On Blending

Discussions of the alphabetic principle and phonemic awareness on social media make me wonder, has anyone read the original sources for these ideas? Here, for example, is a trenchant quote from Liberman, Shankweiler, and Liberman’s classic article “The alphabetic principle and learning to read”:

“As we see it, the letters of the alphabet do not represent sounds as such, but rather the more remote phonological (and morphophonological) segments those sounds convey. This is not to quibble. For surely it must be somewhat confusing to children to be told that the word bag is spelled with three letters, when their ears tell them plainly that it has but one sound. The confusion is only worse confounded if the teacher insists, against the evidence of what the child hears, that bag can be divided into three sounds, and that these can then be “blended” so as to reform the word. For there is, in fact, no way, with or without the marvels of modern technology, to divide bag into pieces of sound that correspond in any reasonable way to the sounds of the three letters, nor is there any way to synthesize the word by somehow putting the letter sounds together. Though bag does truly consist of three segments–it differs from sag in the first, from big in the second, and from bat in the third–these segments are to be found only in the underlying phonology, not in the surface appearances of the sound….”

A little background: Linguistic theories in phonology and syntax distinguish between the “surface” forms of utterances (what is actually said and heard) and their “underlying” or “deep” forms: abstract structures that are posited to account for various empirical phenomena. Phonologists often call the abstract units “segments” to avoid confusions about the term “phoneme”.

Liberman et al. are saying that phonemes do not have pronunciations because they only exist as linguistic abstractions. Phonemes are in your head, not in your mouth. Teaching children how to pronounce phonemes is thus a serious error (it is what philosophers call a category error). The “alphabetic principle” says that in alphabetic writing systems spoken words are represented at this abstract, phonemic level rather than as they are actually spoken. Figuring out that writing could represent spoken words in this manner was a profound achievement that took thousands of years.

Beginning readers need to accomplish this more quickly. The goal is to have them behave in accord with the alphabetic principle, treating spoken words as if they consist of discrete sound segments. This useful illusion makes it much easier to learn associations between orthography (spelling) and phonology (sound). Adopting this abstraction requires exposure to the alphabet itself: seeing that bag is spelled with three letters encourages the illusion that it consists of three sounds.

The practice of teaching beginning readers the pronunciations of phonemes, as recommended in LETRS and many other SoR resources, is based on a deep misunderstanding of this research. It is teaching children a bogus phonemic principle, that spoken words consist of discrete sounds. As in the UK, American educators could safely skip this entirely.

The sounds that children are taught to produce in sounding-out (aka decoding) a word such as bag serve an important function at the outset of learning to read, not because the word is composed of them but as hints to the actual pronunciation, which is the syllable “bag”. Learning to associate a simple sound with each letter (e.g., a is “a” as in “bat”) is sufficient for this purpose.

Why It Matters

So, you might ask, who cares? So the sounds we’re teaching children aren’t bona fide phonemes. Who could possibly care except a linguist? Call them “letter sounds” if you want: what difference does it make? These sounds are needed in order to begin reading words aloud using phonics rules that specify associations between graphemes and phonemes. The more of these mappings they know, the more words they can read. Agreed?

Here’s why I’ve made a fuss about phonemes. It’s not just that teaching the pronunciations of isolated phonemes lacks a scientific basis, is unnecessary, and wastes valuable class time. Doing so has other pedagogical consequences. As Liberman et al. noted, once such “phonemes” are generated, they must be “blended” to yield a correctly-pronounced word. They put “blending” in scare quotes because individual sounds do not merge in the way the term suggests. The “blending” process was discussed extensively in the National Reading Panel report and gets a lot of attention in SoR resources such as UFLI and Fundations. The idea that words are read aloud by generating and blending a sequence of phonemes is so deeply entrenched it has spawned its own pedagogy.

“Blending” is an artificial procedure that only exists because of how children are being taught to read, letter by letter, (pseudo) phoneme by phoneme. If spoken words actually consisted of discrete sounds, combining them to form a word would be trivial. “Blending” is needed precisely because they do not. The need to “blend” is thus a further consequence of the initial mistake about phonemes. Teaching a lot of phonemes (e.g., 44) and consonant blends (50 or so) plus dozens of phonics rules creates a lot of blending. Is all this instruction necessary? Is it effective? Is it the best we can do?

