Thought experiments on language and thought

« previous post | next post »

Keith Chen's recent proposal that the grammar of tense marking in a language has a causal effect on future-oriented financial and health behaviors is too intriguing to resist talking about. In fact, it reminds me of the words of a prominent linguist who once announced during his talk: "The explanation in question is almost certain to be false. However, if it were true, it would be incredibly interesting, so we have no choice but to explore it."

I'm not sure that this is the best argument for, say, how research funding should be allocated. At least, I've never had the guts to put that in a grant proposal. But if Language Log isn't the place to explore almost-certainly-false-but-incredibly-interesting-if-true ideas, then I don't know what is.

The main finding is that there is a robust correlation such that families who speak a language in which future tense marking is classified as obligatory tend to also engage in self-sabotaging behaviors like saving less money, exercising less and smoking more. Geoff Pullum and Mark Liberman have already made some very cogent remarks here and here, airing concerns over the linguistic analysis of the languages implicated in the correlation, and over the likelihood of spurious correlations arising. Keith Chen has graciously responded to some of these remarks. Dr. Chen is especially persuaded by the fact that the correlation holds if we look only at families who have otherwise been matched on a slew of variables that might serve as possible confounds: country of birth and residence, sex and age of family members, family structure, income, number of children, and religion. If families who are identical with respect to these variables but who differ in the language they speak at home behave in strikingly different ways, then surely this is suggestive of a causal role for language structure.

Controlling for such variables offers only mild reassurance if any, I think, and misses some of the sociological reasons for why otherwise very similar families would choose to speak different languages in the first place—I suspect this choice is often driven by the extent to which people value the culture and practices of that particular linguistic community, and by the strength of personal connection they have to it. So if we suppose, as suggested by Mark Liberman, that both grammar and cultural practices have independently spread throughout the same population despite having no causal connection with each other, then we might find that such a correlation would even be magnified if we restricted our analysis to contrasting families who differ primarily with respect to the choices they've made about which language to speak at home—again, all without there being any causal connection between the grammar and cultural practices of that community.

I'm reminded, for example, of Bill Labov's famous Martha's Vineyard study, in which he found a correlation between a specific feature of a regional dialect (i.e. the pronunciation of words like ice and right with a slightly raised vowel) and the extent to which the islanders valued their traditional way of life and resented the growing influx of mainlanders onto their island. It may well be that if Labov had looked at all of the possible confounding variables that Chen has taken into consideration, he would still have found a correlation between vowel raising and local attitudes. But no one has ever suggested that vowel raising somehow played a causal role in shaping people's attitudes. Nor has anyone suggested the converse and offered a phonological theory of how a language's vowel space is determined by the inward-looking culture of its speakers. But the result is far from uninformative. Rather, it's properly understood as a sociolinguistic phenomenon: Labov suggested that people adopt specific patterns of pronunciation as a way of subconsciously signaling solidarity with a particular community and its attitudinal mindset. Vowel raising and islander traditions and attitudes developed independently side by side. But by virtue of their co-occurrence, vowel-raising had become a sort of identity badge for broadcasting one's attitudes. I think it's well worth keeping sociolinguistic factors prominently in mind throughout the whole debate over tense marking and its impact on behaviour. I'm betting it's not going to be possible to really understand so-called Whorfian effects without getting a handle on them.

A causal explanation between vowel raising and distrust of outsiders is a non-starter for the simple reason that there's no plausible mechanism by which the former could influence the latter. But Chen is right to suggest that one could tell a plausible causal story between tense marking and actions that impact one's future self. The trouble is, correlational studies can only get us so far when there are several viable competing explanations for the correlation. So, what would it take to make a compelling case for a causal connection between grammatical tense marking and financial saving?

This is where experiments can help—we can actually manipulate a variable, and look at whether it has an impact on some behavior we're interested in. So let's run through some possibilities. Ideally, I'd love to be able test Chen's hypothesis by treating languages like transgenic mice: tinker with the DNA of a language like Greek or Spanish to tweak its critical grammatical properties, and observe the effects on the organisms learning them. But a knockout gene experiment of language and thought would be a touch impractical, especially in the current research funding climate. It would require legions of randomly-selected volunteers to agree to live out their lives as speakers of an artificially-altered language such as Greek'—just like Greek, but minus the future tense markers. Perhaps a more skilled grant-writer than I could lobby the International Monetary Fund for the money to run such large-scale experiments, holding out the implied promise of fiscal reform through language policy (if a Brazilian government can ban the present participle, surely the European Union can legislate tense marking?)

