August 16, 2006

Hungarian speech rate and the tribunal of revolutionary empirical justice

A couple of days ago, a reader ("aka darrell") sent me a link to a New York Times article citing a claim that "Hungarians speak 41 percent faster" than Americans do (Alex Mindlin, "For a Memorable Price, Trim the Syllables", 8/14/2006):

Consumer researchers know that people are terrible at remembering store prices: two seconds after taking a product from a shelf, the average person has roughly a 50 percent chance of remembering how much it cost. But few researchers have examined why some prices are more memorable than others.

According to a new study, it is a matter of syllables. Each extra syllable in the price reduces the chances of it being recalled by 20 percent, according to the study, which will be published in the September issue of The Journal of Consumer Research. In other words, someone faced with a $77.51 camera (eight syllables) and a $62.30 bookshelf (five syllables) is about 60 percent more likely to forget the camera’s price than the bookshelf’s, after half a minute.

“The way information goes from the environment to your memory, there is this phonetic loop which is a two-second buffer,” said Xavier Drèze, one of the study’s authors.

Hungarians are far better than Americans at recalling long prices; on average, they can recall 19 to 24 syllables with decent accuracy, while Americans can recall only 13. The authors suggested that this was because Hungarians speak 41 percent faster, both out loud and when repeating sounds to themselves “subvocally.”

The cited article in The Journal of Consumer Research won't be available for a month or so, and so my first thought was to leave this on on the to-blog list until then. But I've posted recently about "Sex and speaking rate", and there are some other points here worth discussing; so I'll take a shot at it now, and come back again when the the promised journal article appears. [Update: I found a preprint online, and Eszter Hargittai blogged about this work at Crooked Timber and pointed me to an online version of the final paper, so further discussion is here.]

It's plausible, to start with, that the ability to remember an exact price depends inversely on the length of the phrase required to name the price -- though I'll bet, on common-sense grounds, that memory for approximate prices behaves quite differently. But the piece of this that attracted my attention is the assertion that "Hungarians speak 41 percent faster" than Americans.

I don't have a lot of experience listening to Hungarian, but such experience as I have doesn't leave me with the impression that it's spoken especially fast compared with English. However, it's not going to be easy to measure the difference, if there is one. First, anyone who has ever taken a look at speech rate, or even thought about it much, knows that people vary in their habitual rates, and that lots of factors affect how fast any given individual talks (emotional state, cognitive load, fatigue, etc.), and that a given individual in a given situation can consciously choose to speak faster or slower. Any one of these factors can easily make a difference of 50% or so. Therefore, a believable cross-language comparison would need a lot of subjects, and you'd have to be certain that you were comparing comparable samples from comparable populations of subjects in comparable settings doing comparable things. Second, because languages have different sound inventories, different word and syllable structures, and different densities of information per word or syllable or whatever, it's not entirely clear how to denominate speaking rate in units that can fairly be compared across languages.

The work referenced in the NYT article seems to be using syllables as the units of measure, so let's take a few trivial measurements to give us an idea of how this is likely to work out. From a Google search for Magyar audio, I picked a recent hir-TV segment ("Hankiss Elemér a szolidaritásról", July 24, 2006), in which Ókovács Szilveszter interviews Hankiss Elemér. In the interviewer's first 10 breath groups, I counted 114 syllables in 26.4 seconds, giving us 232 msec. per syllable, or 259 syllables per minute. From a web search for English-language interviews, I picked an MSDN segment from 2004 in which Shawn Morrissey interviews Steve McConnell. I counted 117 syllables in the first 28.08 seconds of McConnell's first answer, for a rate of 240 msec/syllable, or 250 syllables per minute.

OK, so the Hungarian guy was 3.6% faster. Not exactly 41%, but it's in the right direction.

Not so fast, though -- McConnell is an author, not a professional talker, and he's responding to a question, not reading a prepared statement or question. So I picked a Kai Ryssdal segment ("Good day on Wall St."), from yesterday's Marketplace radio show. Ryssdal's opening bit involves 83 syllables in 17.971 seconds, for 217 msec/syllable and 277 syllables/minute. That's 6.9% faster than the Hungarian interviewer. And the start of the body of Ryssdal's segment had 119 syllables in 24.826 seconds, for a smokin' 209 msec/syllable and 288 syllables/minute, or 11% faster than the Hungarian.

This is silly, of course. We can't conclude anything meaningful from one or two short samples from one or two speakers of each language, even in the more-or-less consistent communicative context of a broadcast interview. But at least we can see that even if there's really a speaking-rate difference between Hungarian and English, the essentializing phraseology of the NYT article ("Hungarians are 41% faster") is profoundly misleading.

The reader who sent the NYT link wrote that "[t]his struck me as being the sort of thing you often like to refute". Really, though, I don't enjoy refuting things, of this or any other sort. I'm a positive kind of person, tempermentally inclined to the enthusiastic suspension of disbelief. But I subscribe (out of reluctant conviction) to what what Cosma Shalizi (following Ernest Gellner) calls the "Anglo-Austrian Tribunal of Revolutionary Empirical Justice":

The procedure of the court was as follows: the accused was blindfolded, and the magistrates then formed a firing squad, shooting at it with every piece of possibly-refuting observational evidence they could find. Conjectures who refused to present themselves might lead harmless lives as metaphysics without scientific aspirations; conjectures detected peaking out from under the blindfold, so as to dodge the Tribunal's attempts at refutation, were declared pseudo-scientific and exiled from the Open Society of Science. Our best scientific theories, those Stakhanovites of knowledge, consisted of those conjectures which had survived harsh and repeated sessions before the Tribunal, demonstrated their loyalty to the Open Society by appearing before it again and again and offering the largest target to refutation that they could, and so retained their place in the revolutionary vanguard until they succumbed, or were displaced by another conjecture with even greater zeal for the Great Purge.

Like other political ideologies, this is a sort of secular religion. At least, it's hard to avoid echoing religious phrases in talking about it. Cosma observes that this ritual of refutation "is very reminiscent of The Golden Bough", talks about "our wanderings from the Goshen of superstition to the Canaan of statistical inference", and quotes Popper's self-consciously sacrilegious quip "better our hypotheses die for our errors than ourselves".

Fine points of doctrine aside, I'm a believer. And so when I lead a hypothesis before the Tribunal, it's an act of respect and devotion, not prejudice and hostility.

[Let me add that I'm also skeptical of the article's assertion that

Hungarians are far better than Americans at recalling long prices; on average, they can recall 19 to 24 syllables with decent accuracy, while Americans can recall only 13.

While there are well-documented differences in digit span among languages (whose explanation is, as I understand it, somewhat disputed), this difference seems too large. There could be something funny in the selection of subject populations, or (more likely) a misunderstanding on the part of the journalist, Alex Mindlin; but we'll have to wait until the article comes out to see.]

[Update -- I found a preprint, and Eszter Hargittai pointed me to the official version. The subject selection seems OK (basically business-school students in both countries), and the reporter seems to have taken the claims fairly directly from the paper. The problems are elsewhere -- see here for more discussion.]

Posted by Mark Liberman at August 16, 2006 07:38 AM