Wednesday 22 December 2010

a credulous scientific report

The December issue of Scientific American carries an article entitled ‘A Click of the Tongue: Ultrasound Translates Dying Languages’. (Thanks to the Clinical Linguistics blog for this link, supplied by Madalena Cruz-Ferreira.)

The article is about the use of ultrasound imaging to study articulation.
This portable technology, which became affordable to linguists around 2000, allows researchers to see the tongue as it moves in real time. It is one of the only medical scanning devices that can keep up with speech; MRIs, for example, are too slow.
Thanks to this emerging technology, [researchers] have documented some of the fastest sounds in human speech: the click consonants present in many rare African languages. Because linguists did not know exactly how the clicks were produced, the sound was placed in a “mixed-bag” category of the International Phonetic Alphabet.

Up to a point, Lord Copper. (OK, if you don't get the reference, go here.)

I wonder what evidence there is that clicks are “some of the fastest sounds in human speech”. Impressionistically, I’d have said that clicks (= sounds produced on a velaric ingressive airstream) are no faster or slower than sounds produced with pulmonic or glottalic airstream mechanisms. I suppose the claim is trivially true, in that the postalveolar (retroflex) click [!], for example, is similar in production to the plosive [ʈ], except that it involves a different airstream mechanism. And plosives are fast(ish). The dental click [|], on the other hand, is typically somewhat affricated, which means it is not so ‘fast’ a ‘sound’. And presumably pulmonic-airstream taps and flaps are the fastest of all.

It is not true that previously linguists “did not know exactly how the clicks were produced”. We can quibble about what knowing something “exactly” might be, and who the unspecified ‘linguists’ are; but phoneticians have been familiar with clicks and their production at least since the 1930s. There is a clear account of click production in, for example, Westermann and Ward’s 1933 Practical Phonetics for Students of African Languages. Their schematized diagrams are pretty good, too.
It was Pike, in his Phonetics (1943), who systematized the classification of airstream mechanisms, describing that of clicks as “velaric ingressive” (or “oral ingressive”).

Nor is it true that clicks are ‘placed in a “mixed-bag” category’ on the IPA Chart. They are in a box clearly labelled Consonants (non-pulmonic), along with the implosives and ejectives that occupy the other columns in the box.

Miller’s research, published in 2009, may well have “organized the clicks based on attributes such as airstream (where the air comes from), place (where the mouth constricts) and manner of articulation”. But in doing this she was merely following a long and established tradition. It is fifty years since I was taught this way of classifying clicks (and all other speech sounds), and I passed it on to my own students throughout my teaching career. I assume all teachers of general phonetics do likewise.

And of course ultrasound imaging in no way enables us to “translate dying languages”, though it might aid us in describing them.


  1. One expects journos to simplify and get things wrong, but it's depressing when it's Sci Am journos....

  2. I couldn't read the full article, I don't have a subscription. But here are some reactions to the public part of the text.

    "Fast" has many meanings and the quoted text was vague. It might mean very brief, but duration can be determined from acoustic records. It might mean high velocity, but consonants don't exhibit velocity. Moving articulators have velocities, that can be measured, and usually different parts of, say, the tongue show different velocities at the same moment, and a given part of the tongue can show different velocities in different situations. To be meaningful it would be necessary to state which part of which articulator during which phase of a click consonant has such a high velocity to make it "faster" than any component of any other consonant. In any case, I'm not sure you can compute articulator velocities from ultrasonic scans. The surfaces of the articulators are elastic and different fleshpoints move around with respect to each other. You can't see, and track, single fleshpoints with ultrasound or X-rays. That's why people glue markers to articulators, assuming a marker remains faithful to its particular fleshpoint.

    The Sc. Am. text states ultrasound is one of the only medical scanning devices that can keep up with speech. Let me put some numbers to temporal resolution to show some examples. Traditional motion films and digital video alike consist of sequences of static images. Human vision will fuse 25 image frames/second into a sensation of smooth movement. That sets a lower resolution limit for an impression of keeping up with speech. Phoneme rates can be as high as 20/second, so a temporal resolution of 25 images/sec will give at least one image, maybe two, somewhere during each phoneme, but you won't know where precisely. Frame rates of 48/second for motion films were common 50 years ago, giving 2-4 images per phoneme. Yet I've watched X-ray films made at 48/s that missed entire occlusions. My own X-ray motion films were made at 75 frames/second (3ms snapshots every 13ms), all events are there, and I know precisely where each frame occurred during the speech acoustic record. I'd probably be happier with 100 frames/sec. Whatever the technology, that's how different temporal resolutions manage, or fail, to "keep up with speech".

    Are the illustrations of vocal tract sections from Westermann and Ward based on radiograms? I doubt it. They look like the illustrations in Daniel Jones's Outline of English Phonetics, i.e. only the grey areas move around, everything else is fixed. I see them as visualizations of what the authors say in words, i.e. they are synonyms for the verbal statements, not evidence for the verbal statements. So we're back to John's point about when do we know we know something.

  3. … the click consonants present in many rare African languages. …

    I'm surprised no one was bothered by this, as the Nguni languages (Zulu, Xhosa, etc.) alone have more than 20 million speakers.

    Clicks are fairly rare, and I'm sure some "rare" languages (whatever that means) have clicks, but that doesn't warrant implying that clicks are only present in dying languages.


Note: only a member of this blog may post a comment.