Dr. Charles Perfetti – Word Recognition and Comprehension
Dr. Charles Perfetti is Professor of Psychology and Linguistics and the Senior Scientist and Project Director of the Learning Research and Development Center at the University of Pittsburgh. Dr. Perfetti's central research interest is in the cognitive science of language and reading processes. His work on the relationship between word recognition and comprehension is part of the bedrock of modern thought about reading. Additional bio info
Dr. Perfetti is renowned for his knowledge and expertise and is the best kind of scientist; one who is more interested in dialogue and learning than in professing what he knows.
Video: An Overview of Reading and Possible Future Directions
Note: Remember to click on any word on this page to experience the next evolutionary step in technology supported reading.
David Boulton: I’d like to start with a thumbnail of who you are and how it is that you came to this space of learning. How has your personal path provided you the opportunity to learn what you have learned?
Dr. Charles Perfetti: I’m a psycholinguist by training. I got my Ph.D. in Experimental Psychology at the University of Michigan and I was part of a growing interest represented by two institutions on campus about language. I was interested in the basic processes of language, the nature of language, and at that point it didn’t involve reading very much. After I came to the University of Pittsburgh I continued to work on what we call straight psycholinguistic research questions. Then I got interested in reading and I can almost pinpoint the date; I remember I was driving back with a couple of graduate students from a convention in Philadelphia. There were some issues raised in front of these papers about reading and we starting talking about why some children were having trouble in reading, why some people were better at reading than others.
We started talking about different alternatives. Because we were psycholinguistic researchers, my students and I started talking about things that were kind of traditional concerns of psycholinguistics, like syntax. Maybe poor readers were having trouble with syntax; maybe poor readers were having trouble with semantics, and so on.
I remember reasonably distinctively saying, I think it’s the wrong level; think about what reading is compared to language. Reading is about decoding words so that you can then use your language processes. So, before we start doing studies, which was what was really in the air that day, what kind of studies we might do because we promised to address the problem, we ought to start by looking at word reading because that’s where spoken language and literacy are most different. Once you’re able to read words then presumably you’re engaging in language processes and then understanding spoken language and understanding written language are much more similar. Not identical, but at least similar.
I guess that was in the late 1970’s, 1980 or so, maybe a couple years before then… but we took it to heart. We got back to Pittsburgh and we started doing studies on reading. I did a whole lot of them and the force of them was to demonstrate to my satisfaction that if you looked at two kids in elementary school, beyond the point where beginning reading is taught, (so I am talking about third, fourth, fifth, sixth grade), and one is a better comprehender than the other by some assessment, my prediction was that you’re also going to find that the one who is the better comprehender is better at identifying printed words. And that’s what we found. We had lots of studies that demonstrated that. We published some and put them together in a book in 1985 called Reading Ability. That was my contribution to this problem and my argument.
Word Recognition and Comprehension:
Dr. Charles Perfetti: At that point it was only about comprehension. I was just struck by the fact that kids who had trouble understanding what they read, if you looked at them closely, they had trouble reading words. They either took longer or sometimes we asked the teacher to identify kids having problems and she’d say ‘Well, this kid reads words just fine but he can’t understand what he reads.’ We’d say okay and we’d go test him and we’d find out that, well yes, if you weren’t looking really closely, if you had him read a list of words he would read most of the words accurately. But if you measured the time it was taking him or the difficulty he was having, you almost always found that there was some problem in word reading.
David Boulton: There’s still quite a confusion with that today. People say that the student seems to be able to decode words fine so that isn’t the problem. And yet on closer inspection it seems that the efficiency with which they’re decoding the word, recognizing the word, is definitely having an impact on the resources necessary for subsequent comprehension.
Dr. Charles Perfetti: That’s right and that was the heart of that argument back then. I think the argument has become fairly widely accepted as being a correct argument. I think we have a better understanding, although I continue to be disappointed that researchers don’t look hard at word level processing in this new area of research called specific comprehension deficits. That’s a phrase that has been coined by several researchers. The general idea is that they’re observing children who, in fact, are fine at word decoding and have as their only problem comprehension.
