May 13, 2007

Women and men again, you know?

I'm puzzled. The press is usually all over science stories about sex differences in language use (see "Bible Science Stories", 12/1/2006). And Hillary Clinton is a front-runner for the Democratic party's 2008 presidential nomination. So you'd think that a story about a "scientific" study showing that Hillary "talks like a girl" would get the same kind of play as John Edward's haircuts.

Therefore I'm surprised that the Springer empire's 4/25/2007 press release, "The Power of Speaking Ladylike, has gotten relatively little uptake. It starts like this:

Does gender make a difference in the way politicians speak and are spoken to? This is the question posed in a new study by Dr. Carmelia Suleiman and Daniel O’Connell from Florida International University published this week in the Journal of Psycholinguistic Research. [...]

The researchers found that Hillary and Bill Clinton did largely conform to their gender roles in the interviews, their language reflecting the historic power relation between men and women.

Hot stuff, right? But so far, the mass media have largely failed to rise to the bait. This is just as well, because the study (as we'll see) is deeply flawed. Given the media's general credulity in such matters, I doubt that the (lack of) reaction is due to scientific perspicacity; so perhaps it's just taking a few weeks for the story to percolate into the press. We'll see.

The research report behind this is Camelia Suleiman and Daniel O'Connell, "Gender Differences in the Media Interviews of Bill and Hillary Clinton", Journal of Psycholinguistic Research, published online April 20, 2007 (it hasn't appeared in the print journal yet). Here's the abstract:

Does gender make a difference in the way politicians speak and are spoken to in public? This paper examines perspective in three television interviews and two radio interviews with Bill Clinton in June 2004 and in three television interviews and two radio interviews with Hillary Clinton in June 2003 with the same interviewers. Our perspectival approach assumes that each utterance has a dialogically constructed point of view. Earlier research has shown that markers of conceptual orality and literacy as well as referencing (name and pronoun use for self and other reference) do reflect perspective. This paper asks whether perspective is gendered. Our data analysis demonstrates that some markers of perspective show gender differences while others do not. Those that do include the number of syllables spoken by each interlocutor, referencing, the use of the intensifier so, the use of the hedge you know, the use of non-standard pronunciations, turn transitions, and lastly the use of laughter.

I'm surprised to find this paper in a referreed psycholinguistic journal. The analysis is interesting, but its data has no logical connection whatever to gender differences. There are exactly two subjects, and it's true that one them is female while the other is male. But in addition, one of them is from a suburb of Chicago while the other is from rural Arkansas. So perhaps this is really a study about "Regional Differences in the Media Interviews of Bill and Hillary Clinton"? And the two subjects differ in many other ways as well -- the article could with equal plausibility have been presented as telling us about "Social Class Differences" or "The Effect of Early Family Life". Or just "Individual Differences".

The authors' argument is based on citing the relationship between their findings and the earlier literature on "language and gender". The idea is to show that Hillary and Bill conform to previously-claimed patterns of gendered language use. There are two problems with this method. The first one is that the authors' data is equivocal: the features that they measure in the Clinton's interviews partly agree with the previously-claimed gender differences, and partly disagree. This is exactly what we'd expect if the differences had no particular connection to sex or gender at all. The second problem is that many of the background "facts" about sex/gender differences cited in this paper were originally asserted without any empirical evidence -- and some of them have turned out not to be true. More on this later.

The day after the Springer press release came out, Dennis Baron did his best to blog the story along ("Hillary Clinton: Runs like a man, talks like a girl", 4/26/2007):

Hillary Clinton talks like a girl. That’s the conclusion of a pair of psycholinguistic researchers who analyzed radio and television interviews with Sen. Hillary Clinton and former president Bill Clinton recorded in 2003 and 2004, just after each had published a memoir.

Over the last few days, there has been a bit of marginal mass-media pickup, for example Robin Lloyd, "Clinton-speak reflects political gender: Hillary's use of language is more 'ladylike' than Bill's, researchers say", MSNBC, 5/10/2007:

Some pundits say former President Bill Clinton and Sen. Hillary Clinton operate like a unified political animal. But a study of their TV and radio interviews reveals that gender separates the speech of the power couple, such that he "talks like a man" and she is "ladylike."

