An Interview...

Dr. Keith Rayner – What Eye Movements Tell Us About the Processing Involved In Reading

Dr. Keith Rayner is a Distinguished Professor with the Psychology Department at the University of Massachusetts and the Director of the University's Eyetracking Laboratory.  Dr. Rayner's work focuses on understanding the cognitive processes involved in reading by studying eye movements and making inferences about how the movement of the eyes are related to the underlying processing involved.  Dr. Rayner's study of eye movements during reading is an important cornerstone of today's reading science.  Additional bio info

Note: Remember to click on any word on this page to experience the next evolutionary step in technology supported reading.

(see 'Interview Notes' for more details). Bold is used to emphasize our [Children of the Code] sense of the importance of what is being said and does not necessarily reflect gestures or tones of emphasis that occurred during the interview.

Personal Background:

David Boulton: Let’s start with just a brief sketch of how it is that you come to what you’re doing.

Dr. Keith Rayner: When I was an undergraduate student in the 1970’s all of psychology at the university I was at was mainly either clinical or social, or animal behavior. I was interested in how people think, and I saw an ad brochure for Cornell, where I went to graduate school, about come and study reading and what kids and adults do when they read. So, I went off to Cornell and that evolved into doing research on the reading process.

I mainly work on skilled reading. I’ve done a few things with kids and I wrote a paper that was in Psychology Science in the Public Interest. A version of it was published in Scientific American, about how to teach reading.

The Processes of Skilled Reading:

Dr. Keith Rayner: Most of my own research is with skilled readers. So I’m interested in: What is it that people do when they’re skilled readers? What are the processes that they engage in?

I largely approach it by recording their eye movements while they read, and then try to infer what the mental processes are. We also do things where we make changes in the text, contingent on the eye movements. So, the classic experiment is one where we control how much you can see on each eye fixation via what’s called the “moving window technique.” Or we have also used another technique called the “boundary technique,” where you’re reading along, and there’s a word off to the right of where your eye is, but then when your eye goes into motion that word changes to another word.

David Boulton: So, you say you’re introducing a confusing or ambiguous situation and seeing what to do with it?

Dr. Keith Rayner: Well, it isn’t ambiguous. What we do is vary the relationship between the first word and the second word, so they can be semantically, phonologically or orthographically related. Then we try to infer what kind of information you would be getting from that word before you got to it, as a function of the relationship between the first and second word, and how long you then look at the word that’s there when your eyes get there.

We’ve done a little of this with children, but it’s mostly done with adults to try and see how much information you’re getting different distances from where your eye is currently looking at the moment.

Then the other thing I do a lot of is have people read sentences or text that do indeed have ambiguities in them, so there would be phonological ambiguity or lexical ambiguities or syntactic ambiguities. So they might read a sentence like, “While Mary was mending, the socks fell off her lap.” And there may or may not be a comma after mending, but if there’s not a comma after was mending, then usually people misplace the sentence because they take the sock to be the object that the verb was mending, and then they’re all messed up. So we try and infer what kind of structures are they building as they read to help them parse the sentence.

Or in the case of lexical ambiguity, they might read a sentence where the word bank is there, and then it turns out that it’s referring to a river bank, rather than a money bank. And so again, we try and see — infer what kind of processes they are using to understand by looking at what their eyes do.

David Boulton: Do you also put multiple tasks in to see whether they stack up to consume more time between fixations?

Dr. Keith Rayner: Well, in fact, there are fixation time differences as a function of these manipulations that we make. The task, for us, is typically always just read the text. Then there are variations in how long people look at things, as a function of these kinds of manipulations that we make.

The Criticality of Timing:

David Boulton: One of the things that interests me — I guess you made reference to the dual task paradigm — but ultimately saying that it takes the brain time to do these subprocesses that aggregate into this thing we call reading…

Dr. Keith Rayner: Yes.

David Boulton: …And that if one of these subprocesses takes too long, this can add up to create a dropout in the flow of reading.

Dr. Keith Rayner: Yes.

David Boulton: That’s one of the things that we’re really interested in.

Dr. Keith Rayner: Right. That’s right.