I think that hypothesis is correct. That is, the hypothesis that there are children who have as their main problem something else in the language system than reading words. But I am still a little disappointed when I look at how people look at word identification. So, I think what has happened is people took my argument seriously enough, and the argument of others, that they assess decoding now before they do a study on some other variable. But they still tend not to measure efficiency and speed.
David Boulton: Even the neuroscience hasn’t yet gotten to the point where they can isolate the timing of processes in the assembly processing between the code and the word recognition processes.
Reciprocal Relationship – Comprehension to Decoding:
Dr. Charles Perfetti: Well yeah, I think that’s right. At the same time, I want to acknowledge another angle to this, which is that although there are now training studies you can point to that show that if you increase decoding word efficiency you get comprehension gains, I think one has to acknowledge that’s there’s likely to be a reciprocal relationship here. As you get so that you can comprehend what you read you are going to strengthen your word identification processes.
I mentioned the training studies because those do break through what are otherwise only correlations. That is, generally speaking, a correlation between word decoding and comprehension. You break through that in a training study by showing that if you improve someone’s word decoding you improve the comprehension. There are now a few studies that have shown that that can happen, although I am also impressed by the fact that it’s apparently not so easy to do. Some of the training studies do not work so well; it has to be done right and so on.
But I think that underlying this is an important idea that the reading system, which does depend very much on the spoken language system, its components can develop in tandem and can mutually reinforce each other. So, as you comprehend better I actually think you do get the experience of successful comprehension in strengthening word identification processes; at least on the words that you are reading while you are comprehending what you’re reading. Although I originally argued very strongly for this causal relation, and I still think that’s correct, I do think we need to understand that it’s a little unclear.
You don’t want to be in the position of saying that we have to reach some level of decoding skill before we can expect them to comprehend because that’s certainly not true. Comprehension is something that develops very quickly as children begin to read and I think the behaviors of reading, both reading out loud and silently, are calling on all parts of the system and can mutually support and can strengthen each other. That helps and partly explains why you tend to see a high correlation at some point. Basically, to put it crudely, the more you’ve read the more practice you have had at identifying words and the more practice you’ve had at comprehension. So, you see a positive correlation.
David Boulton: Right, which makes perfect sense in terms of comprehension making it easier for word recognition to ‘catch’ the approximations coming out of decoding-assembly processing such that they feed into one another and become a kind of iterative heuristic that becomes self-optimizing.
A Language Simulation System:
David Boulton: The way that you decoupled the word recognition process from comprehension in your thought model really stuck me because the way we describe it is somewhat similar: reading is a code instructed and informed language simulation system that’s feeding into downstream comprehension processes.And if that first part is working right then the second part, like you said, is very similar to comprehending spoken language.
Dr. Charles Perfetti: Yes, I think that’s a fundamentally correct way to understand it. If you don’t understand it that way I think you’re going to maybe go down some garden path. Getting that first phase to a high level so that you’re able to use the language system – if you’re struggling with that mapping process then you’re not able to engage the language system that you have and that you would otherwise use perfectly well with the spoken language.
Components of Word Recognition:
Dr. Charles Perfetti: That’s why it’s so critical to have the word identification process developed to a very high level. A lot of that, especially in English, and I‘m not sure everyone would see it this way, is rather word specific knowledge. That is, you’re really talking about two things that develop. About developing a strong generalized decoding routine, mapping letters to phonological representation, but your also talking about strengthening specific associations between written letter patterns and a word. I think you’ve got to have both of these to read effectively in an orthography like English, which is highly irregular. So, that’s where the experience comes in. You’ve got to have quite a bit of experience to have a chance to build up associations between a printed form for an irregular word and the word that it’s representing.
David Boulton: Right, and the difficulty for a lot of children entering this is that their earlier language learning environments didn’t sufficiently exercise their brain’s ability to make the kind of sound distinctions and/or make them at the frequency necessary for this to work for them. When they encounter the kind of challenges associated with building this part of the brain so it can generate the simulated word recognitions from this code, then they’re ill prepared to do it.
