The Yale economist Keith Chen has made quite a stir by his claim that there is a connection between grammatical marking of the future and future-related behavior: in short, that speakers of languages which mark the future save less and care less about their health. The purported explanation is that obligatory marking will make the future seem more distant and thus less important to care about. A paper on the topic, forthcoming in American Economic Review, is available from his homepage; in addition, there is a TED talk and numerous media reactions which can easily be found by googling.
This touches me personally since the linguistics that Chen bases himself on is to a significant extent taken from my own work and that of the EUROTYP Tense and Aspect group that I led in the nineties (Dahl 2000a). Moreover, Chen contacted me at an early stage of his work, offering me collaboration; I politely declined, but made a number of comments, which he has at least partly heeded. I do think it is fair to say that Chen has done his linguistic homework, although there are still points to be debated. The disclosure I should make here is that I have for most of my career been a convinced anti-Whorfian, and although I have recently had to admit that the languages we speak may indeed have certain effects on our thinking, I am not too comfortable with becoming associated with the global marketing of a rather strong Whorfian hypothesis – it is not quite what I expected my “fifteen minutes of fame” to be like. This somewhat emotional reaction may of course reflect my own cognitive bias. Some of my colleagues have reacted similarly, however, and have called Chen’s ideas “silly” or “nonsense”. I am not sure this is the best way to respond to these ideas; it is not quite as easy to poke a hole in Chen’s argument as one might think, and to outsiders it may seem like an irrational knee-jerk reaction. It is better if we try to analyze the claims seriously; maybe we can learn something from it – especially if the results are spurious, it may make us be more careful about our own pet hypotheses. Some people (Geoff Pullum, Mark Liberman, Sean Roberts, Amos Teo) have already made their contributions to such an analysis: part of what I am going to say echoes their critique, but I hope also to be able to add a few points of my own.
To start with, a point about terminology. The abbreviation “FTR” stands for “future time reference”. We used it in the EUROTYP Tense and Aspect group as a way of singling out sentences that concern the future, whether they are in any way marked grammatically or not. But Chen rather consistently uses “FTR” for “grammatical marking of FTR”, hence “strong FTR” = “strongly grammaticalized FTR”. Thus, he can say things like “sentence about future events without FTR” (fn. 9), which should be a contradiction in terms. As one of the original promotors of the abbreviation, I find this usage rather irritating. To avoid misunderstandings, I will not use the abbreviation except when quoting Chen (1).
Chen thus divides languages into “strong FTR languages” and “weak FTR languages”, and the idea is that this division is reflected in behavior relating to the future. The main criterion for the division is whether the language obligatorily marks “prediction-based future time reference” grammatically or not. In the EUROTYP Tense and Aspect volume, I gave weather forecasts as canonical examples (Dahl 2000b). I tried to show that there is an area in Northern Europe where languages tend not to have strongly grammaticalized FTR, and labeled it somewhat uncautiously (although in quotes) “the futureless area”. Finnish and Estonian would be the most extreme cases with hardly any future marking connected to the verb; the other languages (basically all the Germanic languages with the exception of English) would display future marking devices of varying degrees of grammaticalization.
Several people have objected to the classification of English as a “strong FTR” language. This objection is partly misguided, but Chen has himself to blame for this, since he exemplifies the claim that English has “strong FTR” by the impossibility of saying I go to a seminar about a seminar later today. It has been pointed out that English speakers would naturally say I am going to a seminar, with the present progressive, in the situation in question – and thus, it seems that English can in fact express this without grammatically marking future time reference. However, the example is irrelevant to Chen’s classification, since it is about what I plan or intend to do and is not a prediction of the kind found in a weather forecast. Furthermore, you could in fact argue that the progressive is an future-marking device in I am going to a seminar, since it does not have the normal meaning of a progressive construction (something on-going at the time of reference). But this raises some questions which may be problematic for Chen’s approach.
First, what does “grammatical marking of future time reference” or “grammaticalized future time reference” mean? Even in Finnish and Estonian, the fact that a sentence refers to the future is sometimes grammatically relevant: in a sentence with an object, it is correlated with the choice of case marking of that object. (I am avoiding getting into the details here.) In Swedish, a sentence such as Det är kallt imorgon ‘(lit.) It is cold tomorrow’ sounds rather unnatural. We would rather say Det blir kallt imorgon ‘(lit.) It becomes cold tomorrow’. (I describe this phenomenon in another paper in the EUROTYP volume.) In other words, we choose a different verb, but since this is a systematic (quasi-obligatory) phenomenon, it borders on being part of the grammar. Russian is treated as a “strong FTR” language by Chen; however, perfective verbs have no present-future distinction; rather, there is just one non-past form which is usually interpreted as referring to the future. Similar circumstances make “strong-FTR” and “weak-FTR” languages less different from each other than the simple dichotomy may imply. This may be something that I ought to have stressed in my article in the EUROTYP volume. However, my goal there was mainly to identify some properties that could be used to delineate an area in northern Europe where languages tend to behave the same way.
