August 03, 2005

Emphatic unreduction again

In my spare time, I've been continuing to chip away at the the-and-a reduction problem that Chris Waigl and I took up a little while ago. The question at hand is, when (and why) do the and a appear with unreduced vowels? One interesting answer is given in the psycholinguistic literature; but (I think) it's wrong, or at least it's incomplete. In fact, we've already seen some examples of phenomena that this theory doesn't cover, and in this post, I'll give some others. For extra bloggy relevance, I'll take some of the examples from an interview with Glen Reynolds, the Instapundit.

Here's the background of the problem, which you can skip if it's old hat to you. (Even if it's new to you, you might want to skip to the examples and come back to this list later.)

1. In standard dialects, English the and a are pronounced as IPA [ði] and [ej] -- sometimes symbolized orthographically as "thee" and "ay" -- when they are used in citation forms ("the word 'the' is spelled tee aitch ee") or when they are contrastively stressed ("it's *A* factor, but not *THE* factor").
2. In fluent speech, when followed by a word starting with a consonant, both words are usually pronounced with a schwa-like reduced vowel, IPA as [ðə] and [ə], sometimes symbolized in conventional spelling as "thuh" and "uh".
2. When fluently followed by a vowel, the is usually pronounced with a higher vowel, roughly the same as in the second syllable of slithy. In most American dialects, this is the same vowel quality as in a stressed monosyllable such as fee, and is sometimes symbolized in conventional spelling as "thee" ([ði] in IPA) In some British dialects, the vowel is somewhat lower, more like the vowel in fin or this.
3. In all dialects, when a is fluently followed by a vowel-initial word, the form "an" is normally substituted.
4. Some fraction of the's and a's, followed phonetically by a consonant, show the forms [ði] and [ej] instead of the forms [ðə] and [ə].
5. A fraction of pre-vocalic the's and a's show the schwa-voweled forms [ðə] and [ə] instead of [ði] and [ej].
6. When followed fluently by a "filled pause" (the various sounds usually written as "uh" or "um" or "ah"), the pronunciations [ði] and [ej] are usually but not always used. Note that this is expected for the (because the pause sounds are vocalic) but not for a.
7. When followed by a disfluent pause, all of the various alternative forms occur, though [ði] and [ej] are fairly common.

The question, again: why are he "full" forms [ði] and [ej] (thee and ay) sometimes used where the standard rule would say that the "reduced" forms [ðə] and [ə] (thuh and uh) should appear?

For the case of the, Jean Fox Tree & Herb Clark gave an answer in their 1997 Cognition paper "Pronouncing 'the' as 'thee' to signal problems in speaking" (Cognition 62 (1997) 151–167). The answer is telegraphed by the title; as David Beaver summarized it in an earlier Language Log post, people "use the full form when they can't figure how to say whatever the hell they want to say next". It would make sense to extend the same model to the pronunciation of a as well.

But we've seen several examples already that don't seem to fit that mould. In one post, for instance, I cited FDR's use of unreduced a five times in his famous "infamy" speech, along with one use of unreduced the, all six cases in fluent performance of a prepared speech, with no signs of compositional or reading difficulty. And in another post, I cited a single non-reduced a in voice-over by George Vecsey, which struck me as expressing a rhetorical underlining of the following noun phrase, not any compositional problem.

A real solution to this sort of problem requires careful compilation and statistical analysis of many examples from many sources, with audio as well as transcriptions available. However, there's some initial value in looking at anecdotal clips, if only to get a sense of the range of phenomena to be counted, and the aspects of the performances that might be relevant.

The next few examples come from Chris Lydon's 2003 interview with Glen Reynolds. In the portion of Reynolds' speech that I've transcribed, 7 of 97 phonetically preconsonantal the's are unreduced ([ði] rather than [ðə]), and 5 of 74 phonetically preconsonantal a's are unreduced ([ej] rather than [ə]). These rates are fairly typical of what we've seen in other cases. The point here is not the rate of unreduction, but its context.

Chris Lydon opened the interview with this long-winded question:

Let me just say, you know when I was in school, my idea of a god of journalism was Walter Lippmann, he had lunch at the Metropolitan club every day, talked to big shots, and then well sometimes talked to them at home next to the National Cathedral there in Washington and ((then)) he turned out these beautifully phrased short essays for American newspapers twice a week, distilling the mood and the mind of Washington, and of course shaping it. Uh today, the Walter Lippmann is a University of Tennessee law professor with a thing about guitars and Mazda sports cars, uh who's reading hundreds, maybe thousands of web sites all the time, and cuing the rest of the world to where the good stuff is. I want you to tell me how the world created this monster "Instapundit".

Glen Reynolds' answer began:

uh monster's probably the1 right word. [audio link]
um
I am uh hardly in the2 Walter Lippman category, uh about all I can say is that my rate of fire exceeds his
[audio link]
um
but uh but that's about all.

Professor Reynolds' first "the" is pronounced [ðə], as we expect in before a consonant; but the second one is [ði], despite the fact that it's produced in fluent sequence with the following consonant-initial word "Walter". Furthermore, "Walter Lippmann" is hardly new information, since the full form of the name was used twice in Lydon's question, just a few seconds before, and it's not very credible that Reynolds is having trouble remembering it. Nevertheless, Reynolds seems to want to emphasize it a bit, and he accomplishes this in in part by the non-reduction of the preceding "the". While he's speaking deliberately overall -- his verbal "rate of fire" in the interview as a whole is rather slow -- I don't hear (or see) any evidence in the prosody of any sort of phrasal juncture before "Walter".