Cognitive Implications of Emotional Aversions:
David Boulton: To make matters worse, if they have too much difficulty with it then they start to have an emotional aversion competing with the cognitive processes necessary for doing it.
Dr. Charles Perfetti: Yeah, none of us handle failure very well. If you’re put in a situation where you’re not performing a task there are going to be negative emotions and there’s going to be avoidance. There’s going to be a spiral of negativity that’s counter productive and that’s a big part of it.
What’s Different About Reading:
Dr. Charles Perfetti: Back to the spoken language part, I think you’re right; the variability in a language environment is really critical. If you take the situation as a whole, then what I think is important about children’s language environments is that they are generally geared toward meaning and communication. In terms of language structures, they’re geared toward meaning rather than to form. And in order to read, suddenly form is important, the form of the letters, the form of the phonological representations. It’s one of the things that makes it really tough.
David Boulton: Wouldn’t you also say, particularly with the English code, that there’s really nothing else a child ever does before that point that is like having to assemble a virtual language experience from a technology?
Dr. Charles Perfetti: Well, … I might have missed your point about what you mean by technology, but I’ll put it in my own words and you can decide whether it connects with what you’re saying.
If not unique, I do think it’s one of a small class of experiences in which there is a radical shift away from being in a tremendously easy, supportive environment. What I mean by supportive is that your attempts to communicate are almost always supported. If you say something, someone understands you. If they say something, you understand them. One of the reasons is that in childhood environments all of these communications take place in contexts that over-determine what’s being communicated. They tend to be conversations about the here and the now; they tend to be about objects from the environment they’re happening in.
David Boulton: And there’s a sensory somatic proprioceptive loop going on.
Dr. Charles Perfetti: Yeah, and the child has not been forced to draw attention to form. With reading, suddenly you’ve got to tend to form for the first time and what counts is that this letter is here rather than that letter. The letter itself has no meaning. It associates with a spoken language unit that also has no meaning, a phoneme.
So, this attention to form kind of pulls the rug out from under you, which is why I think some people want to argue that you ought to make it very meaningful from the beginning so the child can use the kind of support that he’s used to in spoken language.
David Boulton: Which is perfectly legitimate as long as we don’t fail to recognize that we’re talking about something that is different than the way our meaning- oriented natures work in nature.
Dr. Charles Perfetti: Well yes, and that there is a specific learning question here. The question is the child has to learn something, so what does he have to learn? He has to learn the written forms of his language and how they map onto spoken form.
I used the word forms twice there. He doesn’t have to learn new meanings, at least not generally; certainly when you do get skilled at reading you learn new meanings. But the basic thing you have to learn in reading is form mapping. And you try to arrange things so that meaning can help support you in this effort. But at the heart of it you’ve got to learn the forms.
Brain Structures - Connecting Visual Form Perceptions with Language:
David Boulton: Right, and this has to result in a module in the brain that’s processing and translating this form into sound representations faster than you can think about it.
Dr. Charles Perfetti: Yes, that’s right. The brain will develop neuro-structures that generally are adapted from other uses. There’s no reason to think the brain has any module in it in the beginning for reading. Written language is too recent a human development to have allowed brain evolution to accommodate it. So, I think this is another aspect of the difficulty.
The brain has resources it can use for vision, in particular, and it has to recruit these resources for reading. Those visual resources actually are not maximally tuned for distinguishing the difference between an E and a C, for example. There are levels of the visual cortex that can perform that task but it’s quite different from being able to tell one face from another or one building from another, and so on. The high spatial frequencies have to be resolved.
Now, once that part is done and once you connect these visual form perceptions with your language, that part, speaking loosely, could be considered a module. The connections are pretty constrained from visual areas to temporal and frontal areas that together seem to support rapid word identification and so some of those areas are shared with spoken language as well. So, at that point once you’ve got the visual forms, and not just the forms but mapping them to the spoken language forms, then you can talk about a system acquiring some modularity; behaving as if it’s a highly skilled system using sub-brain networks that are in some sense specialized for reading. But in the beginning you have to recruit other resources and apparently that can be a problem.
