In our last post in this series exploring the question, “What is (un)natural about learning to read and write?,” we looked at a paper from 1976 by Ken and Yetta Goodman that argued that written language is a form of oral language and thus, learned naturally in a literate society through exposure and use in the environment.
In this post, we’ll explore a direct counter to that argument made by Phillip Gough and Michael Hillinger in 1980.
- Gough, P. B., & Hillinger, M. L. (1980). Learning to read: An unnatural act. Bulletin of the Orton Society, 30, 179–196. https://doi.org/10.1007/BF02653717
Gough and Hillinger start off by laying out data that show:
“the statistically average child, normally endowed and normally taught, learns to read only with considerable difficulty. He does not learn to read naturally.”
Like the Goodmans, Gough and Hillinger (let’s call them G&H hereafter) lay out how learning to read progresses, starting with “Paired-associate Learning.” In case you haven’t come across this term before, paired-associate learning, in simplest terms, is like a set of flashcards, in which two distinct items, or stimuli, become associated in memory through repeated exposure. This could be implicit (e.g. hearing the bell means it’s time to eat) or explicit; this is the type of learning most commonly associated with learning “tricky words,” or words in English with highly irregular spellings.
In the Goodmans’ explication of reading development, they didn’t call it paired-associate learning, but they describe a somewhat similar phenomena—that children raised in a literate society start to “read” their environment and gain a functional understanding of written language. G&H agree that the initial steps of learning to read involves this “natural” process:
“The selectional paired-associate technique, the child’s natural strategy, for learning arbitrary associations will work well enough for any and every child, in the beginning.”
But for anyone who has tried to memorize many items knows, we quickly hit a limit in our capacity: “With each new word, the difficulty of finding a unique cue to distinguish it will increase,” and a child “must come to recognize that he has been trying the wrong thing, that his natural strategy will not work” for too many words.
It is here that “the child who has been treating the written language as if it were a code must confront the fact that it is a cipher.”
A Code vs. A Cipher
This is a distinction I hadn’t heard before, and it seems worth unpacking since it represents a key pivot for G&H from the notion that learning to read is “natural.”
In fact, let’s return to our friend Ken Goodman for a second. After the Goodman paper (as discussed in my last post), there’s a short transcription of attendees who heard the paper presented asking Goodman questions, and there’s an interesting discussion around the meaning of the word “decoding” and its relation to language that goes back and forth between a few of the researchers.
At the close of the transcription, Goodman states:
“Now, the problem and the confusion is that people have only treated written language as a code, but oral language is a code, too. . . Matching letters to sounds is a kind of recoding operation, because I still come out with code. That is not decoding.”Goodman, K. S., & Goodman, Y. M. (1976). Learning to Read is Natural. https://eric.ed.gov/?id=ED155621
G&H seem to agree with Goodman’s account of “oral language as a code,” which they furthermore agree is learned naturally via associative learning (i.e. paired-associate learning). But where written language departs, according to them, is that it is not only a code: it is a cipher.
In our everyday use of the words code and cipher, there is little distinction between these words. But in cryptography, there is an important, if subtle, distinction.
I’ll admit I reread G&H’s illustration a few times, and found it slippery. I then looked around online, as one does these days, and found this Khan Academy explanation:
A code is stored as a mapping in a codebook, while ciphers transform individual symbols according to an algorithm.
G&H take the analogy of calling written language a cipher seriously, because according to them, the problem is that in order to learn to read, a child has to perform cryptanalysis and we can’t just hand them a codebook.
The problem is that the systematic relationship of the plaintext (oral language) to the ciphertext (written language) is highly complex. One of the interesting tidbits shared by G&H is that in 60s and 70s, there were attempts to build “reading machines for the blind,” which incorporated “as many as 577 rules” to try to map graphemes to phonemes — and even all those rules still “lead to mispronunciations of many of the most common words in English.”
One impulse, as per the Goodmans, would be to say there’s too many rules to learn to decode such mappings, so decoding should not be taught. But G&H aren’t having that:
“We cannot accept this argument. The argument that English spelling is frequently irregular overlooks the fact that the irregularities are not arbitrary. . .”
“We conclude, then, that if the child is to become a fluent reader, he must learn to decode, more precisely, to decipher. He must internalize the orthographic cipher of English.”
But we can’t simply give kids a codebook of 577 rules and we’re done. Instead, kids need their brains to become equipped with the algorithm of the cipher. They must internalize all of those systematic mappings.