[Update -- for completeness, some that I missed earlier: Jim Ritter, "Hillary talks like a 'lady' -- STUDY: Unlike husband, she uses 'non-powerful' words", Chicago Sun-Times. 4/25/2007; "Scientists: Hillary Clinton's a Lady ... When It Comes to Her Speech", Fox News, 4/25/2007; Marilia Duffles & and Jeffrey Lord, "Why Hillary Talks Like Bill", The American Spectator, 5/7/2007 (which deserves examination in its own right).]

And Wesley Pruden, "When Hillary speaks, a lady emerges", The Washington Times, 5/11/2007

When they examined several hours of radio and television interviews of Bill and Hillary, they discovered that Bill inevitably "talks like a man" and Hillary is careful, perhaps subconsciously, to sound "ladylike." [...]

"Even though Hillary Clinton is a politician herself," the researchers found, "she still follows to some extent the historic designation of women's language as the language of the non-powerful."

For example, Hillary is nearly three times more likely to sprinkle her conversation with the linguistic cringe "you know" than Bill is, lapsing into the schoolgirl hedge that diminishes the power of language. Women, the professors say, are more likely to "hedge" than men.

Let's pick up this business about what Pruden calls "the linguistic cringe 'you know'". Suleiman and O'Connell took this from Robin Lakoff's classic 1975 work, "Language and Woman's Place":

Lakoff (1975/1975/2004), however, while acknowledging the stylistic difference, points to the power relation that exists between men and women. Thus, women’s talk is powerless talk. She identifies powerless talk as (a) having a stock of phrases that belong to the domestic domain, (b) a stock of adjectives like divine, (c) rising intonation in statements, (d) use of hedges such as you know, and of intensifiers such as so, (f) use of hypercorrect grammar, and (g) use of polite and indirect statements. Lakoff (1975/2004, pp. 78–81) calls this style ladylike talk.

So use of "you know" is one of the features that S&C measured, with the results as shown in their Table 2 (with the relevant part outlined in blue by me):

(Note that the measure is syllables/you know, so that lower numbers mean higher relative frequency of use of "you know".)

Here's the first part of their discussion:

As Table 2 indicates, Hillary Clinton uses you know much more often than Bill Clinton does (11 < 316 syl/you know). [...] This is in accord with Lakoff’s (1975/2004, p. 79) finding that women use more hedges than men.
[11 < 316 is a typo in the original article -- it obviously should be 111 < 316 ]

The trouble is, Lakoff's "finding ... that women use more hedges than men" was simply an assertion, based on her impressions and not on any counts of the use of hedges by any specific women or men on any specific occasions. But inspired by her ideas, many researchers since 1975 have investigated the matter in detail. I discussed the history of research on one kind of hedges, "tag questions", in a post a few years ago ("Gender and tags", 5/92004). You can learn more of the details from that post and the references it cites, but the basic result is that Lakoff was empirically wrong about sex differences in the rate of use of this particular kind of hedge -- for the kind of tag questions that genuinely express uncertainty, men appear to use them more often than women do.

What about the effects of sex on the frequency of use of "you know"? There may well be some published research on this topic, but I don't know of any. (If you do, please tell me.) So it's time for a Breakfast Experiment.

The LDC Online index of transcribed telephone conversations allows searching by both the sex of the speaker and the sex of the interlocutor. The conversations in question come from several published corpora, including Switchboard, Fisher I and Fisher II. In the current state of the index, it covers 28,274 conversational sides (i.e. 14,137 two-sided conversations). There were 12,589 male speakers and 15,685 female speakers. The speakers range widely in age, educational level, region, ethnic and socio-economic background (though I have not tried to control specifically for such factors in this experiment).