David Boulton: Those resources would also perform, following along your trajectory here, the differentiation inside of word sounds in a way that’s otherwise artificial and that is unique to the kind of distinctions in word sounds necessary to map to the code. So, it’s not the same kind of attention to sound differences that are necessary to participate in oral conversation.
Dr. Charles Perfetti: Exactly. Right, and so again, it’s attention to form, a much more analytic form, little pieces of form that in themselves have no meaning.
David Boulton: So, in a natural setting we’re more attuned to seeing the difference in larger wholes rather than in making differences between such small particulars.
Dr. Charles Perfetti: Right. I think that’s a big part of it.
David Boulton: That’s why overall what were trying to get a handle on is whether we can describe this lower part of the process as if its job is to simulate a stream of language inputs to comprehension and fabricate that simulation according to the way the brain has learned to process the code through the visual and auditory modalities, which is providing it with information and instruction.
Dr. Charles Perfetti: I like that. I think that’s a good way to describe it. I find that very compatible with the way I think about it. Those are your words, not mine, but I think that’s a good way of thinking about it.
David Boulton: There are two things that plug in here that I want to explore with you.
One has to do with assembly processing time. It’s clear in talking to the phonological side of neuroscience that fuzzy representations in the phonemic, phonological dimensions require more processing time to disambiguate and cause a processing stutter – again purely on the auditory processing side.
To the extent that that’s true, then it seems equally true that the time it takes to disambiguate the code is also causing a processing stutter. This is one of the problems I have with terms like ‘alphabetic principal’ or ‘breaking the code’ because they over-simplify what we’d other wise call, more in the computer world so to speak, ‘disambiguation’, which is to take this stream of letters, some of who’s sound values depend on words that haven’t been read yet, buffer them up and construct these approximate word sounds from these fuzzy letter variables.
Dr. Charles Perfetti: Right.
David Boulton: And that the more time it takes to do that, just at that level in this module we’ve been describing as the language simulator, then the more that module is not delivering the language stream in time for comprehension.
Dr. Charles Perfetti: Yes, I think that’s right. Alan Lesgold and I developed an idea that I put in my 1985 book about this. We called it code asynchrony; the idea being that orthographic phonological and semantic codes that had high levels of skill come out all at once. And in low levels of skill there can be an asynchrony in a sense that you’re getting some of the graphemes and some of the phonemes but you’re not getting the whole thing yet and things get all out of phase. Instead of mutually strengthening each other so that at the end of the decoding episode you have a stronger word representation that’s accessible by orthography, you’ve got bits and pieces and only partial success.
Now what you’re adding to that idea. I think specifically what you are saying is if the phonological space is too fuzzy because the letter hasn’t made phonological differentiations that turn out to be relevant for English vocabulary then that’s going to be an additional problem. There’s not going to be a differentiated phonological coding that comes out of any given word reading event. Then the question is how that develops.
Some of the pre-literacy research on children’s development of spoken language, I think, is suggesting that fuzzy phonological representations are normal and characteristic of early language development and that they become less fuzzy, more differentiated and more articulate only in response to increasing demands in the linguistic environment, which usually amounts to having to learn a new word or having to distinguish a new word from one that you already have and that can force you to make new phonological distinctions.
I think the fuzziness is normal and I think what can happen with reading when things are working well is that getting good feedback, either internally generated or externally, on a decoding attempt can have the same effect, that is forcing phonological differentiation. So, you can say an approximation. It doesn’t map onto anything that you know and so you either get feedback that it’s actually this word rather than that word or that the word that you’re trying to decode is novel and has its form and that produces new phonological representation. That, I think, is an interesting possibility for understanding this in general.
David Boulton: Right. So, they’re feeding into each other. I think at one level we’re a fuzzy processor in a myriad of ways.
Dr. Charles Perfetti: I think you’re speaking formally, a formal idea of fuzziness, which is probabilistic category membership. That’s the formal sense.
David Boulton: Yes, but where I’m going with this is that to the extent that the time it takes to get from elemental recognition through disambiguation, through to an approximated word to move to recognition with… if that assembly takes too long then there’s a stutter that radiates through everything.
