On beyond the (International Phonetic) Alphabet

« previous post | next post »

The International Phonetic Alphabet is a useful invention, which everyone interested in speech sounds should learn. But it's much less useful for actually doing phonetics than you might think. Whenever this comes up in discussion, I'm reminded of the Dr. Seuss classic On Beyond Zebra:

In the places I go there are things that I see
That I never could spell if I stopped with the Z.
I'm telling you this 'cause you're one of my friends.
My alphabet starts where your alphabet ends!

Here's a simple example — listen, and contemplate:


And again:


Just a few more:


Those are the seven renditions of sentence SX437, "They used an aggressive policeman to flag thoughtless motorists", from the TIMIT Acoustic-Phonetic Continuous Speech Corpus, created at SRI, TI, MIT and NIST between 1987 and 1990.

And among the many things happening in these examples, the point of interest to us today is the pronunciation of the final /sts/ sequence in motorists, which varies from a full s + t-closure + s sequence in (1):


…through successively weaker approximations to the /t/ gesture in (2), (3), and (4):




…through successively shorter uninterrupted /s/ regions in (5), (6), and (7):




In case you're tempted to think that the problem here is sloppy speaking by ordinary unprofessional talkers, here's an example from a consummate professional voice. Terry Gross is the host of the Fresh Air radio interview program, and this is how she starts her 3/16/2015 interview with Daniel Torday:

And this is not an aberration. In words like motorists, artists, activists, exists, consists, , etc., both professional speakers and ordinary Americans exhibit a similar palette of outcomes for the word-final /sts/. There might be a well-defined silent stop region separating two [s] fricatives, or there might be a region of variably-lowered amplitude in the middle of a fricative, or there might a variable-length fricative without any medial amplitude modulation at all. (And it's worth noting that in fluent speech, the full [sts] variant is quite rare. There are just 4 full [sts] variants out of 61 word-final /sts/ examples in TIMIT, for example (6.6%). In fluent reading of continuous passages the proportion is generally to be lower than that, and it's lower still in spontaneous speech. Overall, the uninterrupted [ss] — or [s]? — version seems to be the commonest case.)

So how should we transcribe these various variants of /sts/?

As far as I know, the IPA doesn't have any symbol for a lowered-amplitude region in the middle of a coronal fricative. We could go "on beyond (IPA) zebra" and invent one — maybe an otherwise unused letter like sigma σ, or maybe taking over one of Dr. Seuss's inventions, say spazz:

But that would probably be a mistake, at least if we interpret

[sts] ↔ [sσs] ↔ [ss] ↔ [s]

as necessarily representing qualitative symbolic distinctions rather than a continuum of degrees of lenition.

For one thing, as we continue on beyond phonetic zebra, we'll find that other variations in strength, timing and coordination of articulatory gestures seem to motivate literally thousands of additional "symbols":

As Dr. Seuss put it

Oh, the things you can find
If you don't stay behind!

But by inventing and deploying those symbols, we'd be giving premature answers to the really interesting and important questions.

My own guess is that the /sts/ variation discussed above, like most forms of allophonic variation, is not symbolically mediated, and therefore should not be treated by inventing new phonetic symbols (or adapting old ones). Rather, it's part of the process of phonetic interpretation, whereby symbolic (i.e. digital) phonological representations are related to (continuous, analog) patterns of articulation and sound.

It would be a mistake to think of such variation as the result of universal physiological and physical processes: though the effects are generally in some sense natural, there remain considerable differences across languages, language varieties, and speaking styles. And of course the results tend to become "lexicalized" and/or "phonologized" over time — this is one of the key drivers of linguistic change.

For more on this, see "Towards Progress in Theories of Language Sound Structure", in Brentari & Lee, Eds., Shaping Phonology, 2018.

[And yes, Dr. Seuss has been steering kids wrong all these years —  duality of patterning ensures that a properly-designed alphabetic orthography can spell any real or possible word in a given language. But extending the alphabetic principle to phonetic interpretation is a mistake, in my opinion.]


  1. Q. Pheevr said,

    April 19, 2018 @ 8:26 am

    The IPA does provide some diacritics that might be useful here, though only if the transcriber has a sense of what’s going on articulatorily. For example, I think one of my possible pronunciations of motorists ends with something like [st̞s] or [ss̝s], with a slight increase in constriction in the middle that doesn’t quite result in closure. But someone working only with acoustic data might not want to use articulatory diacritics, and in any case, it’s still a discrete symbolic representation of an analogue continuum, as you said.