Ambiguity Processing Takes Time:

David Boulton: And you’ve done some work to show that there’s a correspondence between ambiguity and time processing.

Dr. Keith Rayner: Yes.

David Boulton: What can you say about that?

Dr. Keith Rayner: Well, which kinds of ambiguity are you most interested in? There can be ambiguity at all…

David Boulton: In all levels, of course. The thing that most interests us, that it seems to us that the children are most struggling with, is the letter sound correspondence ambiguity.

Dr. Keith Rayner: Uh-huh.

David Boulton: That they’re kind of growing up primed to have an initial response to the letter name or sound of a letter, which, as you know, in the beginning phases, is rarely the case. There’s a dropout. When we watch children, it seems like there’s a direct correspondence between the starts, stops, hesitations or stutters of processing, and code confusion at this level.

Dr. Keith Rayner: Uh-huh.

David Boulton: Do you have anything that shines light in there?

Dr. Keith Rayner: Again, most of our work — I mean, I agree that this is a big problem for children, the ambiguity that exists in the grapheme phoneme correspondence rules for English. Although I don’t think it’s as haphazard as critics of phonics approaches would say.

David Boulton: It’s clearly not haphazard. By the way, just some context: I had a great conversation with Richard Venezky and subsequently, just last week, with Keith Stanovich and Marilyn Adams, all of which point to you. They all got a lot from you. So it’s really a delight to talk with you.

Adult - Child, Reader - Non-Reader Perspective Differences:

David Boulton: Regarding this particular point, while we as adults on the other side of the code may look back upon it with our knowledge of it, and say, “Jeez, it’s really not that confusing, we can see the patterns in it.” But that doesn’t necessarily speak to the kinds of confusions and the time that’s associated with processing those confusions experienced by a child who doesn’t know what we know.

Dr. Keith Rayner: Right. But even for us who do know, there will be longer pauses on words that are irregular, for example.

David Boulton: Yes.

Dr. Keith Rayner: So that indeed when people read a word that’s irregular — do you know what I mean by irregular, like pint? All the other words that end in i-n-t are like hint and things like that.

David Boulton: Right.

Dr. Keith Rayner: But then pint is different — they might think it’s going to be pint with a short sound by virtue of all the other things. When you encounter those kind of words as you read, and look at them, your eyes pause longer on them.

David Boulton: Right. So there is a correspondence between processing time and ambiguity. We could say that it’s possible, though we may not have done it yet, to lay out a map of these nested levels of ambiguity that are involved in the challenge.

Dr. Keith Rayner: Right.

The Time It Takes to Make Meanings:

David Boulton: Some reading researchers speak of the overall time, kind of the net average it takes or is required to process letters into sounds before attention will break down; in other words, before the ability to stay in the flow will just erode, because you spent too much time working on a particular ambiguity. Have you been able to zero in on any time metrics relative to that?

Dr. Keith Rayner: Probably not. I mean, that’s clearly a problem, and it’s a bigger problem for children than adults. The time estimates that we’ve tried to map out are: How long does it take to go from the printed representation to the sort of lexical access to understand the meaning. That’s where a lot of our focus has been. So, we’ve built a model of the reading process. Basically, it’s a simulation, where we predict where people will look, how long they’ll look at a word, which words they’ll skip, and things like that. The model does a pretty good job of doing that. Implicit in the model is the assumption that lexical access is taking place in around somewhere — you know, about 150 milliseconds or so, or less.

But in terms of how long does it take to map the graphemic features onto the phonological features, I don’t know that I’d have a good time frame there. We have done experiments where we try and map out how quickly the different types of information are getting into the system. There it looks like orthography obviously comes first, and phonology comes in pretty quickly.

David Boulton: Right. When we’re first beginning to read — I have a different language for this, forgive me, I’m not in your field of science.

Dr. Keith Rayner: That’s okay.

David Boulton: But I really want to communicate with you. I understand that you really have done something seminal that’s influenced a lot of people here.

Disambiguation and Decoding:

In the early days of alphabetic reading, like in Plato and the Greeks and the Romans, there was a pretty tight correspondence between letters and sounds. That meant: you see a letter, you have an immediate response inside yourself as to the sound of that letter, and you say it together in a blended Gattling-gun kind of fashion, and out pops code-cued speech, and that’s where we began. That’s what it seems like.