Second, it may be noted that conflating the future and the present is not restricted to the languages in the “futureless area”. In fact, Romance languages seem more generous in this respect than English. I discuss this in Dahl 2000b referring to work by Suzanne Schlyter and her associates at Lund University. Thus, in Spanish and French, if the speaker is talking of something that they intend to do in the immediate future, it is highly probable that the present tense will be chosen. In a sense, then, future time reference is not totally grammaticalized in those languages either, a fact that was obscured in my chapter by my insistence on using prediction contexts as a criterion for grammaticalization, which implies a single path that future tenses develop along. Again, I may in retrospect wonder if that was a fortunate choice. In the context of Chen’s project, however, the question that arises is why it should be precisely those contexts that determine our attitudes towards the future, given that we probably more often make intention-based than prediction-based statements. Arguably, it should be the total set of patterns a language provides for speaking of the future that influences how we think of it rather than just the use of future tenses in prediction-based contexts.
Third, and not unrelated to what I have just said, there is a lot more involved in deciding on how future time reference is marked than just the fact that we are speaking of the future. Thus, the distance to the time spoken of, the cognitive status of the proposition put forward and the role of the utterance in a larger discourse context may all influence whether a sentence is marked grammatically and how it is marked. This is in fact implied by the distinction between “prediction-based” and “intention-based” future time reference. One could argue that English will marks the prediction status of the sentence as much as its belonging to the future. If that is the case, a possible Whorfian effect might be to emphasize that the future is less certain than the present – which should lead to more savings rather than less.
Something has to be said about the linguistic data on which Chen bases his claims. His sources include the papers in the EUROTYP volume but also, especially for non-European languages, “several other cross-linguistic analyses” and individual grammars. But he has also made an ambitious attempt to investigate the use of grammatical future time marking empirically by counting such markings in weather forecasts in 39 languages. Since this is only a subset of the languages he was looking at, he used this data only as a check on his main data base. He assures us that doing so produces “results that are nearly identical (both quantitatively and statistically) to the results I report in this paper”.
The choice of genre is clearly inspired by my EUROTYP chapter, where I quote a Finnish and an English weather forecast to illustrate the typological difference between these languages. However, there are certain problems with using this particular genre for classifying languages in general. One is that you may need slightly longer texts than Chen has been able to gather (on an average 46 sentences from each language). Another problem is that it is not excluded that there are genre-specific conventions governing the choice of verb forms. If you are in a weather forecast, you know that reference is made to the future and that anything said is supposed to be the meteorologist’s best guess, hence a prediction with a certain margin of error. One might therefore expect that there might be fewer markings than in an average text.
The results of the weather report study, however, are slightly puzzling. The first thing to be noted is that it looks as if the distinction between “strong-FTR” and “weak-FTR” languages is not at all a binary one but a continuum. This is perhaps not so unexpected for linguists used to working with natural data, but what is a little strange is that some of the alleged “strong-FTR” languages have rather low incidences of future marking. If 90 per cent is taken as a reasonable cut-off point for obligatoriness, only 16 out of 28 languages said to be “strong-FTR” make it. In Polish and Hungarian, only about a third of the future referring sentences are marked. It does appear strange to assume that such a low ratio could count as “obligatory use”. As I said, that we are dealing with a continuum is not unexpected from the linguistic point of view, but it looks to me that it will create problems for Chen’s explanation of the connection between future marking and behavior, which rests on the assumption of a binary choice between obligatory and non-obligatory forms. However, I am not sure that the numbers are correct. I had a quick look at some Polish weather sites, and my impression was that on the one hand, future markings were in place when they were expected, but on the other, there were also plenty of non-future referring sentences. It is possible that these have been wrongly counted as counterexamples.
