abrams R A, Oonk H M, Pratt J.Fixation point offsets facilitate endogenous saccades. Perception & Psychophysics, 1998, 60: 201~208

8 Craig G L, Stelmach L B, Tam W J. Control of reflexive and voluntary saccades in the gap effect. Perception & Psychophysics, 1999, 61: 935~942

9 Forbes K,Klein R M. The magnitude of the fixation offset effect with endogenously and exogenously controlled saccades. Journal of Cognitive Neuroscience, 1996, 8: 344~352

10 McConkie G W, Rayner K. The span of the effective stimulus during a fixation in reading. Perception & Psychophysics, 1975, 17: 578~586

11 Rayner K. The perceptual span and peripheral cues in reading.Cognitive Psychology, 1975, 7: 65~81

12 Ishida T, Ikeda M. Temporal properties of information extraction in reading studied by a text-replacement technique. Journal of the Optical Society A: Optics and Image Science, 1989, 6: 1624~1632

13 Rayner K, Inhoff A W, Morrison R, et al. Masking of foveal and parafoveal vision during eye fixations in reading. Journal of Experimental Psychology: Human Perception and Performance, 1981, 7: 167~179

14 Slowiaczek M L, Rayner K. Sequential masking during eye fixations in reading. Bulletin of the Psychonomic Society, 1987, 25: 175~178

15 Baayen R H, Piepenbrock R, Gulikers L. The CELEX lexical database.(CD-ROM). Philadelphia: Linguistic Data Consortium, University of Pennsylvania, 1995

16 O＇Regan J K. Optimal viewing position in words and the strategy-tactics theory of eye movements in reading. In K. Rayner (Ed), Eye movements and visual cognition: Scene perception and reading. New York: Springer-Verlag, 1992. 333~354

17 Yang S N, McConkie G W. Eye movements during reading: a theory of saccade initiation times. Vision Research, 2001, 41: 3567~3585

18 Liversedge S P, Findlay J M. Saccadic eye movements and cognition. Trends in Cognitive Science, 2000, 4: 6~14

19 Reichle E D, Rayner K, Pollatsek A. Eye movement control in reading: Accounting for initial fixation locations and refixations within the E-Z Reader model. Vision Research, 1999, 39: 4403~4411

20 Reichle E D, Rayner K, Pollatsek A. The E-Z Reader model of eye movement control in reading: Comparisons to other models. Behavioral and Brain Sciences, 2003, 26: 445~526

Acknowledgments

This research was supported by a grant from the Leverhulme Foundation to support K. Rayner＇s stay at the University of Durham as a Visiting Leverhulme Professor, by Grant 12/S19168 from the Biotechnology and Biological Sciences Research Council, and by Grant HD26765 from the US National Institute of Health.