[You can also read this on my Substack, if preferred: Seeking Nisaba]
English does, actually, suck, from a systems-design perspective.
You can be someone who loves languages, be fascinated by them, and even think there’s something like a thread of the divine running through language and language acquisition, and still acknowledge that English sucks, on more than one level.
It almost seems that for every ‘feature’ of the language, you can find a case where it becomes a failing.
I’ve been reading more of Diane McGuinness’ work lately, and while she’s influenced my thinking a great deal, and while I expect she forgot more on this topic than I expect to ever learn, I also think she tended to minimise individual differences in the literacy acquisition equation (which seems a ridiculous thing to say given her relative mastery of the statistics of distribution of skill). But she did seem determined (for the right reasons) to make reading failure less of a ‘problem inherent in the child’ and more of a ‘problem with how we teach an overly complex code’. On the whole, I agree with that, but some of her statements lacked nuance (evidenced by the fact that she seemed to contradict them with other statements, elsewhere… indicating that you can never take just one of her forthright statements to be her whole thinking on the matter).
The Code
McGuinness called reading and spelling reversible processes (e.g 2006a, p.252), and written language a reversible code:
“Of course all codes are reversible; you can put something into a code (encoding) and translate it back out again (decoding), otherwise it wouldn’t be a code. The alphabet code is no exception.” (1997, p.72)
“When you teach the code the way it was designed, it is reversible…” (1997, p.209)
The gist of what she’s saying here is right, but the specifics are wrong, from a cryptography stand point. Codes which have ambiguity are not strictly reversible, and it’s also worth noting that genuinely fully reversible codes are only reversible with a key.
McGuinness was acutely, expertly aware of the ambiguity of English orthography (e.g. 2006a, pp.44-7), so it’s strange that she would be so adamant about its reversibility, which I’ll return to.
English spelling is not reversible except on the most superficial level – that when you say “cat”, someone else could capture your utterance with the letters c-a-t, and then someone else, somewhere else in time and space, could read those letters and know exactly what you said. That, in itself, is a mammoth achievement, and one that’s limited to just one species in around 8 million animal species on the planet. It’s not an achievement to be dismissed… it is a functional code because it can be reversed. But it’s not a reversible code, strictly speaking. Because if your scribe wrote “kat”, instead of “cat”, the messiness is unmasked.
When cryptographers talk about a code not being reversible, they’re usually talking about codes that are deliberately designed to be difficult to reverse. That’s not English. English is supposed to allow “cat” to mean cat, whether it’s read or written.
But it fails miserably at being a genuinely reversible code, on two fronts: the design of the code is inherently bad, if reversibility was the goal, and, secondly, the design of the code also doesn’t take account of the ‘hardware’ it runs on – our brains.
Inherently irreversible
The code of the English language is not one-to-one. That is, there isn’t just one way of spelling one speech sound, there are many. For example, you can represent the ɔ: sound (as in ‘or’) as (at least) ‘or’, ‘aw’, ‘au’, ‘augh’, ‘ough’. So it’s one (speech sound)-to-many (possible spellings).
But it’s not just one-to-many, either. Because that ‘ough’ (as in ‘bought’) can also occur in ‘though’ (the rime of which can also be spelt ‘ow’, ‘o_e’, ‘oe’…) and ‘rough’ (the rime of which can also be spelt ‘uff’).
McGuinness called this a “1-to-many and many-to-1 mapping (spelling alternatives and code overlaps)” (1997, p.161). From a database design perspective (a field in which I do probably have more experience than McGuinness had), this is actually understating the complexity. In relational database design we’d actually call this a many-to-many relationship (many speech sounds-to-many spellings, not just two different co-existing 1-to-many and many-to-1 relationships). She knew the complexity of the code exceptionally well (and mapped out hundreds of sound-to-spelling correspondences), but I wonder if not taking this extra step to naming it a many-to-many relationship underlies her insistence that the code is reversible.
Many-to-many codes can still be reversible (many of us manage it – mostly – with English spelling on a daily basis), but they’re far from the best design, if reversibility is the goal. They are inherently ambiguous, which means information loss is inevitable (which, in the case of English, manifests as errors being inevitable). A genuinely reversible code doesn’t have these problems.
With respect to having a key, the English code doesn’t have one, strictly speaking. The closest we can offer to a key is “it could be spelt this way, or this way, or…” Throw “yacht” into the mix and you see why English learners can’t actually expect to get by on the “key” of grapheme-phoneme correspondences. They need something more.
And, of course, “design” is a rather generous description. Design of the code was incidental. English is a mishmash of borrowings from other languages for all kinds of historical and developmental reasons (my favourite, ‘cynicism-validating’ reason being that typesetters were paid per line, so benefitted financially from making words longer), not something that was designed from the ground up.
