Geoff Pullum's recent post "An aim or a name?" stimulated a surprisingly lively discussion of juncture in English. This morning, I thought I'd encourage this interest in phonetics by posting a random sample of relevant real-world examples.
The distinction between "a name" and "an aim" turned out to hard to find — the 25 million words of conversational (telephone) speech that I searched had plenty of instances of "a name", but only one instance of "an aim". So I picked a similar case where both sides of the opposition are represented by dozens of tokens: "a nice" vs. "an ice", in contexts like "a nice guy" or "a nice one", vs. "an ice storm" or "an ice cream sundae".
Here are four random selections from this little collection — see if you can identify them:
Note that I'm not guaranteeing balance here. Perhaps all four are "an ice", or all four are "a nice"; or any other combination of categories.
I'll give you the answer key, with some discussion, tomorrow.
Meanwhile, Bob Ladd's comment on yesterday's post (describing findings from Ladd and Schepman, "'Sagging transitions' between high pitch accents in English: experimental evidence", J. Phonetics 31(1) 2003), characterizes the true situation fairly clearly:
The biggest acoustic difference we found was [...] that word-final consonants are shorter, on average, than word-initial consonants (in our data, 47 msec vs. 67 msec in the case of /n/). [...]There were glottal stops (at the beginning of e.g. Eason) in only 10-15% of cases. In a subsequent perception experiment, we found that people, given a forced choice, could correctly identify the name intended by the speaker only about two-thirds of the time.
The thing that Bob left out is that the within-category variation (e.g. the variation in duration of word-initial [n]) was no doubt of roughly the same magnitude as the average difference between the categories. (The cited paper, alas, doesn't give variance measures, but only ANOVA F values — a pet peeve of mine…)
Also, although Ladd and Schepman were careful to avoid the obvious effects of facultative disambiguation, they were measuring "laboratory speech", i.e. the relatively careful and formal style of speech that you get when you bring people into the lab and have them read (somewhat contrived) passages.
This is necessary in order to get the orthogonal design required for ANOVA and similar statistical analyses, and therefore we phoneticians have been using such techniques for the past century or so; but this approach tends to produce somewhat stereotyped version of the differences under study, human nature being as it is, and to produce results that may be artificial in other ways as well. This practice is starting to change, as we learn to approach similar questions using data from collections of thousands of hours of natural speech.