Henry Davis told me on Friday night about something very interesting. In Lillooet, there is a distinction between velars and uvulars. This distinction is very robust; Henry says it carries a “high functional load”. Speakers consequently reliably distinguish and identify them in speech.
When writing, on the other hand, Henry claims that some consistently get this robust distinction right, as I think we’d all expect: they use velar symbols for velars and uvular symbols for uvulars. But others tend to mix them up: uvular symbols are, unpredictably, sometimes used to represent velars and vice-versa. So Henry asks me: why should that be?
Honestly, I don’t know. But here is the guess I ventured.
Keep in mind that I’ve only read a little bit about the study of writing systems – just enough to know that “grammatology” is the term for it, coined by I. J. Gelb, promoted by Peter T. Daniels, and later co-opted for other purposes by Jacques Derrida (referring “with reservations” to Gelb). To be precise, the little I’ve read on the topic has been by Daniels: his contribution to The Handbook of Linguistics and bits of The World’s Writing Systems (co-edited by Daniels and William Bright). OK? That’s it.
The issue with Henry’s Lillooet example, as I see it, is this: what elements and/or levels of linguistic information do people use to encode a writing system? Specifically, what elements and/or levels of phonological information are used? At least in the case of alphabetic writing systems, I had always assumed that, roughly, the (shallow) underlying level and the surface level of the phonology are both accessible, and that different orthographies are more-or-less surface-y or more-or-less underlying-y. Y. There are plenty of examples of each; here are just a few I can think of:
- Orthography may encode surface-level information.
- In Turkish, voiced obstruents are devoiced syllable-finally.
This neutralization is reflected in the orthography:
kitap [p] ‘book’,
kitabım [b] ‘my book’ - In Spanish words, nasals assimilate to following labials.
This neutralization is reflected in the orthography:
inútil [n] ‘useless’,
imposible [m] ‘impossible’
- In Turkish, voiced obstruents are devoiced syllable-finally.
- Orthography may encode underlying-level information.
- In German, voiced obstruents are also devoiced syllable-finally.
This neutralization is not reflected in the orthography:
Rad [t] ‘wheel’,
Rade [d] ‘wheels’ - In Spanish, voiced stops are spirantized post-vocalically.
This neutralization is not reflected in the orthography:
dentro [d] ‘into’,
adentro [ð] ‘inside’
- In German, voiced obstruents are also devoiced syllable-finally.
(I say “surface” and “underlying” instead of what I think is the more usual “phonetic” and “phonemic” in this context, because (as I claim below) there is a crucial distinction between the surface phonological level and the “true phonetic” – that is, non-phonological – level.)
It’s been said (though I don’t know where or by whom, so maybe I’ve just heard it said) that more underlying-y orthographies are in general easier because the underlying level of information is, in the relevant sense, more accessible to people than the surface level. This might explain something that I think is probably relatively common problem for speakers/writers of languages with more underlying-y orthographies. Martina Wiltschko, a speaker of Austrian German, reports that she had a hard time learning to correctly transcribe devoiced syllable-final voiced stops. I had a similar problem learning to correctly transcribe spirantized post-vocalic voiced stops in Spanish. But, this is kind of a red herring; we need to balance it with (equally anecdotal, if necessary) evidence that speakers/writers of languages with more surface-y orthographies have a hard time learning to write their language using a more underlying-y orthography. (Maybe such research exists.)
But getting back to the point: note that neither underlying nor surface phonological information helps in the Lillooet case. Velars and uvulars are distinct beyond a shadow of a doubt, both underlyingly and on the surface:
|
Orthographic key
|
(Henry adds that there are no contexts in which velars and uvulars don’t contrast, and that both velars and uvulars are “automatically rounded before (and usually after) round vowels”.)
So, armed with this robust underlying and surface phonological information about this highly functional distinction, why do some Lillooets get it wrong only when writing it down? It seems like it should be easy to make this orthographic distinction, but it isn’t.
This suggests to me a completely different hypothesis about what kind of phonological information is accessible to people for encoding an orthography: none. Suppose instead that the only information accessible is the actual output of the vocal apparatus as felt in the mouth and as heard by the ear (and as interpreted by the brain, of course, but crucially not accessing any kind of phonological information). The distinction between velars and uvulars is highly subtle, both articulatorily and acoustically; if we all only have access to “truly phonetic” information (whatever that means here), then some folks are just going to be better at noticing these distinctions than others.
For what it’s worth, this hypothesis is compatible with the (anecdotal) problems I noted that some of us have transcribing our more orthographically underlying-y languages more narrowly. The phonetic distinctions in the German and Spanish cases I cited are each nearly-but-perhaps-not-quite as subtle as the velar/uvular one in Lillooet, but we know that our level of phonetic awareness is influenced by our parochial linguistic experience and can be reinforced by our orthographic knowledge, so there’s no need to invoke direct access to actual underlying phonological information.
(I suppose that accessibility of phonological information could also differ from person to person, but I somehow have a harder time imagining this one. Then again, I’m making this whole thing up, and boy will my face be red if I hear this has all been said (and debunked?) before. But I would like to know anyway – it’s interesting. What does the research behind the phonics industry say about this?
I’ve had similar experiences with Yolngu Matha and Bunuba writers. They have underlining for retroflection, speakers reliably differentiate the two series, but they always leave off the underlining. I assumed it was because speakers knew which series it was and didn’t feel the need to encode it. There isn’t a strong prescriptive tradition for orthography in either place. Another consistent set of errors in Yolngu Matha is the use of voiced or voiceless consonants variably after the glottal stop. There’s no voicing distinction in that environment, and speakers vary in their preferences for the voiced or voiceless series. fwiw, speakers seem not to be bothered by the ambiguity when read things back.
When (and how) do Lillooet speakers typically learn to write their language? Are they already (at least somewhat) literate in English before they start learning the Lillooet orthography?
If so, the answer could have to do with a strong prior bias to “spell things out” (meaning spell them out with English rules, modified slightly by the knowledge that fancy Lillooet spelling is full of
When it comes to intuitions about how sounds are made, I think there’s some evidence that the further from the lips you go, the harder it gets.
This might be because of less visual and tactile experience, or because of fewer intrinsic proprioceptive cues, I don’t know.
Then again, it might not even be true. But consider, for example, thinking about consonant place between the dental and palatal positions, vs. thinking about front vs. back velars. Or the murkiest thing of all, tongue root position. Does anyone have any instincts about the fact that the tongue root is advanced in (most) English voiced stops? And what do you think, is [n] like [d] or like [t] in that respect?
I certainly can’t “feel” or “sense” the answer to that question.
Intuitions about how different two sounds are, either articulatorily or auditorily, depend very heavily on the speaker’s and listener’s linguistic experience. As Janet Werker has demonstrated for the velar:uvular contrast specifically, and many others have demonstrated for other contrasts, contrasts are easy to detect when they can be mapped onto native contrasts and hard to detect when they can’t. Linda Polka has shown that English speakers treat velar:uvular contrasts as a good “k” vs a not so good one, but not as a “k” vs a contrasting category.
This evidence leads me to be very hestitant to apply my intuitions about how different two foreign sounds are to explaining why native speakers of the language that distinguishes them fail to write them reliably with different symbols.
Instead, I think that failure stems either from imperfect learning of the writing system, or perhaps more likely from indifference to the distinction when the choice can be inferred from context.