An Alternative: Pronounceable Parts

Perhaps these questions will assume greater urgency if I point out that beginning readers already know how to “blend phonemes” because they can talk. Consider a child learning to read aloud a word containing the letters ST. Instruction about how to pronounce such “consonant blends” will focus on pronouncing the sounds associated with each letter and then “blending” them. However, pre-reading children know how to say “st” because they use words such as stop, step, and stand. Why are they being laboriously taught something they already know? Only because they are being taught to decode words serially, left to right, grapheme by grapheme, generating and then blending a sequence of phonemes.

Here is an alternative approach that I have been working on for a while called pronounceable parts instruction. The core idea is that instruction can make better use of children’s knowledge of spoken language, minimizing the use of artificial elements such as the “phonemes” that are currently taught.

In the SoR children are taught phoneme-like sounds to use in sounding out words. Using the “correct” pronunciations of phonemes is supposed to facilitate blending them into a word, but what actually happens is that the reader jumps from the sounds to the word.

Skilled readers do not read letter by letter. When the eye fixates on a word, the letters are analyzed in parallel. Every modern model of visual word recognition works this way. Parallel processing allows the pronunciations of letters to influence each other in the brief moment before the word is recognized. The pronunciation of a vowel, for example, can be affected by letters that come before and after it.

Beginning readers need to learn to sound out some simple words letter by letter to get reading off the ground. Letter sounds can be used for this purpose. However, compelling children to become proficient in using this procedure to read many words is unnecessary. Sounding out words functions like training wheels. It’s an interim step, a way to get started on the path to riding a bike. The goal isn’t to become good at riding a bike with training wheels. It’s to be able to ride without them. In reading the goal is being able to see a word and immediately access its meaning, pronunciation, and other knowledge associated with it. That means shedding the training wheels as soon as possible.

After some initial experience reading simple words using the standard letter by letter procedure, children could transition to reading that emphasizes what I am calling “pronounceable parts:” components of words that are pronounced the same as or very close to the way they are in talking. For example, the word it is pronounced “it”. “It” is a pronounceable part because it is a syllable (and word) that people actually say. The letter sounds for i and t, “i” and “t”, are not pronounceable parts of words but they are close enough to start out with. Once the child learns the pronunciation of it, it can be used as a pronounceable part of words such as fit. The pronunciation of fit is then “f” + “it”. Combining “f” and “i” is easier than combining the three individual sounds because there are fewer of them, the three sounds can’t actually be combined, and “it” is an actual syllable. With a little practice the learner can also begin to treat “fit” as a pronounceable part, which is what a skilled reader does. There is no reason to restrict the child to reading words letter by letter, phoneme by phoneme, blending the results, and redoing the process if the initial pronunciation is incorrect, as it frequently will be given the nature of written English. The goal is achieving the ability to read in bigger chunks such as words.

On my pronounceable parts proposal, the child begins to read aloud using a few letter sounds which (like “phonemes”) are artificial. They rapidly progress to using larger, more naturally-pronounced units such as rimes (the it in fit) and syllables. The goal is to read in chunks that are words (such as fit) or parts of longer words (such as fitness, outfit, fitting). It isn’t necessary to learn to read every word this way, only enough to begin reading sentences, which enables learning about and from that level of language. In other words, enough to achieve escape velocity, enabling the reader to begin learning from a broader range of reading, spelling, and writing activities, as I’ve discussed before.

Is there evidence that this approach is effective? I know of none. However, the same is true for methods that are widely used in the SoR. It’s all a big experiment at this point. We do not have the luxury of outcome studies showing how effective specific methods are or how they compare to each other under real-world conditions. The best we can do is look at the support for the assumptions they incorporate. The method I’ve described certainly has a solid rationale, teaching reading by building on children’s knowledge of spoken language and viewing reading development as a progression from smaller units to larger ones. It minimizes instruction about components (such as “phonemes” or letter sounds) that are both artificial and literally meaningless in favor of an emphasis on more natural, meaningful units such as words and morphemes.

Were I still running a research laboratory, I would be dying to get some direct evidence about the efficacy of this pronounceable parts approach. I am not, alas, and so I offer the approach for others to consider. This approach is in keeping with other arguments for making instruction more efficient as well as effective. I’ve obviated the need for extended instruction about phonemic awareness and now blending. Taken with the observation that phonics instruction only has to enable implicit learning from other kinds of experience, these considerations yield a developmental progression that makes appropriate use of explicit instruction and provides a way to transition to reading words the way skilled readers do. This sequence would make more time available for reading itself, and for learning the many other things that go into comprehending texts.