Looking at it analogously, From a systems design point of view, English makes orthographic demands of us of a similar nature to the semantic demands created by this phrase:
“My child has gone home”
…if the speaker had more than one child and more than one home, and the homes can all house all of those children at different times (a many-to-many relationship). To understand what this phrase means, specifically, more information is required or, alternatively, it requires an external referent to anchor to, which eliminates the unknowns. You can’t possibly interpret that sentence correctly except by chance, or with more information. The same is true for spelling – to get it right, amongst all the possible options, it requires more information (than just the phonology of the word) or a referent to compare it to.
Spelling clarifying language – maybe.
The words “taught”, “taut”, and “tort” all sound the same. That’s arguably a ‘failing’ of the language. But the feature of spelling in these words is that the different spelling (which is enabled by the ‘failing’ of the many-to-many orthography) signposts a different meaning. As such, you can reasonably argue that a deep orthography does have its advantages. But this ‘feature’ fails in sets like “bear” (in the woods), “bear” (my burden), and “bare” (chested); while it can be, there’s no guarantee that phonological and semantic ambiguity will be, disambiguated by spelling – another ‘design’ inconsistency.
The language itself has capacity for standalone sentences of unresolvable ambiguity, the parsing of which is more likely to come down to statistical frequency. A sentence written as “She can’t bear the pain”, from dictation, is perfectly spelt, unless the sentence happens to have actually been about a Chinese glazier unsuccessfully trying to peel off the protective covering from a sheet of glass (“Xi can’t bare the pane” – a far less common meaning, less likely to suggest itself). In that sentence, spelling does disambiguate. i.e. that sentence is only likely to be understood by itself when it’s written; a good reader reading that sentence will not fall prey to the misunderstanding that would likely occur had it only been spoken. The many-to-many orthography has advantages.
“He’s mean” – either verbal or written – can conceivably be describing someone as cruel or average, and spelling offers no help here. In the sentence, “I mean that the mean seems mean”, spelling can do nothing to disambiguate – it’s syntax that does the heavy lifting here.
So, along with the orthography being a many-to-many relationship between sound and spelling (which results in sounds being represented inconsistently, and spellings being pronounced inconsistently), that characteristic can lead to further inconsistency in interpreting meaning, because sometimes spelling disambiguates, and sometimes it doesn’t.
But the biggest problem isn’t even these enormously inconsistent characteristics of the code, it’s the hardware the code has to run on.
Incompatible hardware
Our brains are very good at recognising things, and not very good at recalling things. You generally recognise your own vehicle in a car park quite quickly (though choosing between two similar cars may be a bit more challenging), but if you had to draw the tail-light of your car (which may be so familiar to you that it serves as a feature by which you recognise your car in a car park), or the exact design of the side mirror (something you’ve looked at many times), you’d likely struggle to form a complete, accurate mental image of the proportions and shape of the lenses.
In literacy, this limitation means readers of a certain skill level and experience will always be able to read onomatopoeia, diarrhoea, psoriasis, and ophthalmologist at a glance, but may still struggle to nail the spelling, unless they’ve made a concerted effort to learn them.
That’s particularly problematic for a deep orthography, featuring many-to-many sound-to-spelling relationships, because there are so many valid spelling options that could be correct, but aren’t. Fright could be frite. Freight could be frate. Smoke could be smoak. Many readers will acknowledge these legal spellings but think they’re unlikely. Other readers will not only acknowledge them, but also recognise them because they work with kids and such spellings are actually likely for a subset of kids with particular challenges – those who are weak orthographic processors.
For weak orthographic processors, the disparity between the challenge inherent in a recognition task (reading) and a recall task (spelling), is particularly large, and it’s exacerbated by an orthography in which the number of options of valid (but wrong) spellings is so high.
The ‘something more’ that I mentioned English learners need, beyond knowing grapheme-phoneme correspondences, is strong orthographic processing. The key to the code is of limited use, because the code is ambiguous. Orthographic mapping does the heavy lifting once the basics of how the code works, is known. Some of us seem to be geared for that (our hardware in that respect is relatively good, but still not great – we still read individual words more easily than we spell them), many of us are not.
Given the hardware that the code has to be run on, I would argue that the code is not even fit for purpose (if universal education is the goal) and, unlike Diane McGuinness, I don’t think these problems are all down to how the code is taught (though that is certainly a factor).
Workarounds
In IT terms, the best we’ve been able to do with this arguably faulty code, is come up with workarounds.
Learning spelling patterns can be a very effective workaround – especially for low orthographic processors. Knowing that we use an initial c to represent /k/ before a, o, u, but k before e, i, y, is very useful for these kids.
