Uh accommodation?

« previous post | next post »

In the course of an enjoyable conversation over lunch yesterday, Michael Chorost asked whether disfluency is contagious, in the sense (for example) that talking with someone who uses "uh" a lot would tend to lead you to behave similarly.  This seems plausible, since such effects can be found in most other variable aspects of speech and language use, so I promised to check — with a warning that causation is especially difficult to infer from correlation in such cases.

As data, I used the transcripts of the Fisher English conversational speech corpus, recorded in 2003 and published by the LDC in 2004 and 2005. There are 11,699 conversations in all, or 23,398 conversational sides. As transcribed, the average number of words per conversational side is 960.

The median rate of uh usage was about 2.6 per thousand words. But 8,494 of the 23,398 conversational sides (36%) had no instances of uh at all, at least as transcribed; and some had a lot — one conversational side had 80 uhs in 592 total words, for a rate of 135 per thousand.

The correlation between the rates of uh usage on the two sides of a conversation was r= 0.383.  This is not especially high — it means that only about 15% of the variance in overall rate of uh usage is explained by knowing the interlocutor's rate. Still, it's certainly statistically significant.

And some ways of looking at the relationship are more striking. Thus of the 4,235 conversations in which the A side never used uh, in 3,507 cases (83%) the B side also never used uh. In contrast, in the 7,464 conversations in which the A side used uh at least once, there were only 752 (10%) in where the B side failed to use uh.

One graphical way to look at the relationship would be to compare the distribution of uh-usage rates for conversations where the partner's uh-usage was greater than the median, to the distribution of rates in conversations where the partner's uh-usage was less than the median. Here's a "bean plot" that does this:

I've plotted the square root of the rate of uh usage, just to spread out the distributions a bit. If we leave out the zero-rate cases, a comparison of the log of the uh-rate distributions is suggestive:

A number of trivial explanations for this pattern come to mind. For example, transcribers sometimes edit out disfluencies (though the transcribers were instructed not to do so in this case), and so it's conceivable that this correlation reflects variation in the behavior of the transcribers rather than variation in the behavior of the speakers.

Assuming that the correlation is really (at least in part) a fact about the behavior of speakers, "accommodation" of filled-pause usage is not the only possible explanation for the correlation. There might well be shared factors (complexity of the subject-matter, for example) that would influence both participants at once.

Finally, we know that rates of uh usage vary with age and sex. There's no guarantee that this set of conversations is balanced for pairings of ages and sexes — in fact, the distribution of topics might well lead to sex or age correlations among speakers. So it would be helpful to use multiple regression to see if the interlocutor's rate of uh-usage still had an effect, when sex and age were also included in the model.

Still, I'll take these results as tending to confirm Michael's conjecture.

(For the purposes of this little experiment, I didn't check on uses of um or ah, and I should have included them as well.)



16 Comments

  1. Sili said,

    April 2, 2010 @ 5:45 pm

    I would expect any effect to be most pronounced at the end of conversations when the uh-less side has had a chance to adjust to the uh-er – or vice versa.

    Anecdata: A fellow post grad. was very prone to hesitative "like" to the point that even I – a notorious uh-er – was distracted. Unfortunately pointing out "hey, you really say 'like' an awful lot" only resulted in my taking up the verbal tick a few moments later …

  2. Morten Jonsson said,

    April 2, 2010 @ 7:40 pm

    I've never come across a verbal tick, but I imagine it would be awfully distracting.

  3. Randy Alexander said,

    April 2, 2010 @ 8:51 pm

    For some speakers, it could have the opposite effect. When I hear people saying uh, um, or like, I become conscious of it and avoid it in my own speech.

  4. Bobo Linq said,

    April 2, 2010 @ 9:44 pm

    Who does these transcriptions, following what guidelines?

    I know from experience that court reporters vary widely in the degree to which they clean up speech. Some court reporters eliminate all ums and ahs; others leave them all in.

    If one person transcribes both sides of these conversations, then the best explanation for the correlation might be the transcriber's style.

  5. Ted Jordan said,

    April 3, 2010 @ 1:29 am

    Despite admiring the writer as involved with the topic, and as knowing more statistics, on the pet peeve level — "and i should have included them as well".

    and further, this not unusual scenario of repetitive redundancy is growing more common as well.

