Whenever I’ve taught phonetics, I’ve been mildly uncomfortable about the fact that many if not most of the phonetics texts that I have to consult, even ones with a decided focus on English and relatively narrow transcriptions, don’t really note that /tr/ and /dr/ clusters are pronounced with an initial affricate of some sort (as opposed to a stop, that is). When I have seen some mention of this, the voiceless cluster is transcribed as [ʧr] and (more narrowly?) as [ʈʂɹ].
It’s such an easily observable phenomenon, and a student or two typically asks me about it before I have a chance to mention it in class myself. A wee small bit of the research on this phenomenon can be found with a quick web search (which I cite below the fold — if you know of more, please comment). Some of this research is concerned more specifically with the “retracted /s/” in /str/ clusters observed in some varieties of American English.
Unfortunately, there’s no standard way to test whether there’s an underlying /tr/~/ʧr/ contrast (or /d/~/ʤ/, in the voiced case) that is (nearly-)neutralized by this affrication. But President Bush committed a wonderful speech error in his State of the Union address earlier this evening that must say something about either the perception or the implementation (or both) of these clusters.
The speech error occurred somewhere around halfway through the address. (I’m waiting to clip the relevant portion of the audio.) (Thanks to Travis Bradley for sending a clip — amendments follow.) Here’s the transcript — search for “drug trade”, and you’ll find this sentence:
We also show compassion abroad because regions overwhelmed by poverty, corruption, and despair are sources of terrorism, and organized crime, and human trafficking, and the drug trade.
The bolded portion came out as:
and the jug … drug trade.
That’s [ʤʌg], not [dʌg] (or [ɹʌg], for that matter). In other words, the conditioning context for the affrication (_ɹ) is absent in the speech error, yet the “/d/” is realized as [ʤ].
[ Added Feb. 1: Here are spectrograms of (the speech error) jug and (the word) drug, extracted from the audio clip. Though I didn’t measure with any sort of real precision, drug (.34 sec) is *maybe* a couple hundredths of a second longer than jug (.32 sec), but of course they’re in very different contexts, one’s a speech error that is immediately corrected, etc.
jug
drug
Just eyeballing these is hard, and I’m not very good at this sort of thing anyway. If anyone has anything to say about these (the n=2 issue aside), feel free to comment. ]
I couldn’t say for certain exactly what this says (if anything) about Bush’s (or anyone’s) lexical representations of words with the relevant clusters, or the interaction of phonological rules/constraints in his production of these clusters, or whatever. One would presumably want to (figure out a way to) record (acoustically and articulatorily) multiple instances of this type of speech error and systematically compare them with recordings of multiple utterances of words that no doubt begin with affricates (such as the actual word jug to compare with the speech error [ʤʌg] from drug).
I was talking with a UCSD colleague recently about the affrication facts, and dug up a small sample of work on the topic (which I found with a quick Google search for {“/tr/ /str/ affrication english”}). This is a lightly edited version of an e-mail I sent my colleague at the time.
[I]n most if not all varieties of (American) English, the initial stop in /tr,dr/) sequences is not produced in the same way as other instances of /t,d/ […] due to coarticulation with the following /r/ […]. I know there are studies out there about all this […]. For example, the following footnote in this paper:
“Lance & Howie (1997: 356) note that a “fronto-palatal”-affricate occurs in /tr/ for many dialects of at least American English (e.g., beginning spellers sometimes write chrain rather than train), but it is not clear that one finds such affrication after a sibilant. That is, (alveo)palatal affrication of the /t/ in str and its assimilatory anticipation by the preceding /s/ should in principle yield [∫t∫r], but we are not familiar with any reports documenting this pronunciation — which in any case would need one or more clean-up rules/processes in order to yield attested [∫tr].”
The reference is:
Lance, Donald M. & Stephen M. Howie. 1997. “Spectrographic Analysis of English Phonemes and Allophones”. In John S. Kenyon, American Pronunciation (12th edition, expanded) [D. M. Lance & Stewart A. Kingsbury (eds.)]. Ann Arbor, MI: George Wahr, pp. 267-344.I also found the following Linguist List post from 1999, which summarizes responses to a query about /(s)tr/ clusters […]. And here’s a recent impressionistic analysis of /s/-retraction in /str/ clusters (and /l/-vocalization in other contexts) in the African American Vernacular speech of Columbus, Ohio.