Dr. Charles Perfetti: Right, because it turns out you’re actually not assembling a unit that you can then use as a representation. Things are too disconnected.
David Boulton: Yes.
Dr. Charles Perfetti: I talked about this in terms of the theory of what it is that children learn to represent when they learn to read, something I called specificity representation. So that before acquiring specificity as a characteristic representation, I would put in these formal terms: it has variables instead of constants. Instead of always having this sound, it has sort of something like this sound or something like that sound.
David Boulton: It’s very much like a quantum wave collapsing to a particle in the context of the process.
Dr. Charles Perfetti: Yeah, maybe. I haven’t thought about that. That’s an interesting way. Okay.
David Boulton: I’m with you and I think your work has separated out these processes so that we can get our minds around them as logically distinct units even though we realize that these are all concurring processes happening in real time.
There’s something special and unique about learning a ‘code’, which is an artifact, a technology, something that we, not nature, created. If you look at the history of how the particular English writing system came to be in the condition that it’s in, it’s pretty clear that it was a series of accidents that have resulted in a code system with a number of layers of complex ambiguity. It requires a lot of processing power to resolve that ambiguity and the brain has precious little time to do it before the whole assembly process is going to stutter up the flow to comprehension processes.
Dr. Charles Perfetti: Well, yes. I think that is well put. I agree.
Reading in Different Orthographies:
Dr. Charles Perfetti: You know, a lot of the work I have done in the last few years, this gets back to being a little more of a biographical story as I mentioned stuff that got me interested, but I also became interested over ten years ago in the fact that so much of what we’ve learned about reading has been in English and I started directing my attention to what it’s like to learn Chinese, to read in Chinese compared with reading English. I think there are some lessons there. Particularly, this assembly process that you’re talking about is not present when reading Chinese.
You’ve got to learn associations between characters and their meanings and between characters and their spoken language form. But despite the fact, and this I think is an important discovery, despite the fact that when you read Chinese you don’t assemble in the way you do when you assemble an alphabetic writing system. You, nevertheless, find that the reader activates phonological representations. So, Chinese readers, when they read characters, activate the sounds of the Chinese words that associate with the characters just as we do in English or Dutch and what have you.
I think that’s a very universal perspective on the code. The nature of the code can vary, but that code, if you want to call Chinese a code, it’s a really simple kind of code; it’s not complex like the alphabetic code is. But after that coding process does its work the outcome is rather similar. You get connection to the spoken language, even if you get a system like Chinese, which allows you, in principle, to go directly to the meaning.
David Boulton: Right. It still maps to meaning through phonological representation.
Dr. Charles Perfetti: At the same time, I think in Chinese what seems to be the case is that you get simultaneous mappings to meaning and phonology, but you don’t just eliminate the phonology. If you draw a diagram to phonology the link looks like it ought to be a vestige or something. It’s not logically necessary; you could imagine doing without.
But the fact is no matter what your language, when people read they have learned to use their language. Reading goes through the language system no matter how it’s written.
David Boulton: Right. I’ve always been interested this. I once asked Reid Lyon, in terms of the international comparison of reading across the different languages, whether anybody has actually built a matrix that compares performances in relation to the distribution of challenges unique to each of the languages. Because comparing English and other languages without in some way drawing attention to the fact that the processing challenges for languages are distributed differently, seems to be missing something vital.
Dr. Charles Perfetti: Yeah, that’s a good point. Did Reid come up with anything?
David Boulton: No, I think he said they haven’t looked that close at it. Which tends to make me feel like the international comparison of reading doesn’t have a lot of ground. I think it’s fascinating, like the work you’re doing, I think we should study this to a much greater degree than we have. But I don’t take any comfort from the fact that other countries are reading better or worse because we are comparing apples and oranges.
