Tuesday, 26 July 2011

hearing plosives

It’s an old truism of acoustic phonetics that you can’t hear the hold stage of a voiceless plosive (as such).

What you get in the middle of a word such as happy ˈhæpi or lucky ˈlʌki is a short period of silence, as the airstream is for a moment prevented from moving through the vocal tract and out of the body. No air movement means no sound.

How, then, can we identify the place of articulation? How do we know that in the first word we have a bilabial p but in the second a velar k?

We know because of the formant transitions created as the organs of speech move into place for the complete closure (the ‘approach’ phase) and then again as they separate (the ‘release’ phase). You identify the p in happy through its effect on the end of the æ and on the beginning of the i. You identify the k in lucky by what you hear in the course of the ʌ and the i.

In the case of a fully voiced plosive, all you can hear during the hold phase is voicing. Again, you identify the place of articulation through the information contained in the formant transitions before and after, i.e. in the approach and the release. That’s how you know that abbey has a b, ladder a d, and lagging a g.

Let’s get back to the plosive clusters we were discussing on Friday apropos of Gdynia. In an English word such as acting ˈæktɪŋ we normally have the same ‘masking’ phenomenon we observed for the voiced plosives in hugged and Ogden.

In the English kt or gd the plosives typically overlap, in that we make the approach for the second plosive before releasing the first. The sequence of events is velar approach – velar hold – alveolar approach (inaudible, because the velar hold is maintained) – double hold (velar and alveolar) – velar release (inaudible, because the alveolar hold is maintained) – alveolar hold – alveolar release. The only audible phases are the velar approach and the alveolar release. In these the formant transitions supply the clues to the places of articulation.

To identify the place of a plosive it is sufficient to hear either the approach or the release. You do not need both. That is how we can tell that the word is ˈæktɪŋ rather than, say, ˈæptɪŋ or ˈætkɪŋ or ˈættɪŋ.

My impression from my first visit to Poland was, then, that for Polish gd- the two plosives did not overlap in the English way. Rather, the sequence of events was straightforwardly velar approach – velar hold – velar release – dental approach – dental hold – dental release. Rather than taking place during the velar hold, the dental approach was delayed until after the velar release. The tiny transitional nonsyllabic schwa between the plosives is created in the tiny interval of time between the two articulatory gestures, velar and dental.

We can leave the native speakers of Polish to debate whether this non-overlapping is usual, as I supposed, or only found in careful or overenunciated pronunciation, as some seem to claim. My impression is that if we compare English actor with Polish aktor it is typical for the English plosives to overlap but for the Polish ones not to. Similarly with the name Magda.

In strongly Japanese-accented English, on the other hand, a word such as actor tends to have a greater interval between the release of the k and the approach of the t. This space might be identified as a Japanese voiceless ɯ̥ (thus “ アクター”). Typically, it seems to be much longer than the momentary mini-voiceless-schwa of the Polish kt. It reflects the Japanese mora-based timing in which equal time is allotted to each of a, k(u), ta, a.


  1. In rapid colloquial Castilian Spanish, the [k] in "actor" (with late stress) is usually pronounced as a velar (it can sometimes be an interdental) approximant. In careful speech, the sequence [kt] is pronounced in the English way. Pedantic newsreaders do it in the Japanese way.

  2. Inspired by the earlier post and responses to it, I sought to go beyond introspections and check out some real data. I've got a good few CDs here that accompany various recent Polish textbooks and I chose one which, in my opinion, has pronunciation typical of good Polish professional speakers - careful, but not overdone.

    In all cases of the word 'gdy' (if), the /g/ was released before the /d/ was formed (quite a few examples in the lesson on conditionals, of course) and, fortunately, Gdynia came up too. Again, the plosives didn't overlap.

    More interestingly, there were are good few examples of a phenomenon I've noticed before - that /nn/ doesn't have to be a single geminated long /n:/, but the first can be released (with an unavoidable schwa) before the second is (re)formed. This occurred in 'Anna', 'panna' (miss), 'inne' (others) and 'odmienna' (different).