Dr. Keith Rayner: Uh-huh.

David Boulton: But in the English language and other romance languages, where there was a mapping of the local oral system to the Latin system and the Roman Alphabet, correspondence problems developed. The consequence is that now, rather than just seeing a letter and popping with the sound, the letter information somehow needs to be suspended and worked out in a growing or expanding contextual memory field. It’s almost like, in computer speak, that it has to be ‘buffered’ and operated upon, rather than just seen and responded to. And it takes time to do this. One of the problems with our reading science language here is that when we talk about decoding, buried inside of what we normally call “decoding,” there’s a disambiguation overhead. That’s what I’m trying to get at.

Dr. Keith Rayner: Okay.

David Boulton: So I understand you’ve not done any work with children, that has brought you to a point where you can actually see the time consumption and its relationship to the overall processing flow and attention of a coherent reading stream relative to the letter sound correspondences. But based on what you do know, what else could you say about that space?

Word Recognition Timing Thresholds in Skilled Readers:

Dr. Keith Rayner: What I know for adults, which is where all of our work is done, is that it takes about fifty milliseconds to get the information into the system. We’ve done experiments where people are reading, and if we mask the text or we take it away, but they get to see it for fifty milliseconds, that’s enough time.

David Boulton: So, by controlling how long something appears, you can find out what the recognition threshold is?

Dr. Keith Rayner: We can get the threshold for: How long does this stuff have to be there to get it into the system? If we leave it up for less than fifty milliseconds then reading falls apart; it’s really disrupted. But if we give them fifty or sixty milliseconds to see the text, then the orthographic information that’s there in front of their eyes gets into the system. It doesn’t matter if the text goes away after that. In fact, there are interesting things like if they’re looking at a low frequency word — so when you read normally, right, and we record your eye movements, you’ll always look longer at a low frequency word than a high frequency word. Even when we take the text away or mask it so it’s not there anymore, the eyes will stay in place longer for the low frequency word than the high frequency word, which is pretty interesting. So, that’s basically saying the mental operations that are involved in figuring out what that word is are driving the eye movements through the text.

David Boulton: So, there’s a loop going on that the eyes are a part of, being directed by, freed up, told to move on, at each iteration of some kind of coherent capture of object.

Dr. Keith Rayner: Yes. We’ve done other experiments where we present — do you know what the word “prime” means?

David Boulton: In the sense that it’s cuing up the — priming you for what’s coming next; it’s creating a decreased field of possibility around what’s coming?

Saccades and Fixations:

Dr. Keith Rayner: Yes. So, imagine this scenario: You’re reading a sentence, and think of a target word in the sentence somewhere, say it’s the fifth word in the sentence. Before your eyes get to it, there’s just a string of random letters there. So we’re using this technique where we’re going to change things contingent on when your eye moves to that word. We make this change always during a saccade, during the eye movements, so that vision is expressed, you never see this change take place.

David Boulton: A saccade is a…

Dr. Keith Rayner: Saccade is a fancy name for eye movement.

David Boulton: Okay. So, when the eye is in motion is when you create this change so that you don’t actually get an attention blip when you’re on a more static eye fixation.

Dr. Keith Rayner: Right. The change takes place during saccade, you don’t see it. So in our standard boundary paradigm, that’s what would happen. We’d present a string of letters; or we’d present a word, and you’d move into the word, and as soon as your eye starts approaching — you know, coming into a word, it changes to another word.

Now, then in a slightly different variation of that, we present a string of random letters where the fifth word should be. Your eye moves into that location. Now we present a prime word, which only stays on the screen for the first thirty or forty milliseconds of the eye fixation, and then it changes to the word that’s going to be there the rest of the time. Now, this change will get seen, because it’s happening during a fixation. There are two changes in this particular paradigm. What we do in those experiments is we vary the orthographic phonological or semantic relationship between the prime word and the target word.

David Boulton: So, you’re actually wanting to see how they recover from their own prediction?