    [(myl) All true — except that I suspect this is the wrong articulatory story, or at least an imprecise one. The [s] constriction is made with the sides of the tongue blade against the post-alveolar palate — at least mine is — and a [t] constriction, in general and in this case, involves the tongue tip. So what's happening (I think) is that there's a more-or-less lenited waggle of the tongue tip towards (or against) the alveolar ridge, while the tongue blade remains in position for [s]. Said waggle can be longer or shorter as well as achieving a more or less complete closure.

    I guess you could treat this as a version of an alveolar tap [ɾ], but …]

  2. Q. Pheevr said,

    April 19, 2018 @ 10:15 am

    My /s/ is apical—but of course this kind of variation is exactly why we should be wary of choosing articulatory symbols based on acoustic data. Sitting here saying “motorists” to myself repeatedly, I think that what I’ve got is a brief raising of the middle of the tongue tip, flattening the constriction to produce something like the slit fricative [t̞] that occurs as a variant of /t/ in some Irish and Irish-influenced varieties of English (in, e.g., Liverpool or Cape Breton).

  3. Mitch said,

    April 19, 2018 @ 10:32 am

    The phenomenon also exists/should be checked for other inter-fricative voiceless stops, eg 'risks', 'wasps'

    [(myl) Indeed. I had already verified that TIMIT has only 1 word-final /sks/ cluster (pronounced in a pretty canonical way), and no word-final /sps/ clusters. However, the much larger LibriSpeech dataset has 6057 word-final /sts/ instances, 499 word-final /sks/ instances, and 80 /sps/. So more on that later: sufficient unto the day is the breakfast experiment™ thereof.]

  4. J.W. Brewer said,

    April 19, 2018 @ 10:36 am

    On Beyond Zebra is one of the Seuss canon's more underappreciated works, imho.

  5. DWalker07 said,

    April 19, 2018 @ 11:15 am

    It's too bad that the most recent Microsoft-supplied browser, on Windows 7, doesn't play these audio clips… Oh well, I'm about to upgrade to a newer version of Windows. And don't tell me about other browsers; I know about them.

  6. Neil Dolinger said,

    April 19, 2018 @ 11:32 am

    Ted Geisel died in 1991, I think before LLog was a twinkle in the eyes of Marc and Geoff, but I'd like to think if he were alive today he would be an avid participant. He was by far my favorite author as a kid, and animated adaptations of his works were among my favorites as well.

    Like probably a lot of us here, my language geekdom started the day I discovered the language section at my elementary school library (good ol' Dewey Decimal section 400). On Beyond Zebra was never the one that my parents or teachers pulled out to read to the kids, so discovering it in the Seuss section was a very happy accident.

  7. Neil Dolinger said,

    April 19, 2018 @ 12:12 pm

    This post is a classic, for so many reasons — IPA vs. phonology vs. Seuss in a 10-round match, with no clear winner.

    It was helpful for you to point out that unlike orthography and IPA, where there are generally agreed conventions regarding the boundaries between the sounds/articulations each symbol represents, phonology is full of spectra that make representation using symbols an almost impossible task — at least using IPA or conventional orthographical systems. You brought up a few of these somewhat established (within linguistics) specta vis a vis the continuum between [sts] and [s], but I wonder whether even with these measures we have an incomplete picture because of physiological differences that exist from region to region, or even from person to person.

    I wonder whether we might eventually come to an IPA that is extended to include various super/subscript numbers to capture all of the phonological spectra.

  8. Jerry Friedman said,

    April 19, 2018 @ 12:24 pm

    Another extension could be to "first step" and the like. I think I usually have less of a [t] there than in "fists", "costs", etc.

    And then there's my favorite English tongue-twister, "Ten-twelfths equals five-sixths."

  9. Neil Dolinger said,

    April 19, 2018 @ 1:14 pm

    "Ten-twelfths equals five-sixths." Love!

    Does any other language outside the Caucasus have as many consonant cluster pileups as English?

  10. Sili said,

    April 19, 2018 @ 2:07 pm

    Cry of fear in danish is angstskrig.

  11. David L said,

    April 19, 2018 @ 2:39 pm

    Ten twelfths equals five sixths

    That's not much a tongue-twister for me, because I say (I think I say, anyway) "ten twelths equals five sixxs," where 'xxs' represents something very blurred.

  12. bks said,

    April 19, 2018 @ 3:36 pm

    On Beyond Zebra was a consciousness-raising experience for me in the second grade.

  13. Neil Dolinger said,

    April 19, 2018 @ 3:48 pm

    "On Beyond Zebra was a consciousness-raising experience for me in the second grade."

    After that, was you woke?