    I expect this is a feature of more careful speech, but, in my experience, it's not restricted to broadcasters and actors.

  3. Just a quick question from someone who knows next to nothing about Japanese phonology: why is 'actor' "a, k(u), ta, a", and not a, ku, ta?

  4. @Paul Carley: Yes, practically all geminates can be pronounced as either long or "double", as a function of style, by some speakers at least. (Not necessarily professional speakers.) And it doesn't have to be super-careful; just plain old "everyday careful". I find it a bit jarring, and I seem to remember a Polish pronunciation dictionary recommending the long pronunciation; but it's a fact of life...

  5. @wjarek

    Further nasals:

    Haven't I heard this result in a sequence of different nasals? For example, might not 'w wannie' (in the bath) be pronounced [v vanə̯ɲɛ]?

    Are 'double' nasals down to the influence of literacy, do you think?

  6. Half my transcription disappeared. I wanted a non-syllabic symbol with the schwa, then a palatal nasal, then cardinal vowel 3.

  7. @Paul Carley: Yes, you can definitely get a released /n/ in wannie, which, almost by definition, gets you a schwa-like vocoid in there. As described in John's post.

    The second option is to have it place-assimilated to the palatal, resulting in a long /ɲ/. But full-on place assimilation in plosives is rare (if at all present). Thus, with plosives, you will always get release.

    BTW, I can see the whole of your transcription OK, just that the non-syllabicity mark is not aligned properly under the schwa (as is usual in a browser).

  8. Actually, now I've thought more about it, I think you can also have an unreleased /n/ going straight into the /ɲ/. Polish /ɲ/ has dental/alveolar contact (i.e. it's not a prototypical palatal), so it's not such a big challenge.

  9. I think that name Magda, a short form of Magdalena (Magdalene) is a perfect example as it is a widely known name. I don't know how it was pronounced in original Hebrew (or Aramaic? I have the strong impression that it was the case of non-overlapping plosives though). When we look at how it's pronounced by different people in different languages (Magda, Magdalena, Magdalene) we can see a wide range of deletions, dissimilations, overlapping, and non-overlapping of those two plosives.

  10. @Gassalasca

    I'm sure someone else will come up with a better answer. But just for a quick response.

    My impressionistic guess as an average native speaker of Japanese is that it initially comes from an American pronunciation, ie. the last "a" corresponding to the word-final r.

    In other words, if it comes from RP, it can be arguable but it could have been without the last "a". This may be tested through a phonological translation task with children who may not have much influence from other Japanised English words.

    However, at least in writing, there are cases that the last "a" can be dropped.

    For example, in a fairly recent trend in the IT industry, the English words such as 'server', 'player' as in 'GOM player', 'reader' and 'writer' as in 'card reader' are written as either "sa.a.ba.a" or "sa.a.ba", "pu.re.e.ya.a" or "pu.re.e.ya", "ri.i.da.a" or "ri.i.da" and "ra.i.ta.a" or "ra.i.ta" respectively. So if the trend gets extended to other words, "a.ku.ta.a" may become "a.ku.ta" in the not too distant future?

    Note that their initial form tends to be WITH the final "a", not the other way round.

  11. Some varieties of Spanish (that spoken in rural El Salvador, for example) have syllable-final conflation of plosives, seeming to favor velars in most cases (especially when followed by [s]):
    Pepsi → [peksi]
    pizza → [piksa]
    Maritza → [maɾiksa]
    absolutamente → [agsolutamente] (which is exceptional, since /g/ normally elides to [ɣ] in syllable-final position)

    This velar-substitution can also occur in the Biblical name Job:
    /xob/ → [xog].

    Other alternations, such as in the name Henoc (/enok/ → [enop]) may be hypercorrections.

  12. @Andrew

    That sounds very familiar from my EFL days.

    During the first few lessons some Spanish students would say 'like' for 'light'. It usually didn't take much for them to get over it. It only happened at the very beginning.

  13. This intervening schwa must in part have something to do with the way in which languages tend to divide out syllables that underlyingly respect sonority hierarchy.