Pronouncing a word as you’d expect it to be pronounced, based on its spelling, can be an effective phonological ‘hook’ via which orthographic mapping can be strengthened. I still say “fry-end” in my head, as I write “friend”. There has been research on this “spelling voice” approach which supports its effectiveness.
The often misspelt “does” can be explained as a base word and a suffix. The verb “do” is the base word, and we add the suffix “es” to it as a second-person form, just as we do for “boxes” and “munches”. “Does” is orthographically regular in that sense – it’s the pronunciation that’s odd.
Humans cognitively privilege narrative. That is, we attend to and learn things very well if they’re based on stories. A passage based on a cause-and-effect sequence is remembered better than one based on description (though a very emotive description can also create associations, due to the emotion, not the description). So, etymology – the story of how words came to be spelt the way they are – is also effective. Knowing that we used to spell “queen” as “cwen” prior to the 1066 Norman Invasion, when the French influence supplanted that “cw” consonant blend with “qu”, is both memorable and transferrable – no English word I can think of now uses “cw” to represent that consonant blend (unless it’s actually Welsh).
Words with initial /r/ sounds that are related to twisting, are often spelt with “wr”: wrestle, wring, wrench, writhe, wrack, write, wrist… Words with initial /n/ that are related to something having appearing to have been ‘bunched to a point’ are often spelt with “kn”: knob, knee, knuckle, knife, knoll, knap…
But it’s hard work
Such workarounds are useful – vital, even, for some kids. But they create a significant instructional load and a massive cognitive load for the kids. It’s great being able to explain that you drop the silent “e” in “fame” when adding the vowel-initial suffix that means ‘full of’, to get “famous”, but that we can’t do the same for “courage” to make “courageous” because if you do drop the “e”, the /j/ will be ‘hardened’ to a /g/ – couragous. But it’s even better not having to think that through because you just know how it’s spelt.
And that’s the privileged reality for strong orthographic processors – they just know how it’s spelt.
And you can only go so far ‘triangulating’ every word with all these workarounds, in a language of some 200,000 words. For example, you can argue that “please” has to be spelt that way, and not “pleese”, because the ee vowel team represents a ‘long’ vowel sound, while ea can represent long or short (as in ‘eat bread’), and we know that “please” (‘long’ vowel) is related to “pleasure” (‘short’ vowel) and “pleasant” (‘short’ vowel), so you need to use a vowel team that can be used for both pronunciations – hence we ‘have to’ use ‘ea’. And that may even satisfy a kid until you have to explain why the semantically similar “speak” and “speech” are spelt with different vowel teams.
You can encourage kids to think about related words to remember spellings… “autumnal” and “condemnation” may help someone to recall that the respective base words have an otherwise silent “n”. But thinking of “example” won’t help them to spell “exemplified” (and indeed it looks as much like “exempt”).
It’s great to know that the Latin derivatives cede/ceed/cess (meaning go/yield) are quite reliable in English verbs: only three are spelt ‘ceed’ (exceed, proceed, succeed) with the remainder not using that form. But it’s far more nuanced to delve into why the /ee/ sound in ‘proceed’ and ‘procedure’ are spelt differently.
Yes, you can (usually) ‘search and destroy’ such apparent contradictions with deep word study, but your work is cut out for you. Your weak orthographic processor students’ work is piling up even higher than yours, because they don’t already know the spelling; the explanation you’re providing isn’t – to them – just an interesting anecdote or bit of trivia accompanying a spelling they already know; they’re clinging to it as their only hope of remembering how to spell that word.
And that’s why orthographic processing is queen in a deep orthography. The kids who have to be thinking through all this stuff because they have poor memory for spelling, are taxed to the max by any writing task. By contrast, most strong adult orthographic processors couldn’t even articulate the spelling rules, spelling-semantic patterns, morphological consistencies, and etymology of the words in their written vocabularies. Because they don’t have to know them. If they do know these things, it’s more likely because language study is like a hobby for them, and certainly not because their survival in a highly literate society depends upon it.
So, what to do?
Indeed.
Overhauling English to be a phonetically regular language has been proposed more than once, and I’m yet to be fully convinced that it’s not a good idea (or that some version of such an overhaul couldn’t be). With a shallower orthography, yes, we’d lose some of the ‘richness’ of the language, but I’d argue that the folks who can appreciate the richness are those who are have zero problems with literacy skills. From a utilitarian perspective, I think I’d prefer to lose some of the richness to make it more accessible.
My concerns would be with losing the ‘islands of regularity’ that morphology provides. For example, “courage” has an “a” representing the sound that sits closer to a ‘short’ /i/, while “courageous” has an “a” representing a ‘long’ eɪ. These words would be spelt differently under a phonetic system, despite being the same base word, and strongly semantically related.