The first result is that males use "you know" about 10-11% more frequently, on average, than than females do. If I search for {"you know" & sex:male} I get 173,321 instances in 11,753 conversational sides, or 14.7 per conversation). Searching for {"you know" & sex:female} I get 198,086 instances in 14,849 conversational sides, or 13.3 per conversation. That's if we count only the conversational sides where "you know" was used -- if we count all conversations, we get 173,321 in 12,589, or 13.8 per conversation for the guys, and 198,086 in 15,685, or 12.6 per conversation for the ladies.

Either way, I think we can reject the idea that "you know" is strongly associated with female speakers in the U.S. these days. Far from being a "schoolgirl hedge that diminishes the power of language", you know might instead be viewed as a guy thing, a marker of male social solidarity. (Though let's not get carried away -- the sex differences are not very big.)

It's interesting that (in the interviews that Suleiman & O'Connell surveyed) Hillary used "you know" three times as frequently as Bill -- but there's no reason to think that this is because "gender make(s) a difference in the way politicians speak and are spoken to".

Suleiman and O'Connell suggest that we also need to look at the other party in the conversation -- they observe, for example, that "Hillary Clinton's mean syllables/you know indices in these interviews are notably different with men TV (M=72.5), woman TV (M=113) and Radio (M=150.5) interviewers". I need to point out again that we're talking about unreasonably small numbers to generalize from -- two male TV interviewers and one female TV interviewer, in this case.

So here is the 2X2 table from searching the same LDC-Online conversational speech corpora, expressed in terms of the average number of instances of "you know" per conversational side. (The numbers are in parentheses divide the counts by the total number of conversational sides of the relevant type, whether or not any instances of "you know" occurred there.)

Speaker Female
13.6 (13.0)
12.7 (11.8)
13.4 (12.9)
15.2 (14.3)

In addition to the result noted earlier (males tend to use "you know" more than females), overall, this susggests that participants in same-sex conversations tend to use "you know" more than participants in mixed-sex conversations. (I'm not taking the time to do statistical tests this morning, but with 28,274 subjects, these differences are likely to be statistically significant.)

The tendency for you know to be used more frequently in same-sex conversations is the opposite direction from the effect noted by Suleiman & O'Connell, which was that Hillary Clinton used "you know" more often in two TV interviews with male interviewers than in one TV interview with a female interviewer. Does this mean that Senator Clinton is somehow acting against dialogic stereotype? Well, my first guess would be that it doesn't mean anything at all about dialogic gender, but relates to the individual personalities or subject-matter involved in the interviews. Or perhaps it's due to what Senator Clinton had for breakfast.

Since I've expressed my results in terms of numbers of "you know" per conversational side, you may be wondering whether the excess male use of "you know" means that guys use it more often per unit time or per syllable or per word, or whether it's just that guys talk more.

In an earlier post ("Gabby guys: the effect size", 9/23/2006), using a subset of the same data, I found that in single-sex conversations, men used about 3.2% more words, on average, than women; while in mixed-sex conversations, men used about 6% more words, on average, than women. The male excess in "you know" per-conversation rates in single-sex conversations was 15.2 vs. 13.6, for an excess rate of about 12%. In mixed-sex conversations, it's 13.4 vs. 12.7, or about 6%. For comparison with S&C's syllables/you know numbers, the male/male conversations in Fisher I involved about 930 words per side, so that 15.2 you know's per side translates to about 61 words per you know.

I don't have the time to count syllables just now -- I've finished my second cup of coffee, and now I've got a Mother's Day picnic brunch to prepare -- but a quick estimate would be that both male and female speakers in this conversational corpus use "you know" somewhat more often, in terms of syllables/you know, than the rate that S&C observed in Hillary Clinton's interviews.

You can take it the interpretation from there. As usual, when we look at the facts instead of relying on stereotypical impressions, the overall sex differences in speech and language (leaving aside biologically-determined features like pitch) are pretty small. And as is all too often true, the empirical differences are in the opposite direction from some supposedly scientific "findings".

Posted by Mark Liberman at May 13, 2007 07:10 AM