Dr. Keith Rayner: Well, it’s not necessarily their prediction. It’s how much use they’re able to get in that early time frame of something that’s orthographically similar, or phonologically similar, or semantically similar to the target word.

David Boulton: Okay.

The Milliseconds Between Orthography and Phonology:

Dr. Keith Rayner: What we find is that the orthography kicks in very quickly, so that over the whole priming interval that we look at the orthography, if the prime word is orthographically similar to the target word, you get some facilitation. But what’s really interesting to me is the phonology kicks in really quickly, too. I mean, it’s lagging behind the orthography only by about five or ten milliseconds, so that the orthographic information is getting in very quickly, but you’re quickly converting that to a phonological representation.

We’ve also done experiments, going back, now to the first paradigm I mentioned, where there’s just one change. So think of this: You’re reading and the word beach is your target word b-e-a-c-h. Now, before you get to that word, it could say, b-e-e-c-h, which is a homophone of b-e-a-c-h; or is could say b-e-n-c-h, which is orthographically very similar. Or it could just be another word like house, which is just totally unrelated to the prior word.

David Boulton: Right. Different levels of disruption to the expectation field that’s unfolding.

Dr. Keith Rayner: Yes. So, what we find is that you get facilitation from having the homophone, above and beyond having the orthographic control word, so that when people’s eyes land there, one of those things was there first, either the homophone beech or the orthographic control word bench or the unrelated word, or the identical word is there, so b-e-a-c-h could be there.

David Boulton: Yes.

Dr. Keith Rayner: Then we look at: How long does it take to process this target word b-e-a-c-h as a function of what was there before? If the homophone was there before, you’re about as fast as if the identical word was there, and you’re faster than if b-e-n-c-h was there. So before you get to a word, before the eyes actually fixate on it, there’s a preview benefit; you get some facilitation of the upcoming word, and phonology, again, is kicking in very quickly.

David Boulton: So, is that suggesting, you think, that somehow the phonological processing is being primed?

Dr. Keith Rayner: Yes, the phonological information is being primed. You’re getting a head start on the phonological information before your eyes actually get to the word.

Five to Twenty Milliseconds:

David Boulton: That’s fascinating. So you’re saying for a good reader there’s only about a five millisecond delay between the orthographic information clicking and the phonological pattern of that coming in.

Dr. Keith Rayner: Yes. Right.

David Boulton: That’s exactly where the children would have trouble, is that it’s taking them longer to get the orthographic information to make a recognizable phonological pattern.

Dr. Keith Rayner: Right. I wouldn’t take “five milliseconds” literally. It’s something like five to twenty milliseconds.

David Boulton: Okay.

Dr. Keith Rayner: Something like that.

David Boulton: But it’s still the window that I’m interested in.

Dr. Keith Rayner: Yes, right. Here’s another thing that may be relevant to you. I’ve got a graduate student here, Kathryn Chace, who’s doing that same experiment that I just described to you, where there’s a homophone or an orthographic control as a preview for the word before you get there.

David Boulton: Right.

Dr. Keith Rayner: She’s doing it college students. There are good and poor readers and so she gets their reading ability by using the Nelson-Denny Test, which is a test to evaluate reading skill. What she finds is that the good readers all show the pattern that I just described to you. The poor readers don’t. The poor readers don’t show facilitation from the homophone preview. So somehow, for them, quickly getting this head start on the phonology isn’t happening for them.

David Boulton: That also fits right in here. Have you done anything with ESL students or adults that would be otherwise good language processors in their native language? People that would be struggling with the orthographic phonological ambiguity of the English language, and allow you to test against this point?

Dr. Keith Rayner: Well, that’s interesting, because we were talking about that the other day. We haven’t actually done it. We thought it would be an interesting thing to do to try and find people who are learning English and aren’t proficient in it yet and do some of these kind of experiments with them, but we haven’t done it yet.

David Boulton: I think that would be great. I would love to hear about that, and particularly this one zone – this lag time between orthography and phonology.

Dr. Keith Rayner: Right.

Phonics and Whole Language:

David Boulton: So, is there anything else in this particular space we should talk about? Suppose that you have the microphone, and that there’s an opportunity to talk to other reading researchers and teachers and the public in general about what you’ve learned about reading, and what you think, in various degrees, people are not getting out there about what you’ve come in to understanding.