The schwa would riddle the lexicon as well.
I would assume that such changes would have an impact on vocabulary acquisition via reading, and it could potentially transform what ‘set for variance’ means (i.e. decoding pronunciation should be more accurate, so demand less of a jump for oral recognition of a word, but for a word that is unknown, the link between it and its differently-pronounced base would become more tenuous).
Spelling, too, would have to change to keep step with pronunciation changes, though this needn’t be a bad thing, and nor does a deep orthography insure against divergence between pronunciation and spelling, or changes in spelling.
But, based on the history of past attempts, for now, at least, it seems we’re stuck with the most complicated alphabetic orthography on the planet, and the workarounds are all we have to assist the kids for whom the code poses most difficulties.
Can we fix the problem with instruction?
We can certainly minimise the problem with good instruction, but it’s true that the level of granularity required for low orthographic processors is not required for all, so a judicious approach within a multi-tiered system of supports is wise. McGuinness tended to lean toward the position that dyslexia was more likely dysteachia, and that teaching the code following the statistical regularities within it would ensure success.
The trouble with that view is that we know dyslexics tend to have weaker statistical learning ability; that is, they are not so good at picking up patterns. Explicit instruction around those patterns may help kickstart it, but there is still a weakness there. Kids with learning disorders often also display lower paired associate learning (vital for grapheme-phoneme correspondences). Dyslexics also tend to be weaker in overnight consolidation of taught knowledge. In other words, they are often weaker at learning on multiple fronts, and exactly those fronts that are vital for literacy acquisition. That doesn’t mean we shouldn’t base instruction around statistical regularities – quite the opposite, because we want these kids to have the best chance possible of picking up on those regularities – but I think we have to accept that at the left of the bell curve there will be kids who continue to struggle despite everything being tailored.
It’s also worth noting that the kind of disambiguation examples I gave above – taught, taut, tort (wherein different spellings of the same sounds appear to be an advantage that help hook into specific, different meanings) often don’t actually help dyslexics. Instead, such discoveries present to them as some kind of sick joke, creating more white noise, not less. If you can’t process spelling well, and you can’t anchor meaning to spelling well, having multiple ways of spelling the same-sounding word just adds to the difficulty. Kids can lean on mnemonics in cases like these, but mnemonic devices can often be ambiguous in themselves, and this adds yet another layer to the cognitive load when they have to think through them.
McGuinness’ seeming avoidance of these kinds of “inherent in the child” issues is perhaps exemplified by her comment on a 1990 study (Bruck and Waters) finding 13 sixth graders out of 175 who had normal reading comprehension but very poor spelling:
“As a rule, children tend to score similarly on reading and spelling tests… One should not get carried away theorizing about this unusual population of children… ” (2006a, p.273)
There is a literature on “unexpectedly poor spellers”, with some cited occurrence figures as high as 7%. It seems likely that there’s something at play here that’s worth attending to.
On the other hand, McGuinness makes some pretty compelling arguments for her position that:
” ‘What’s wrong with children who can’t read?’ is a bad question scientifically, logically, and pragmatically, and has been extremely unproductive. ” (2006b, p.426)
She’s highly critical of the science behind the science of reading, which I hope to return to in another post soon. For now, I’ll say that I think many of her criticisms of science and instruction are well-founded, but – as with her strange insistence that the code is reversible despite her expert knowledge of it, and of humans’ incompatible hardware for ambiguous codes – her forthright claims suggest a reductionism that lacks the nuance of which she was clearly capable. I expect that, were she around to answer, she’d say, “well of course that’s not what I meant…”
I also admire her tenacity and motivation in insisting reading difficulties aren’t a ‘problem with the child’ – and I think there’s a good deal of truth in the idea that we’re over-medicalising a difficulty to master something completely unnatural – but there are very clearly kids for whom this is a very tough gig, despite tip-top instruction.
At very least, some understanding would go a long way, for the kids who are working so much harder than their peers to try to master the multi-horned beast of English orthography. Our understanding of this cohort is not great, and a high literate’s ability to relate to their difficulties is almost non-existent, as evidenced by the number of times I’ve had to explain that providing a spelling dictionary is no guarantee of good spelling for some kids – they may not even be able to copy a sequence of letters reliably – or the number of comments educators make about kids needing to ‘try harder’ to improve spelling.
References not hyperlinked to in text:
McGuinness, Diane (1997) Why Our Children Can’t Read: and what we can do about it. Touchstone 1st Edn.
McGuinness, Diane (2006a) Early Reading Instruction: What science really tells us about how to teach reading. MIT Press 1st Edn.
McGuinness, Diane (2006b) Language Development and Learning to Read: The scientific study of how language development affects reading skill. MIT Press 1st Edn.