  6. fs said,

    April 3, 2010 @ 5:53 am

    Ted Jordan: Wrong. "and" links the statement that he should have included them to the statement that he didn't include them. "as well" links their inclusion to the actually realized inclusion of "uh", one of their compatriots. This is not repetitive or redundant.

  7. Mike Chorost said,

    April 3, 2010 @ 11:07 am

    Thanks for putting up this note, Mark. It was a pleasure talking with you over lunch too. I was telling my fiancee about your distinction between "um" and "uh", and we were talking about how important insights often come out of noticing things that no one had noticed before. Anyway, your post makes me wonder if patterns of speech are contagious in populations. Perhaps one person "infects" a group with a brief run of collective disfluency and it takes a while to run its course. It wouldn't surprise me if such a thing happened. You might be interested in Gordon Bell's recent book "Total Recall" where he proposes that wearable computers could collect every word a person says in their lifetimes. Now that would make for an extraordinary corpus of language for mining.

  8. Bob Ladd said,

    April 3, 2010 @ 11:40 am

    Bobo Linq is probably on to something in suggesting that the correlation is partly a function of the transcriber rather than the speakers. It's known that disfluencies can be hard to detect even in a task like transcription, where people are trying to record everything faithfully. See an old paper by Robin Lickley on this specific topic, and various other Google-able but hard-to-link papers by Lickley and others on related aspects of disfluency detection.

    [(myl) Yes, as I said in the post,

    A number of trivial explanations for this pattern come to mind. For example, transcribers sometimes edit out disfluencies (though the transcribers were instructed not to do so in this case), and so it's conceivable that this correlation reflects variation in the behavior of the transcribers rather than variation in the behavior of the speakers.

    One reason to suspect that this is not the dominant factor is that the correlation between uh-frequencies on the two sides is in fact fairly low (0.383) -- and the correlation between uh and um frequency (across conversations as a whole) is even lower (0.032). If (say) half the transcribers always noted disfluencies, while the other half never did, we would expect both correlations to be higher.

    Also, as noted in the post, there are fairly large effects of speaker sex and age on both uh and um frequencies, which again suggests that transcriber variation (which surely did exist) was not large enough to disguise this influences.

    However, it would prudent to check, especially to see whether the large number of zero-uh-frequency transcripts was partly due to transcriber variation. I should note, though, that (as you can see in the beanplots) there is still a non-trivial correlation (0.21) in uh-frequencies among the conversational sides where both participants were transcribed as using uh.]

  9. Victor Mair said,

    April 3, 2010 @ 12:51 pm

    I've been annoyed at myself lately for saying "There you go" far more often than I would like to or would normally do. The reason for my starting to say it with unaccustomed frequency is clearly the result of being around some people for whom it is a KǑUTÓUCHÁN 口頭禪 ("pet phrase" — literally, "oral / verbal Zen").

  10. Troy S. said,

    April 4, 2010 @ 4:38 am

    It's only recently I was tipped off that there is a spelling convention conflict Americans prefer uh and um, while the British prefer er and erm. They are they same sound, due to British non-rhoticity. I had imagined some people actually say a rhotic "er" as a stutter. Incidentally, at last I have a satisfying explanation for A.A. Milne's description of "Winnie-ther-Pooh" It's a non-rhotic spelling (for emphasis's sake) of "Winnie-the-Pooh"

  11. Private Zydeco said,

    April 4, 2010 @ 7:31 am

    Setting aside pettifoggery on the matter of how much and in what circumstances utterances such as "like" and others are to be imputed (and/or disputed) as dysfluent/cies and considering them from a memic standpoint, that practically anything sayable can gain wide commonality through inchoate repetitions made by but a few people is by no stretch of statistics-manipulating a wholly undocumented phenomenon. Language is a normative force in its own right, and people of course adopt or try to adopt commensurate levels of formality of register. The scarceness of cases in which "uh" was a feature of one but not the other person's speech patterns can't be considered all that bizarre. As just one example of exceptions to this rule one may conjure to mind a scenario like that in which one of the two persons has undergone formal training in oratory of some kind and the other not.