Dr. Keith Rayner: I think ten years ago there was this problem where — you know all about the whole language and phonics debate, and so forth?

David Boulton: Yes. Both of which are compensations for this underlying code ambiguity.

Dr. Keith Rayner: Right. I think that for a long while too many people were deceived by the whole language approach — which has some positive features to it, such as having kids read and have it be enjoyable and so forth; that’s a good thing to do. There are other aspects of it that are positive as well. But they made a lot of assumptions which I think are not accurate, and I think that phonics really does make explicit to the kid what the relationships are that exist between the graphemes and the phonemes, and that’s why, at the end of the day it’s the best way to teach reading.

David Boulton: It certainly is at the moment. But it’s also a patch code, right?

Dr. Keith Rayner: Yeah. I think lately people have — the pendulum has swung back the other way towards people realizing that phonics is important, and that you can’t just throw that kind of instruction out. There’s clearly other things, as you suggest, that could be done. I think that the level of beginning reading instruction, that’s a good thing that people are more aware of those kind of things now.

Moving Windows and Perceptual Spans:

Dr. Keith Rayner: Most of my research, again, has focused on adults and skilled readers. The one thing that I did with children along these lines was we did these experiments with moving windows. There what we do is: You’re reading, and then a region around your fixation point, normal text is available to you, but everywhere else we’ve replaced the letters with garbage. So, if you’re standing behind the person, it looks like a window is gliding across the text. But the window is being controlled by the reader’s eye movements. Wherever the reader looks, they get to see the window. And then outside of the window there’s nothing, or there’s garbage. Then we vary the size of the window, so we make it very small. With adults we’ve made it as small as one letter, so that means letter by letter reading, literally, which is very hard.

David Boulton: Right.

Dr. Keith Rayner: But the real rationale behind all this is: How big does the window have to be so that people don’t read any differently than if all the text is there? What we find is that for skilled readers you need about three or four letters to the left of your fixation point, and about fifteen to the right. We call it the “perceptual span for readers.”

David Boulton: And that’s necessary because of the recursion that’s going on, yes?

Dr. Keith Rayner: Yes, and because of the lack of acuity, so that — vision drops off pretty rapidly from the center of vision.

David Boulton: But I mean, the asymmetry; the fact that it’s three back and fifteen forward, is just that that’s the difference in their recursion in the buffer field, so to speak, to go back and double check? Why would it be so…

Dr. Keith Rayner: Why is it asymmetrical?

David Boulton: Yeah.

Dr. Keith Rayner: It’s because that’s the direction that you’re moving your eyes in, and that’s the direction your attention is going. But if we do the same experiment with Hebrew readers, which would then…

David Boulton: Be read in reverse.

Dr. Keith Rayner: It reverses, right.

David Boulton: Okay.

Dr. Keith Rayner: This span is also pushed around by the nature of the writing system, so the values that I gave you are for English skilled readers. But if you look at Hebrew skilled readers, then the span is smaller, because Hebrew is more densely packed than English, right?

David Boulton: And less ambiguous.

Dr. Keith Rayner: Yes. So the span for Hebrew readers is about three to the right and about, I think, about eleven to the left. So they have a smaller span in terms of numbers. But in terms of meaning units, it’s probably about the same. Then if you look at Chinese, then the span is something like one character to the left and about three to the right.

David Boulton: Right, because of the density of meaning that’s implicit in Hebrew writing.

Dr. Keith Rayner: Yes, exactly. If you translate it into concepts, then the span is probably fairly constant. But it’s being really driven by the orthography or the characteristics of the writing system.

David Boulton: Right, and how that relates to the construction system, what’s going on inside to deal with it.

Dr. Keith Rayner: Right. So, we also did these experiments with children. I tested kids at the end of first grade, at the end of third grade, and at the end of fifth grade. What we found was that by the third grade, the kids looked pretty much like adults in terms of their perceptual span.