In a slightly more serious vein, we should be able to get some mileage from more constrained and realistic experimental scenarios. The idea behind Chen's proposal is that every encounter with a distinct grammatical marker for future time creates a little mental nudge that leads to a conceptual partition between present and future time, and hence, a de-valuing of future benefits relative to present benefits. If that's so, then we should be able to see the effects of these little mental nudges on specific behaviors that we target experimentally. Luckily, the experimental psychology literature is riddled with precedents for an experimental paradigm in which bits of language serve as effective behavioural nudges—for instance, a well-known study by John Bargh and colleagues has famously shown that you can prime meek behavior by having subjects unscramble sentences laced with words like graciously, respect, and honor, and that undergraduates exposed to words like gray, Florida, and bingo undergo accelerated aging, performing badly on memory tests, and walking more slowly. Here on Language Log ("Don't read this post: Be a Language Log reader", 7/22/2011), David Beaver has discussed a particularly interesting study in which people were more likely to vote in an election depending on the wording of a question they'd been asked the previous day: If subjects were asked "how important is it to you to be a voter?" they were dramatically more likely to cast their ballot than if they'd been asked "how important is to you to vote?"

So, we could easily set up a study that looked like this: To control for variables other than language, we might target a group of bilingual speakers of one language that obligatorily marks future tense (e.g. French) and one that does not (e.g. German). This group could then be randomly split, with half of them being made to describe or read about some future event in German and the other half in French. Both groups could then be given a mock investment task that involves making decisions about how much to save versus how much to spend now. In theory, those randomly assigned to the German experimental group should be eager to save more than the French group. Just to be sure to control for any spurious differences between our random groups, we could make each group come back the following week and do the same tasks in their other language, predicting the theoretically-appropriate increase or decrease in their inclination to save for the future.

Suppose we got a nifty result like this. I'm afraid we still wouldn't be able to conclude that grammatical tense marking of a language has an impact on the financial planning habits of its speakers. Why not? Because, it turns out, linguistic nudges can be incredibly coarse and belong to a very broad class of stimuli that prime behaviors by virtue of their association with certain concepts or behaviors. It's true that you can elicit rude or polite behavior with certain words. But you can also prime behavior with just about any stimuli that elicit social stereotypes—people become more aggressive after seeing images of African Americans, perform better on tests after being primed with thoughts of professors rather than soccer hooligans, resist the pressure to conform socially more often after they've seen a photo of a punk rocker than a picture of an accountant, and behave more competitively if they're in a room with a briefcase rather than a backpack. Asian girls, apparently, do better on math tests if primed to think about their ethnic background, and worse if primed to think about their gender. Heck, even logos can prime behavior by virtue of their associations—in one study, subjects who saw subliminal images of the Apple logo performed better on a subsequent creativity test than those who'd seen subliminal images of the IBM logo.

All of this should lead us to worry that our hypothetical research subjects are being primed less by the grammar of their languages, and more by the cultural associations of their languages. In fact, there's some good evidence that language can serve as just this sort of generic social cue, as discussed in Language Log a few years ago ("Non-Whorfian linguistic determinism", 1/2/2009). A 2010 study by Dirk Akkermans and colleagues serves as a particularly nice illustration. In the Akkermans et al. study, Dutch subjects played a business variant of the Prisoner's Dilemma game, intended to test the degree of cooperative versus competitive behavior that subjects would choose as a strategy to maximize profits. (The game is set up so that you reap the highest profits if both you and your partner choose a cooperative strategy of keeping prices for your products high, and the lowest profits if you play cooperatively but your partner chooses to undersell you.) Half of the subjects played the game in English, and half played the game in Dutch—the idea being that the English language is more closely associated with highly individualistic and competitive cultures than Dutch. The subjects who played the game in English did indeed choose a more competitive strategy than those who played it in Dutch. But the effects of language on strategy choice were especially prominent for those who'd lived in an Anglophone country for at least three months; among this group, those who played the game in Dutch played cooperatively 51% of the time, while those who played it in English did so only 37% of the time. In contrast, among those who hadn't spent more than three months in an Anglophone country, the rates for cooperative behavior were 48% for Dutch, and 45% for English. So, the effects of language were strongly mediated by how much direct exposure to Anglophone culture the subjects had. Actual proficiency in English turned out to play no discernible role at all.

Results like these, showing cultural associative effects of language, are sometimes referred to as Whorfian effects (and indeed, the authors of the above paper interpret their results as supportive of the weak Sapir-Whorf hypothesis). But this is puzzling to me. It's not at all clear that this is really a linguistic effect at all. I suspect you would get very similar results if you primed subjects with images of American versus Dutch flags, or recognizable national figures, or perhaps even national symbols like bald eagles versus tulips. It's certainly interesting that languages can serve as repositories for cultural associations—in fact, it's a phenomenon that's quite worthy of study in its own right. But it's not likely to be related in any way to grammatical structure. If we must, perhaps we could refer to such results as socio-Whorfian effects.