    So whether or not the schwa may be reported by native speakers in the Polish examples with gd- may be more moot than the fact that this involves two plosives with equal sonority. The word-initial g- is a semisyllable prefixed to the main mora-bearing syllable. Young-mee Cho and Tracy King explain the features of semisyllabicity in The Syllable in optimality theory.

    I believe that dividing syllable structure in this way, with semisyllables and respecting sonority peak, helps to more intuitively account for all of this, including more "difficult" languages like the Salish grouping that appear (only superficially) to have "unusual" syllable structure or even (gasp!) no structure at all!

    And as for Japanese, it seems to me as though "u" has a tendency to be whispered as in desu or arimasu. In akuta(a), the "u" is again prone to whispering but it nonetheless keeps the original timing of the syllable. I presume in Future Japanese, the weight of the mora of "u" could feasibly be fully transferred to the previous syllable, thereby lengthening the first vowel (but this is just all fun speculation on my part).

  14. Oh and by the way, the differences between English and Polish concerning the realization of kt and gd are in part due to the fact that these sequences are unnatural in word-initial position in English while perfectly natural in Balto-Slavic languages. We also have to take into account the histories of these words and note how the sequences were formed in the first place.

  15. This thread has reminded me of the existence of phoneme "gb" in a lot of
    languages spreading from Liberia to Nigeria (and in Zaire's Lingala) :



    The last name of former president of Ivory Coast "Laurent Gbagbo" (of Bété origin) contains it twice.

    Here below are some recordings of it ("Gbagbo") by various speakers :



    Proper name "Gbagbo" as pronounced in a French sentence
    by Ivorian Guillaume Soro (at the 41st second) :
    ...and by Ivorian artist Gbi-De-Fer (at 1mn22) :

    Russian (transcribed there either as "Гбагбо=Gbagbo" or as "Кбакбо=Kbakbo") :


    I'm under the impression that the speed of realization of /gb/
    sounds the slowest in Russian and the fastest in the speech
    of Ivorians Guillaume Soro and Gbi de Fer.

  16. How do we identify the place of articulation? Certainly in more than one way, as demonstrated by the McGurk effect.

  17. In fact the Lechitic languages support consonant clusters that don't exist in other Slavic languages as a result of different outcomes from the fall of the yers. Where the Eastern languages got epenthetic vowels or metathesis, and Czech and Slovak got syllabic consonants, in Polish consonants got trapped in giant, near-Georgian clusters.

  18. (Oops, saved too soon)

    Compare Polish trzcina 'reed plant', which has two syllables, with its Czech couterpart trstina 'id.', which has three because of the syllabic r: the Common Slavonic form is trъstina. This example is from Tobias Scheer's paper "Syllabic and trapped consonants in (Western) Slavic: different but still the same".

  19. @Glen: Dutch doesn't allow such initial consonant clusters, yet has a velar release in words like "acteur" (/Ak.tU:r/), and not the English velar hold followed by dental/alveolar release. So your reasoning that these clusters in Polish might have something to do with it may not hold up.

    @John C: I read a study recently that showed that Polish actually does have syllabic consonants, but they aren't recognized as such because of the Polish stress rule ("always stress the penultimate"). But e.g. "zdzdblo" (sorry, can't type the accents here) was found to have three sequences that could be described as syllabic (but since it is stressed on the "blo", not recognized as such).

  20. @Kilian Hekhuis:

    (a) Dutch: That's useful to know! Does that apply to all plosive-plosive clusters across the allowable contexts?

    (b) English: As a matter of fact, it's worth remembering that in English, the non-release is just one of the options. The other is, well, "normal" release ;) I've just listened to act, fact, pact, tact, duct, actor, sector, doctor, proctor, rector from the LPD, and guess what, actor is the ONLY one with unreleased /k/...

    (c) Polish: Interesting. Would you have a pointer to that study? The argument looks a bit circular to me. The standard approach would be to claim that syllabification is prior to stress assignment (=stress assignment depends on syllabification); thus, źdźbło is stressed on the /o/ BECAUSE there is just one syllable...