But if this is the case, then the child is confronted with a serious problem, for there is no way we can give him that cipher. . . . In phonics, we try. . . But we believe that the rules of phonics bear only a superficial resemblance to the rules which the fluent reader has internalized.
… the rules of phonics are conscious and explicit — we state them in English — while the rules which the reader uses are unconscious and implicit. . . . the implicit cipher is too fast for phonics.
This is where the terminology orthographic mapping, which G&H weren’t equipped with yet, comes in handy!
The Cipher Must be Internalized
This leads to a conundrum. A child needs to be able to “break the code” but internalizing the cipher is implicit. As G&H state:
Here, then, is the crux of the child’s problem as he enters the second stage of reading acquisition: he must acquire the orthographic cipher, but he cannot get it from his teacher. [bold added]
There’s an interesting parallel here to the Goodmans’ claim that “Instruction does not teach children to read.” The Goodmans’ ALSO state: “Our contention is that we can explain both acquisition and lack of acquisition of literacy in terms of the internalization of the functions of written language by children.”
But G&H diverge substantially in what they mean. While Ken Goodman said in no uncertain terms that he would NOT teach letter-sound correspondences (as he believed it would be detrimental and inhibit natural learning), G&H, instead, are pointing out that the process of breaking the code is so difficult that a whole lot of explicit instruction will be needed to get kids to the point where they can step off on their own.
For G&H this is when things get unnatural. The process of cryptanalysis entails that a child can:
- Recognize that the printed message is an encoded version of a spoken one (‘cryptanalytic intent’)
- Recognize that the ciphertext is composed of letters
- Note each and every letter of every word
- Recognize that written words are composed of a sequence of phonemes
- Recognize that spoken words are, in turn, composed of phonemes
- Decompose a spoken word into its constituent phonemes (‘metaphonological awareness’)
- Sufficient exposure to paired spoken and written messages (plaintext and ciphertext)
According to G&H, because these four factors are unnatural, many children will require explicit teaching of them:
“These things must be explained to him, or he must figure it out for himself. . . And please note this is not a natural thing for the child to do. We confess that we cannot think of another instance in the child’s experience where the child must recognize some visual stimulus as composed of a particular configuration of commutable, permutable, elements. (This is not true of faces, or houses, or animals, or anything else we can think of.)
I think this is an important point to highlight in relation to another recent paper we investigated on language learning, in which author Michael Ramscar challenged the notion that language is learned by its parts, or compositionality, and instead is learned via computing probabilities. I think he’s right, and this echoes the “natural” argument of the Goodmans. But what G&H draw out here is that having to notice each letter and letter sequence in a written word is completely different than the type of learning that we engage with in learning language or paying attention to our environment. Furthermore, phonemic awareness (not yet terminology at the time this was written, apparently, they call it ‘metaphonological awareness’) requires an unnatural recognition and ability to decompose the parts of sounds in a spoken word.
“Whether recognition of individual letters causes difficulty or not, the recognition that each ciphertext word is composed of a sequence of meaningless elements must be hard for the child to achieve. The requirement that he note the same fact about the plaintext, that he recognize that each spoken word is composed of a sequence of meaningless elements, may be even more unnatural.”
So despite the fact that a teacher cannot just hand over a codebook, and phonics may be an artificial vehicle, G&H stress that children will need all the help they can get to be able to internalize and automate the cryptanalysis required to decipher written language.
“. . . we do not believe that phonics teachers the child the rules of the cipher which he must master. But it does provide the child a virtually indispensable tool for collecting data on his own, for discovering what spoken word goes with an unfamiliar written word.
We would note, though, that in our view, phonics is theoretically dispensable. It gives the child artificial rules by which to get the data he needs to learn the real rules.”
The field has come a long way since this 1980 paper, which is to me what makes it all the more remarkable how clear-eyed this account remains, given that G&H were riffing off an analogy to cryptography.
But this analogy gave Gough and Hillinger a firm and testable basis to counter the Goodmans’–and whole language’s–unempirical belief that learning to read could be achieved without a systematic approach to teaching letter-sound correspondences.
So in our exploration thus far in this series, we’ve looked at the argument that learning to read is natural, now countered that it is not. Gough and Hillinger have helped us to see that while the Goodmans may be right about those aspects of the written language that are most similar to spoken language, learning and applying the alphabetic principle to decipher letter-sound sequences and decompose and recode spoken words is no easy feat.
I am still left wondering: Is it that gaining an overlay of a writing system is unnatural? Or is it that it is more abstract, and thus, presents a higher bar to gain fluency with?