The appendix to the paper includes what Chen characterizes as “a large table of all languages included in this study and their coding”. The number of languages is 126, but that includes some languages with a very small number of native speakers and thus hardly relevant to his hypothesis, such as the Aboriginal Australian languages Alawa and Bandjalang (taken from my 1985 book), and the Turkic language Karaim spoken in one town in Lithuania (taken from the EUROTYP volume). I am slightly skeptic about the possibility of determining whether all the languages in the list are “strong-FTR” or “weak-FTR”. When preparing the WALS map “The Future Tense” we decided to refrain from classifying futures in any other way than by whether they were expressed morphologically or not, since we found that the information in grammars was usually not sufficient for anything else. Chen does not give a full list of references but only lists the works that have been “most important”, so it is sometimes hard to judge about his sources. Another problem is to what extent the languages in the list cover the countries studied. For two of his major data sets (from OECD and SHARE), basically all the countries have majority languages which are included in the list. The World Values Survey is more problematic. The survey covers at least 86 countries, but Chen chooses to study “a set of 76 countries for which language data are available”. He does not list those countries, but some conclusions can be drawn from Figure 1, which is a plot of the languages with “Share of Households Speaking a Strong-FTR language” on the x axis. Most languages are at one of the ends of the diagram and are not named, but Malaysia, Estonia, Ethiopia, Singapore, Switzerland, Nigeria, and Burkina Faso are indicated as having different proportions of “strong-FTR” and “weak-FTR” speakers. Among these, Ethiopia, Nigeria, and Burkina Faso are highly diverse linguistically: according to the Ethnologue, they have 87, 522, and 68 languages respectively, in other words, more than five times as many languages as the total number listed in the Appendix. For Nigeria, I can only find the four most widely spoken languages in Chen’s list; these — Fula, Hausa, Igbo, and Yoruba – account for about half the Nigerian population. The value of “Share of Households Speaking a Strong-FTR language” apparently refers to the percentage of the participants in the survey, not to the percentage of the country’s population. But the question then arises: were all the participants in the survey speakers of one of these four languages? If not, how were the others treated?
More unclarities relating the choice of languages: In Figure 1, ten unnamed countries taken from the WVS are said to have “almost no strong-FTR speakers”. But this is the same number as the clear cases of “weak-FTR” OECD countries in Figure 2. Taking into account that Iceland is not included in the WVS survey, this leaves room for exactly one new “weak-FTR” country when extending the sample from 35 OECD countries to 76 in the WVS. This is somewhat unexpected. I will return to this question. On p. 2, “nine multi-lingual countries with both weak and strong-FTR populations” are singled out for within-country analysis of saving behavior. However, in Figure 1 and Table 4, this number has decreased to seven: Belgium and the Democratic Republic of Congo have mysteriously vanished. It is probably too paranoid to assume that this is because they did not behave the way they should according to the hypothesis. But the whole thing is a mystery, since with the 59 unnamed countries that “have almost no weak-FTR speakers” there are exactly 76 countries altogether in Figure 1, which is the total number of WVS countries studied by Chen – implying that the two missing countries have been somehow re-classified.
I do not think that any of the above considerations wholly invalidates Chen’s argument. It does appear that there is some kind of statistical correlation between the marking of future time reference and the non-linguistic parameters he has investigated. The weakest point, as has been noted by others, is the absence of independent support for a causal link grammar->thinking->behavior. Chen’s idea, that grammaticalized future marking somehow makes the future “more distant” is rather crude and hard to make sense of. Distant compared to what? Does it mean that the whole future, from the next instance on, is moved away mentally from the present? And suppose that we had found a positive correlation between future marking and savings, wouldn’t we then be able to construct an equally plausible explanation for that? People who don’t mark the future might be supposed to be more sanguine about the future since there is no reason to think it’s different from the present. A commenter on Language Log says “Chen opens up fascinating insights into the language of squirrels, who we now know hide their acorns because they have no future tense!”
So why do we then get the correlation? It is unlikely that it is due to mere chance. But as has been pointed out by several of the bloggers cited above, cultural phenomena come in bundles, and typically, there is no very profound explanation behind this, rather the bundling is due to several features happening to take part in the same historical process of diffusion. To add to the examples already given elsewhere, consider the fact that all the nine full member countries of the International Cricket Council also have left-hand driving. (I am not counting the tenth member, the West Indies, which is not a country.) There is no universal principle connecting cricket and left-hand driving, rather both these phenomena happened to be part of British culture at a specific point in time and spread with the growth of the British empire.