But Chen is interested in the impact of grammar on behavior, and so, to experimentally test his hypothesis, we have to go a bit further than simply demonstrating forward-looking financial planning after exposure to German as opposed to French. We'd have to show that exposure to the tense marking of future time in a language has an impact over and above any effects from mere exposure to the language, since the language as a whole may bring along with it various cultural associations. This would be perfectly do-able: If we tested half of our subjects by having them tell a story about a past event, and had the other half describe a future event, we'd predict that the impact of language on financial behavior should be markedly more robust for the group talking about a future event, and thereby forced to use a future tense marker.

Personally, I'm not willing to plunk down any-sized bet that these predictions would come out. If I had to bet any portion of my Language Log royalties, I'd put my money on a socio-Whorfian effect showing an effect of language, but no additional impact of future tense. But Chen has identified a not-crazy hypothesis, and some correlations that are worth thinking about, even worth experimentally investigating. Besides, if the study did show a true Whorfian effect, it would be, well, incredibly interesting.



18 Comments

  1. Chris said,

    February 22, 2012 @ 2:51 pm

    Would be nice if we could get good, historical economic data for languages whose tense systems have changed and compare spending habit of speakers from 500 years ago and speakers now. Might be a natural database, if the economic data exists.

  2. GeorgeW said,

    February 22, 2012 @ 5:13 pm

    Thanks for the interesting post.

    Maybe the EU should have conditioned the financial bailout package on the Greeks adopting an obligatory future-tense marker. Obviously, they don't have one now.

  3. Coby Lubliner said,

    February 22, 2012 @ 6:16 pm

    …bilingual speakers of one language that obligatorily marks future tense (e.g. French) and one that does not (e.g. German)

    I am not aware that future tense marking is any more obligatory in French than in German, except that French offers a choice between a synthetic (je parlerai) and an analytic (je vais parler) form, and German has the analytic only. Both languages allow the use of the present tense with a future meaning: l'année prochaine je vais en Allemagne, nächstes Jahr fahre ich nach Frankreich.

  4. Kenny Easwaran said,

    February 22, 2012 @ 6:17 pm

    One other possible way to structure such an experiment would be to use a language like English in which the future tense is not obligatorily marked, and then have the two experimental groups read descriptions of some future event, with one group reading a description that uses "will" and "going to" everywhere, and one group reading a description that avoids these markers in every case. Of course, this would test a subtly different claim – it would be testing the claim that particular instances of tense marking affect one's saving behavior, rather than the claim that languages with obligatory tense marking affect one's saving behavior differently from languages in which it is optional (I think that was the distinction being looked at?) But this actually seems like a relatively easy experiment to run, at least for those people who (unlike me) are used to running these sorts of experiments.

  5. Julie Sedivy said,

    February 22, 2012 @ 6:49 pm

    Indeed, the problematic aspects of definitively categorizing languages as either obligatorily marking future tense or not has come up several times in the LL discussions—as Kenny points out, in an experimental setting, the presence/absence of a future tense marker could be controlled for rather easily in an experimental setting in which subjects read the materials.

  6. Jerry Friedman said,

    February 22, 2012 @ 7:13 pm

    @GeorgeW: I think you have it backwards. According to Keith Chen, the people who believe in the future are the ones who don't have a future-tense marker; they see the future as present.

  7. GeorgeW said,

    February 22, 2012 @ 8:10 pm

    @Jerry Friedman: Whoops, then the condition should be to ban the future-tense marker.

    (I hope the sarcasm is obvious).

  8. Tim Martin said,

    February 22, 2012 @ 8:18 pm

    Great post! Thank you.

  9. Keith M Ellis said,

    February 22, 2012 @ 8:44 pm

    Your "more competitively if they're in a room with a briefcase" link is incomplete.

  10. Jess Tauber said,

    February 23, 2012 @ 9:30 am

    I haven't had time to look closely enough at arguments being made in this and other related posts with regard to Chen's claims. However, what I have read made me think of the issue of backgrounding of information in grammaticalization as a possible motivation for effects (assuming their truth, which I'm not). Has anyone mentioned this?

    Assuming some validity, hypothetically, wouldn't a search for languages that foreground the same sorts of information linguistically help to balance out the perspective?

  11. Phil said,

    February 23, 2012 @ 10:43 am

    I think Kenny's got it right. If we want to test this effect, we can use a language where overt future tense marking is optional, but not obligatory.

    It also seems to me that this would test Chen's thesis directly, even though we would actually be looking at how specific instances of tense marking affect behavior rather than looking at how the obligatoriness of tense marking affects behavior. I haven't read the Chen paper in full, but assuming that Sedivy's explanation is accurate, then the mechanism by which obligatory tense marking affects behavior is through "mental nudges" induced by particular instances of the tense marking. Such nudges would presumably occur when future tense marking appears but is not obligatory.

