October 16, 2006

Guys are a bit gabbier in Dutch, too

Hugo Quené, "Multilevel modeling of between-speaker and within-speaker variation in spontaneous speech tempo" (ms., 11 Oct. 2006) "investigates several factors affecting tempo in a corpus of spoken Dutch, consisting of interviews with 80 high-school teachers". Among the findings: "male speakers produce longer phrases (containing 11.1 syllables on average) than female interviewees (10.4 syllables)"; and as a result (since longer phrases are well known to have a shorter average word or syllable duration), "male speakers produced significantly shorter syllables (i.e. faster tempo) than female speakers". Also:

"...the sex difference is also observed in the total number of syllables in each interview: for female speakers, the average interview length is 3541 syllables (s = 765), for male speakers 3855 syllables (s = 499); t(67) = −2.173, p = 0.033. Thus male interviewees are indeed more talkative than female ones."

So here is more data, from a different sort of talk in a different language, that contradicts Louann Brizendine's widely-quoted but empirically unsupported assertions that women talk more than men, and also talk faster.

Hugo goes on to point out that

[s]milar gender differences in talking behaviour have been reported for a large corpus of telephone conversations, where male speakers produced more words than female speakers (mean 926 words vs. 867 words, respectively, in mixed-sex conversations; Liberman (2006)), as well as for formal meetings, where male participants talk longer, and interrupt more often, than female participants (Holmes (1995)). These small but significant differences are most likely coupled to gender differences in the speakers’ social dominance and status.

Liberman (2006) is a Language Log post ("Gabby guys: the effect size", 9/23/2006). Holmes (1995) is Janet Holmes, Women, Men and Politeness (Longman, London).

Hugo is working with a published corpus, so you can check his work or add to it:

The Corpus of Spoken Dutch (Oostdijk (2000)) was used to investigate which factors contribute to (variation in) speaking rate. For this purpose, we concentrated on the subcorpus containing interviews with N = 80 high-school teachers of Dutch in the Netherlands (Van Hout et al. (1999)). Interviewed speakers (‘interviewees’) were stratified by dialect region (four regions within the Netherlands), sex, and age group (below 40 vs. over 45 years of age), with n = 5 speakers in each cell. All speakers are assumed to speak a variety of Standard Dutch as used in the Netherlands. All interviews were conducted by the same interviewer (male, age 26), and similar topics were discussed across interviews. Hence, language variety, conversation partner, and conversation topics were eliminated as confounding factors, and the speech samples were highly comparable among speakers.

Note that there's some reason to think that the women in this sub-corpus might have been slightly more talkative if the interviewer had been female -- there's some discussion of this point in the cited LL post -- but we're a long way away from the claim that women produce two to three times more words than men do, or that they talk twice as fast.

Sex (anti-)stereotypes aside, Hugo's paper offers an interesting lesson in scientific reasoning. He starts by replicating a previously-published result about the effects of region, sex and age on speaking rate:

Results for this model (1) confirm previous analyses of speakers’ average tempo in this corpus (Verhoeven et al. (2004)). First, comparisons of the four regional means show that speakers from the West region (the linguistic center of the Netherlands) produced significantly shorter syllables (i.e. faster tempo) than did those from the other regions (Χ2 = 16.6, df = 3, p = .001). Second, male speakers produced significantly shorter syllables (i.e. faster tempo) than female speakers (Verhoeven et al. (2004)). Third, older speakers produced significantly longer syllables (slower tempo) than younger speakers. For each additional year of age, ASD increases with 0.87 ms. With a grand mean ASD of 212 ms, the tempo difference between speakers aged 25 and 65 is (40 × 0.87)/212 or about 16%. This age effect is well above the JND for speech tempo of about 5% (Quené (2006)).

However, when he also models the effect of phrase length, he finds that the "facts" turn out to be different. Maybe region, sex and age don't really influence speaking rate after all, except indirectly via their influence on phrase length:

First, results confirm that phrase length indeed has a large and highly significant effect on speaking rate, as known from previous research ... Speakers produce longer phrases with shorter average syllable duration, hence with faster speech tempo. Secondly, the effects of the between-speaker predictors (sex, age, and region) all disappear, if phrase length is included as a predictor in the model.

In other words, it seems that speakers of different ages, sexes and regions are producing phrases of somewhat different lengths, on average; and phrases of different lengths vary in tempo; but once this indirect effect is allowed for, there are no longer any statistically-significant differences in speaking rate among the classes of speakers.

Issues like this come up in any correlational study, and responsible scientists spend a lot of time thinking them through, and trying different sorts of analyses (or different sorts of experiments) to sort them out. It's nice to see such a clean example of this process.

[The reasons for the phrase-length effect are discussed in another Language Log post, "The shape of a spoken phrase", 4/12/2006, and in a paper, Jiahong Yuan, Mark Liberman and Christopher Cieri, "Towards an Integrated Understanding of Speaking Rate in Conversation", ICSLP 2006.]

Posted by Mark Liberman at October 16, 2006 08:53 AM