Another example, about 13 minutes in, occurs when Reynolds is talking about future "horizontal models" of journalism:

and I suspect we will see that sort of thing grow, as the1 software gets better and as the2 network gets larger. [ audio link]

Here both of the the's are unreduced, without any indication that Reynolds is having any trouble fetching the words "software" and "network".

In the other direction , there are several examples in the interview of Reynolds' using reduced (though elongated) articles in front of quite long pauses-for-thought, for example at about 2:50 of the recording:

uh though I'm not sure that the
tone is all that different
than it would have been if I had a couple of hundred readers
because for me the experience is the same.
[audio link]

There's a 980 msec. pause between "the" and "tone", and the vowel of "the" is 340 msec. long, but "the" is still pronounced [ðə]. The word "tone" is new to the interview, and Reynolds appears to be giving himself a "think-pause" before choosing it, but the prepausal article is still produced with a schwa.

That's not to say that disfluency and uncertainty are never relevant. However, it's neither a necessary nor a sufficient condition for unreduction. We already know that uncertainty (real or feigned) is not an essential ingredient in unreduction, because of citation-form and contrastive pronunciations. What we're adding here is the idea that there's a species of article-unreduction that is mainly about vocal underlining of the following word or phrase. (There's clearly another form of article-unreduction as well, a sort of reading pronunciation that can occur even in quite fluent reading from some speakers, especially those that are less well educated.)

Sometimes when Reynolds uses unreduced articles, it does indeed seem plausibly to be linked to uncertainty about what to say next. Here's an example from about 7:20 of the interview:

uh in fact I got an email just today about that, I had linked to the1 blog of a2 military guy in Iraq
uh named L T Smash, that's not his real name, I
actually know his real name, but
but he blogs anonymously
[audio link]

Here both the "the" and the "a" are unreduced; there are no silent pauses or overt disfluencies, but Reynolds slows down as he thinks about how to describe Lt. Smash and his blog, perhaps inhibited by the problem of internally swapping the pseudonym for the true name.

Switching away from Reynolds for a moment, here's another example of emphatic unreduction of a, from NASA's 7/29/2005 Mission Status Briefing. Phil Engelauf is answering a question from the AP's Marcia Dunn, about 17 minutes into the briefing. I've divided (this small piece of) his answer into breath groups:

There has been some discussion about whether or not we might send the crew
to uh take a close look at or remove one of those gap fillers that's protruding
uh that is a1 very very preliminary discussion at this point, ((it-)) we've been sort of asked to uh
[audio link]
take a look at what the impact of doing that would be
uh I don't think that there's a consensus that that's required yet
it's really just a-2 a preliminary "what if" discussion
[audio link]

Case 1 is "a" pronounced [ej] without any pause or pseudopause and without any indication of disfluency or uncertainty. Nor is the following word technical or rare or hard-to-understand -- it's just plain old very, somewhat emphasized. In my opinion, this is basically the same phenomenon as the unreduced a in George Vecsey's comment that Lance Armstrong "goes out as a great champion with a clean record".

Case 2 is "a" followed by a short pause and a repetition. Despite the disfluency and the speaker's clear momentary uncertainty about how to go forward, "a" is pronounced [ə] here.

And for another interesting bit of anecdotal phonetics, this time from Britspeak, here's another example that I heard this morning as I was writing this post (from the BBC Newshour 8/3/2005 12:00 GMT edition, about 48 minutes into the hour). Former Ford president Sir Nick Scheele is being interviewed:

BBC: ... the American car companies are terminally uncompetitive, aren't they?
Scheele:   ah th- they have a huge problem
there is no question that the health care cost problem
allied to
declining profitability
is causing a major squeeze -- however
I think to say that this is terminal is m-
a vast exaggeration.
[audio clip]

In this case, "a major squeeze" has [ej] (or really in this case more like [e]), while "a huge problem" and "a vast exaggeration" have[ə]. The three phrases "huge problem", "major squeeze" and "vast exaggeration" are all reasonable candidates for being underlined, while the only one of them near a disfluency is "vast exaggeration". So the emphatic unreduction theory can't claim any sort clean sweep here; but the think-pause theory doesn't help at all. In fact, these few data points might make you think that Sir Nick has some sort of vowel harmony thing going on...

In a later post, I'll take a critical look at the details of the Fox Tree & Clark paper. In particular, I'll look at their finding that

"About 20% of the time, speakers continue after THIY without further disruption, apparently able to repair the problem in time. But about 80% of the time they deal with the problem by pausing, repeating the article, repairing what they were about to say, or abandoning their original plans altogether"

which was based on counts made from the transcriptions in a British speech corpus for which audio was not available to them, but seems quantitatively very far away from the numbers that we've been seeing in material for which we have the audio. It's hard to tell whether this is because of dialect differences or because of some sort of transcription bias.

Posted by Mark Liberman at August 3, 2005 03:08 PM