  12. Levi Montgomery said,

    February 23, 2012 @ 12:49 pm

    I'm just a novelist, not any sort of linguist at all, so don't throw stuff at me if I'm crazy here, but surely any putatively causal link between obligatory future tense marking and future-oriented financial decisions could be ruled out fairly quickly. Given any sufficiently large number of families (and I must admit I'm not certain why the reference is to families and not to individuals) who speak a given language and only the given language, I would be willing to bet that I could find a statistically significant spread of future-oriented financial decision-making.

    In other words, surely there poor future-planners and good future-planners among the speakers of any single language. Rather than try to determine the relative merit of the theory's claims across two or more languages, why not examine its merit in one language?

  13. marie-lucie said,

    February 23, 2012 @ 3:54 pm

    Controlling for such variables … misses some of the sociological reasons for why otherwise very similar families would choose to speak different languages in the first place—I suspect this choice is often driven by the extent to which people value the culture and practices of that particular linguistic community, and by the strength of personal connection they have to it.

    This excerpt assumes that families have a choice in what language to speak. This may be true in places with large numbers of bilinguals – as in Québec or some regions of the US – but not in places like Greece where the vast majority speaks only one language, the one they have always heard around them, and any acquaintance with a second language taught in school but not significantly used in the community is minimal. I understood Chen's research to apply to mostly monolingual families.

  14. Jerry Friedman said,

    February 23, 2012 @ 4:09 pm

    @GeorgeW: I did get your sarcasm (and I'm aware that my country has a lot of difficulty with fiscal responsibility).

    @Levi Montogmery: Chen is arguing, I believe, that habits such as saving are statistically correlated with the way the future is talked about in people's language. He doesn't say that not having a future tense makes everybody a saver. In my post above, I oversimplified Chen's position drastically, and I apologize if that led you astray.

  15. Jerry Friedman said,

    February 23, 2012 @ 6:19 pm

    @Chris: I bet a nickel that nothing like the required data exists more than 100 or 150 years ago, unfortunately, and for most of the prudent behaviors studied by Chen, the time is probably less. (Now someone will win my nickel and I'll learn something.)

  16. Ben Hemmens said,

    February 23, 2012 @ 6:50 pm

    a) What about countries that have gone through spectacular changes of the level of financial prudence exercised by their inhabitants without speaking a different language? e.g. Ireland.

    b) Will the Greeks have to learn a different language before they can get their act together?

  17. Ponder Stibbons said,

    February 26, 2012 @ 2:18 am

    @marie-lucie:

    I doubt Chen's research was intended to apply to mostly monolingual families. On p. 12 of the paper he explicitly states that the seven countries that provided significant data were those for which there was significant intra-country variation in FTR, which means countries like Greece are ruled out. For at least two out of seven of the countries he used for his data (Malaysia and Singapore), monolingual families who speak Mandarin, Tamil or English (three out of the four languages studied in those two countries) are quite rare. Thus it seems likely that multilingual families were a significant proportion of the families studied.

  18. Egal said,

    February 26, 2012 @ 9:52 pm

    Socio-Whorfian? Really, do we have to? Cringe.

    "All of this should lead us to worry that our hypothetical research subjects are being primed less by the grammar of their languages, and more by the cultural associations of their languages."

    There is just no way to separate these, none. Take close dialects of the same language, one with the grammar variant, one without; those dialects will be associated with cultural variants in their native-spoken space. (The culture of Greek' would be altered by the mere fact of its being spoken by self-aware experiments.) Take the same language, over a span of time in which it's dropped or added the variant; shockingly enough the culture of the people who speak it will also have changed over time. Take contemporaneous families bilingual in the same two languages? Well, you said it all above.

    Whorfian hypotheses are to race and nationality as evolutionary psychology is to gender; the very few legitimate inquiries are all but drowned out by the endless studies searching for some pseudo-scientific confirmation of a broadly-held and often harmful political stereotype.

    Women can't do math? Must be their birthright as women. Better get cracking on a study about estrogen and left-brain function. Greeks don't save money? Must be their language as Greeks. Let's see if we can prove that Greek causes financial irresponsibility. That'll be an easy and convenient way to explain the Euro crisis that doesn't blame or prosecute anyone! (Except the Greeks, as a whole. Screw them, right?) Bonus: the study will boil down to a nice eight-word headline that'll justify a lot of racism and therefore get a lot of press.

    Yes, Whorfian hypotheses are tempting. Let's just say it: they would make the entire field of linguistics more important and relevant than all the other How People Think humanities fields combined. We would all get a lot of funding and everyone would have to listen to us.

    It's vitally important to recognize this temptation for what it is.

RSS feed for comments on this post