Dr. Charles Perfetti: Well, right, but the first part of what you’re saying is we’ll understand that better if we really understand the challenges that each system is providing, and I think that’s true. There’s quite a bit of work now and the work is not, from my point of view, comparing which country has kids making higher achievement but rather exactly what you said, how are these, in staying in the family of alphabetic orthography, how are these orthographic variations playing out in the challenges that the child is facing in learning to read? There are some very fascinating differences that have been observed. People are beginning to try to make it rather systematic, basically in terms of how easy it is to assemble the phonological representation.
David Boulton: For example, something like Spanish. One of the things that makes this challenge so unique is that throughout the emergence of the alphabet and its movement through Greece and Rome anyway, for the most part it was phonetic. Look, say, blend it together and you’re reading. No internal assembly required. But the past 500 years of English, in particular, and probably French, German and other northern European languages that fused with the Latin sound systems and Roman alphabet in various ways… they represent a different kind of complexity in the internal assembly required by this simulation module to generate this inner experience.
Dr. Charles Perfetti: I think that’s right and so that’s created this kind of variability through alphabetic systems. Someday you might enjoy, and maybe you know this story already, but to take a look at Korean, what alphabet was invented. Korean is written in alphabet, it used to be written in Chinese characters that didn’t fit the language well and they invented the perfect alphabet in the 1500’s. They’ve made some changes since then and actually the direction has made it somewhat less perfect, but it was perfection.
David Boulton: I will look into that. Do you have a short on how effective that’s been in terms of the efficiency of their reading?
Dr. Charles Perfetti: Don’t have the comparisons.
David Boulton: Okay. I know most of the languages of the world have gone through adjustment processes to try to reconcile or better adapt the kind of dialectical pronunciational shifts from the spelling patterns so as to bring them back together a little bit. And that English, of course, has rejected every attempt to do that in the past 400 years, except for Noah Webster’s few exceptions.
Dr. Charles Perfetti: Well, I think it hasn’t always been successful where it’s been tried. They did it in the Netherlands and had relatively small changes but it’s caused a new set of problems, actually. So, trying to make things a little bit more consistent actually winds up causing other problems.
First of all, writing is very conservative and as far as popular perception is concerned people don’t really want it changed, so it’s very resistant. But the interesting thing is that it’s only when you start to make the changes that you realize there’s a cost for the changes as well as what you’re trading off for. You’re trading off something for something else.
David Boulton: Right. Charles Hockett, a linguist you’ve probably encountered before, once said, “People are more likely to change their religion than change their writing system”. It certainly seems true.
I’m really enjoying our conversation and we’ve just scratched the surface, obviously. I am glad we’ve talked about the time precariousness of the processing happening in this language simulation module in terms of the assembly and disambiguating the letters and sounds and so forth.
The Downward Spiral of Shame:
David Boulton: The other dimension we haven’t talked about is what happens when instead of being on automatic pilot, with a kind of cognitive entrainment to the task of translating letters into sounds and priming this pump and moving through there, the child or the adult who is struggling with this triggers into shame.
The common consequence for people that have trouble learning to read is self-blame. They think it’s a reflection of something wrong with themselves. The more that they experience shame, the easier it is for it to trigger (if you look at it from the affect science point of view). And everything I can get so far, although there isn’t any neuroscience that goes this deep, is that the moment one of these negative emotions, these affects, trigger, it switches attention away from what was being processed and it undermines the cognitive processes necessary for the reading.
Dr. Charles Perfetti: Yeah, I think that’s right.
David Boulton: So, we’ve got a complex code assembly process dependent on a robust language processing backbone, so to speak, and in a precarious emotional threshold before shame-out can cause a fundamental disruption to the cognitive processing.
Dr. Charles Perfetti: Yes, that sounds like a right analysis. It’s devastating. It’s a double whammy. The trouble with decoding is one whammy and because you have trouble you have a response that essentially makes it harder to learn what you need to learn.
David Boulton: Yeah. More on this, I hope, when we get together face to face.
Dr. Charles Perfetti: Okay.
David Boulton: Thank you so much, sir, I really appreciate your time and all the great work you’ve done.
Dr. Charles Perfetti: You’re very welcome. Nice talking to you.
Special thanks to volunteer Lauren Groover for transcribing this interview.