  14. JPL said,

    April 19, 2018 @ 6:35 pm

    @Q. Pheever:

    I agree that whatever the specific solution to indicating these phonetic distinctions is chosen, it should probably be done using diacritics rather than segmental symbols. E.g., a diacritic for [t] lenition, as you suggested. However, as Mark points out, there is the question of whether one should be aiming to describe the pure acoustic and articulatory facts, or whether one should identify a sound (via symbol) by considering the intention of the speaker to produce a particular known sound; e.g., in this case one suspects that the speaker is making a half-hearted attempt to produce [t]. In a field work situation on the other hand, one might not know anything about speaker intentions.

  15. Rhona Fenwick said,

    April 20, 2018 @ 1:50 am

    @Neil Dolinger:
    Does any other language outside the Caucasus have as many consonant cluster pileups as English?

    Even in the Caucasus the famous allegations of consonantal cluster complexity seem to be in practice found predominantly in Georgian, with examples of lesser magnitude in its sisters and occasionally in the neighbouring (but unrelated) Abkhaz and Armenian. Of course, some languages in the Caucasus also have large consonantal inventories (Ubykh has 84 segmental consonant phonemes, Archi apparently 81), but that's not the same thing.

    For others outside the Caucasus, see Nuxálk, which occasionally produces forms such as cktskʷc "he had arrived", sc’qctx "that's my animal fat over there", xłp̓χʷłtłpłłskʷc̓ "then he had had a bunchberry plant in his possession". Some languages from other regions also see complicated consonantal clustering: Sipakapense Maya ʃtqsɓχaχ "they're going to whack it", Tashlhiyt Berber tsskcftstt "you dried it (feminine)", and even Polish wszczniesz "you will initiate" and Slovene skrbstvo "welfare".

  16. Penny H said,

    April 20, 2018 @ 8:46 am

    "He thrusts his fists against the posts and still insists he sees the ghosts": exercise taught in an eighth or ninth grade speech class circa 1963 to get students to articulate those consonants.

  17. Chris Button said,

    April 20, 2018 @ 12:03 pm

    maybe an otherwise unused letter like sigma σ

    Incidentally, the symbol σ was actually used in the original "Principles of the IPA" (1912) for one of the "whistled sibilants" in Bantu languages (the others also had their own symbols). It's interesting how the labialization was deemed intrinsic enough in the sibilants to merit their own distinctive symbols.

  18. Andrew Usher said,

    April 21, 2018 @ 2:04 pm

    '… sees the ghosts' is apparently a very old tongue-twister that's been presented as a speech exercise and even as poetry. Since the previous lines end with 'frosts' and 'boasts', I would _guess_ that it dates to when that was considered an acceptable rhyme – 17th c.?

    'Ten-twelfths equals five-sixths' doesn't seem hard for me; of course I have the usual variation in the last word, but I'm sure I can't elide the 'f' in the first. Is there a similar continuum there? Anyway, prominent renditions of twelfth, and even more so fifth, without any /f/ are truly substandard.

    I also see no problem in extending the IPA in this manner; to me, that's how it should be used. If you're using it properly – as phonetic rather than phonemic (for which a respelling system would be better) – then you pretty much have to indicate somehow whatever phonetic distinction you're currently interested in, whether that's canonical IPA or not. Even if you were going to go the Canepari route you're still going to have such cases.

    k_over_hbarc at yahoo.com

  19. William Berry said,

    April 21, 2018 @ 10:17 pm

    I just wish they’d stop using the IPA in English-language Wikipedia; or, at least, supplement it with a more traditional style of pronunciation guidance. I suspect that the vast majority of Wikipedia users don’t know the IPA. They are badly served by the editors who have made the decision to use IPA almost exclusively as a guide to pronunciation.

    I should have learned it myself by now, and might just break down and do it one of these days, but there is a long list of things that have prior claims on my time.

  20. William Berry said,

    April 21, 2018 @ 10:23 pm

    To be clear, the IPA must be quite useful (imperfections notwithstanding) for you linguistics-type folks, but for the rest of us the traditional dictionaries have worked well enough.

  21. Robert said,

    April 22, 2018 @ 9:25 am


    In Dutch it is angstschreeuw.

    @William Berry
    If you mouse-over Wikipedia's IPA, a tool-tip comes up with "oo as in goose" and such-like. An example is the page for Foobar:

  22. Catanea said,

    April 23, 2018 @ 5:38 pm

    I was quite annoyed by "On Beyond Zebra" as a child, because the extra letters were obviously unnecessary as both their spelling and pronunciation could be representated with just the "regular" alphabet.

RSS feed for comments on this post