Wouldn’t it make sense to view the entire tr (and dr) cluster as a phonetic affricate instead of affricate+rhotic? As you know, a similar phenomenon happens in some Spanish dialects, such that otro ‘other’ and ocho ‘eight’ start to sound homophonous–the former sequence is typically described as a voiceless alveolar quasi-affricate. Treating English ‘dr’ as a phonetic affricate might explain its interchangeability with the phonological affricate ‘j’ in Bush’s speech error. A following rhotic need not be present to condition the affrication of ‘d’ in the speech error because affricate ‘j’ is somehow felt to be equivalent to affricate ‘dr’. By the way, another reference is Greg Lamontagne’s 1993 UMass dissertation. r-assimilation in English ‘tr’ is dealt with in chapter 3.
Thanks, Travis. I thought about the Spanish facts — which occurs in my father’s-side-of-the-family’s speech — and put it aside precisely because, e.g., otro and ocho are “more homophonous” than, e.g., drug and jug. There’s still an /r/ in the English case (though not in the speech error!), but not in the Spanish case. I guess I was thinking this difference might have something to do with the differences between the two /r/-sounds, but now I don’t know why I would have thought that.
And I had forgotten that Lamontagne discussed this in his diss.! It’s been a while. Thanks!
You know, I’ve always wondered about that myself. The phonology textbooks I read for classes never mentioned this (to me) glaringly obvious fact. When I taught about allophones of /t/ or /d/, I always included the affricates occurring before /r/. You could probably make some kind of argument based on the fact that if these affricates weren’t allophones of /t, d/, you should be able to create a minimal pair of nonsense words like drack and jrack, but I don’t think you can.
I think that’s precisely the rub, Neal. Standardly, one says that in an initial cluster of C+liquid in English, C must be either a stop /p, t, k, b, d, g/ or a voiceless fricative /f, θ, s, ʃ/, noting “exceptions” such as */tl, dl, sr/, the “special status” of /ʃl, ʃr/ (in e.g. Yiddish borrowings), and other “peripheral” examples like [vru:m] ‘vroom’.
Since there are no alternations to show anything relevant, it’s impossible to tell what would happen if we simply assumed that, say, an affricate actually underlies the examples we tend to say are /tr, dr/. Do stops and affricates neutralize in this position, or are affricates simply banned from this position in underlying representations, or does something else happen to potential affricate+liquid clusters?
It seems to me that we need good experimental methods for addressing basic questions like these (if we don’t have them already — I could just be ignorant in this regard), since at least some phonological theories make claims about which of these interpretations of the facts is correct. That’s why I found the speech error interesting; this may be a window into how to approach the problem experimentally.
I was starting to think no one else noticed this!
The fact that the sibilant may be pronounced retroflex is new to me, though. I would love to meet someone who speaks a language that differentiates between those sibilants, maybe they could tell us which one it is.
Certainly the spelling mistakes suggest that for some speakers those sequences are identified as /ʧr/ and /ʤr/. I don’t know that I would go so far as to say that they’re coalesced into a single affricate (presumably /ʈʂ/ and /ɖʐ/ or something like that). I’ve been working on a project dealing with these sounds, and in Praat, if you listen to the portion of a tr- or dr- word after the affrication (which is easy to see on a spectrogram) you can still hear an /r/ (that is, you can cut out the /ʧ/ in “try” and hear “rye”). The sound might not be exactly like a plain /ʧ/ or /ʤ/ but it seems a simpler analysis to say that they are simply affected by coarticulation with the /r/ than to propose a new phoneme altogether (there’s probably some lip-rounding due to the influence of the /r/, for example). Certainly “Patrick” and “Patch-Rick” sound the same to me, though, aside from stress.
Also, Polish has three coronal affricates (none of which is /ʧ/ or /ʤ/) and none of them sound like an English /tr/ or /dr/.
Aside from children’s/learner’s spelling mistakes (and possibly mistakes in speech), I can’t think of any way to demonstrate that the /t/ and /d/ of /tr/ and /dr/ are actually identified with /ʧ/ and /ʤ/ phonemically, however.