  12. Private Zydeco said,

    April 4, 2010 @ 9:35 am

    Clarifying some, and in the hopes of not being redundant also…

    Even being more than just somewhat out of the professional league of actual computational Linguists of which a goodly proportion of the L.L. council proper is a party, anyone who is a member of one run-of-the-mill speech community or other could confidently vouch for the validity of the theory imputing many of the hedges, interjections, voiced pauses, et cetera so often found populating quotidian conversation as being pathologic in nature.

    Speaking from experience, after taking up the habit at-best-un-consciously of producing "like" every twenty words, trying to be rid of it again has taxed heavily upon self-discipline. To some there is no reason to be concerned over doing so, but, after all, when abused thus the word carries a stigma of being a sign of undereducation, bad breeding, laxness, etc. It might indeed be classed as a disorder of sorts, and by the same token virulent.

    That said, should the statement that "there's no guarantee that this set of conversations is balanced for pairings of ages and sexes" be construed as meaning that it's likely no effort was made during the study to match young with old and male with female partners any more often than would occur at random? If so, given that and that "rates of uh usage vary with age and sex", and that the transcripts were recorded with a degree of accuracy leveled at including utterances of the kind in question here, there may be insufficient evidence for inferring much beyond observing that like paired with like behaved similarly. As it is a question of detectable alteration of the speech habits of participants who don't usually resort to "uh", alterations in intra-speaker "uh" frequency across time are the element to look for. Determining just whom they were talking to that they resorted to it more than usual is another, as are that other person's speech habits.

  13. Simon Spero said,

    April 4, 2010 @ 9:00 pm

    There is some evidence that some disfluencies may have a positive role in discourse in signaling complex information or new referents. Is there are relationship between the use of a disfluency in your sample, and the introduction of a new referent (either in the conversation or by that party?)

    See e.g.

    Arnold, J. E., Fagnano, M., and Tanenhaus, M. K. (2003). Disfluencies signal theee, um, new information. Journal of Psycholinguistic Research, 32(1), 25-36

    Arnold, J.E., Hudson Kam, C., & Tanenhaus, M.K. (2007). If you say thee uh- you're describing something hard: the on-line attribution of disfluency during reference comprehension. Learning, Memory, and Cognition, 33, 914-930.

  14. Private Zydeco said,

    April 5, 2010 @ 6:33 pm

    Excuse an amateur's narrow-mindedness, much of Lingustic
    science still being incconu to this humble first-year…

    but a broadening of the term's scope here, in consider-
    ation of all dysfluency types, rather, to also include
    instances of "what-is-it-called"s that appear w/ post-
    jacent pauses and subsequent production of whatever then-
    remembered topical referent, could be, along with counts
    for four or five other very similar phenomena, applied
    at some profit to extended analyses of the speech-data
    which Professor Liberman is relying on in the post above,
    with special relevance to the note Mr. Spero just added.

    That is, if the suggestion might be made…

  15. Major Tom said,

    April 11, 2010 @ 10:22 pm

    Though it be terrifically old news to many long-adherent Language Log readers and practically axiomatic to a quorum of the Upper Echelon,
    as another somewhat tangential side-note, the following excerpt from
    Stephen Pinker's "The Language Instinct" (ch. 1, pp. 18-19), a summary
    of some sociolinguistic fieldwork by WIlliam Labov:

    "Another project of Labov's involved tabulating the percentages of
    grammatical sentences in tape recordings of speech in a variety of
    social classes and social settings. "Grammatical," for these purposes,
    means "well formed according to consistent rules in the dialect of the
    speakers." [...] "Ungrammatical" sentences, by this definition, include randomly broken-off sentence fragments, tongue-tied hemming and
    hawing, slips of the tongue, and other forms of word salad. The results
    of Labov's tabulation are enlightening. The great majority of sentences
    were grammatical, especially in casual speech, with higher percentages
    of grammatical sentences in working-class speech than in middle-class speech. The highest percentage of ungramatical sentences was found in
    the proceedings of learned academic conferences."

  16. Online Book Store and News - The Next Question You Beg May Be Your Last said,

    April 30, 2010 @ 10:42 pm

    [...] For guidance, I wrote to Mark Liberman, a professor in the department of Computer and Information Science at the University of Pennsylvania who co-authors an amazing blog called The Language Log. The Language Log will tell you things like what's wrong with (or funny about) the headline "Greece fears batter markets again" and whether disfluency is contagious. [...]

RSS feed for comments on this post