For first graders the span was smaller; their span extended to the right only about eleven characters or so. What was interesting was that it was asymmetric. So, a year’s worth of reading instruction results in this asymmetry, where you’re getting much more information to the right of your fixation than to the left. You might have thought that for beginning readers, because the decoding process is so hard, this asymmetry wouldn’t be there, but it is.

David Boulton: Is there a correspondence to the stretch in the number of letters surrounding the current one that corresponds to the speed of reading, how fast they’re moving through? In other words, I would imagine that the younger student, that it’s taking them more time, they have to be more intentionally focused, kind of point by point, than the…

Dr. Keith Rayner: Oh, sure.

David Boulton: Smoother flow, and that that would affect this envelope your describing.

Code Work and Comprehension:

Dr. Keith Rayner: Exactly. Yes. Incidentally, another thing I did do there was I took third graders and gave them text which was appropriate for them, and then they’ve got sort of a normal looking window. But then if I gave them text that was too hard, then the window shrinks.

David Boulton: Yes.

Dr. Keith Rayner: It’s very small, because they’re spending so much effort in decoding the fixated word that they don’t get more use from the information further away.

David Boulton: Yes. That makes perfect sense.

Dr. Keith Rayner: There are also experiments done with skilled readers, that were done by students in our lab, where they showed that if what you’re looking at is hard to process, then the span of perception shrinks down again, gets small, because, again, you’re devoting all of your resources to figuring out what this word is that you’re looking at.

David Boulton: That indirectly shows the relationship between the code work and comprehension, yes?

Dr. Keith Rayner: Yes.

Summary of Research:

David Boulton: Fascinating. Is there a place to go – kind of like Steven’s Handbook of Experimental Psychology  or the like, a web page or particular resource location that summarizes the findings of your research and your comments about their implications?

Dr. Keith Rayner: Do you know about the Psychological Science in the Public Interest paper?

David Boulton: I read it a while back. I need to go back to that one.

Dr. Keith Rayner: That would be the best statement about the implications. There’s another — do you know what the Journal of Behavioral and Brain Sciences is? I just got the reprints today of a paper that we have in there. This doesn’t have anything about implications in it, this is just the most recent description of our model, where we simulate a reader in terms of where they fixate and how long they fixate. It’s called The E-Z Reader Model. So if you go to the Brain and Behavioral Sciences’ web page, they should have that posted there.

Oral Language Processing Issues:

David Boulton: Good. In regards to the work of Jack Fletcher and Paula Tallal and others, it seems to me that there is a model that’s emerged that says, “Look, there’s a need for the capacity to make phonemic or auditory — oral language processing distinctions — at a level of granularity and at a rate of speed that until you cross a certain threshold it’s insufficient to take off and read.

Dr. Keith Rayner: Right.

David Boulton: And then, reading has to create a kind of simulated player piano playing of that by this code.

Dr. Keith Rayner: Uh-huh.

David Boulton: Then that has to all operate because the code is not self evident, not self contained; it requires external contextual information and rules and all this other stuff to interpret it. Then it’s got to loop through comprehension, which has got to feedback into both of these two processes.

Dr. Keith Rayner: Uh-huh.

David Boulton: So there’s this really multi-stage, complex, timing precarious interrelationship between these concurrent processes.

Dr. Keith Rayner: Yes.

David Boulton: My sense of what Tallal is saying is that if the basic sound distinction processing frequency side isn’t working well, then the core engine on which this other stuff gets built isn’t there. By exercising people’s distinction in sound in time, we build up that layer. And all that does is make one level work so you can go on. It’s not the key to anything, other than it’s necessary, but insufficient to do anything else.

Dr. Keith Rayner: Right.

David Boulton: Ultimately what it looks like to me is that a great many, if not the majority of our children are growing up feeling confused. They’re confused. They’re in an unnatural technological environment that has a number of levels of artificial confusion in it, or an artificial kind of confusion the human brain hasn’t dealt with before. And the children, because of the context we’ve created, feel like there’s something wrong with them because they’re confused. So they’re learning to be ashamed of their minds. We have to do something about that.

Dr. Keith Rayner: Yes.

David Boulton: Very good. Thank you so much, sir.

Dr. Keith Rayner: All right. Take care.