Let us look at the “weak-FTR” countries in Chen’s material. As noted above, he does not specify which they are, except for one source – the 35 OECD countries. In Figure 2 in his paper, 13 of these are indicated as “weak-FTR” countries but two of these, Switzerland and Belgium, are said to have “significant within-country FTR variation”. (It seems that Estonia was forgotten here, although it is treated as an in-between case in other places.) What is striking is that all these countries except one – Japan – is situated in a contiguous area in Northern Europe, which also happens to coincide with the now infamous “futureless area” identified in my EUROTYP chapter. It stands to reason that the countries in this area may also have other things in common. This turns out to be confirmed by one of Chen’s own sources – the World Values Survey website, more precisely the page called “The WVS Cultural Map of the world“, authored by Ronald Inglehart & Chris Welzel. “This map reflects the fact that a large number of basic values are closely correlated; they can be depicted in just two major dimensions of cross-cultural variation.” The two dimensions are called “Traditional/ Secular-rational values” and “Survival/Self-expression values” and are together said to explain “more than 70 percent of the cross-national variance in a factor analysis of ten indicators-and each of these dimensions is strongly correlated with scores of other important orientations” – presumably including the features studied by Chen. So how does his division of countries into “strong-FTR” and “weak-FTR” relate to the dimensions identified by Inglehart & Welzel? The map below shows how. The red squares represent the ten clear cases of “weak-FTR” countries according to Chen’s Figure 4. (East and West Germany are represented separately.) The green diamonds are the three intermediate cases in Figure 4 and the blue diamonds are all the other countries in the WVS. (The figures represent slightly older data than the maps on the webpage, but changes are only in the details.) As we can see, all the red squares are found in the upper right-hand quadrant of the diagram, except Japan which is slightly outside. We can also see that the area in the diagram where we find the “weak-FTR” countries corresponds fairly well to the cultural area called “Protestant Europe” by Inglehart & Welzel. Japan is outside this area; it has a negative value on the x-axis (“Survival/Self-expression”) but is one of two outliers on the y-axis (“Traditional/Secular-rational values”). The other outlier is my home country Sweden which is also an outlier on the x-axis.
What does this mean? It appears that there is a fairly strong correlation between being a “weak-FTR” country and having a positive value on the two dimensions in the WVS Cultural Map. Using Chen’s logic, we could argue that the existence of a future tense influences not only whether you save and smoke but also your whole outlook on life, including whether you believe in God, are pro-life or pro-choice and a lot of other things. I checked the statistics on the Wikipedia page “Demographics of atheism” and could conclude that the percentage of believers in a personal god is on average twice as high in the “strong-FTR” OECD countries in Europe, which is consistent with the WVS dimension “Traditional/secular-rational values”. I also checked the Wikipedia page “List of countries by intentional homicide rate” and found that the “strong-FTR” countries in the OECD have a murder rate that is about 60 per cent higher than that of the “weak-FTR” countries. I guess Chen could argue that having a future tense makes the possible punishment seem more distant, so you are more prone to kill your neighbor. What it is about the future tense that makes you believe in God is a bit harder to pinpoint. All kidding aside, it does appear that we are indeed dealing with bundles of cultural features that have developed and spread together without there being an general underlying principle.
Chen assures us that he has controlled for “spatial variables” and factors such as religion, and that they do not change his results. It is hard for me to see how the fact that virtually all “weak-FTR” countries come from one specific region with rather special properties can fail to bias the results rather seriously. It may be that spatial variables such as distance from the Equator and religion in the sense of the traditional religion (as opposed to present-day degree of secularization) of the country in question does not single out the region referred to as “Protestant Europe”. In addition, I wonder what the effect is of the fact that out of the ten “weak-FTR” countries, five are small Scandinavian countries with a total population of around 25 million.
What the above does not account for is Chen’s probably strongest card, the intra-country correlations between grammatical marking and future-related behavior. However, as noted above, there are some unclarities in the data there, and it should also be pointed out that apparently the results shown in Chen’s Figure 4 are not significant except in one case, Switzerland – where the “Röstigraben” (the border between the German and the Romance speaking areas) also delineates the northern European “futureless” zone. (Chen says diplomatically that “small samples reduce statistical signiﬁcance”.) Maybe the discussion can continue when we get a clearer account of the data.
Dahl, Östen. 1985. Tense and Aspect Systems. Oxford: Basil Blackwell.
Dahl, Östen, ed. 2000a. Tense and Aspect in the Languages of Europe. Berlin: Mouton de Gruyter.
Dahl, Östen. 2000b. “The Grammar of Future Time Reference in European Languages”. In Dahl 2000a, 309-328.
Thieroff, Rolf. 2000. “On the Areal Distribution of Tense-Aspect Categories in Europe”. In Dahl 2000a, 227-264.
(1) Note on Chen’s quoted definitions: Chen is sometimes not quite literal in his rendering of the EUROTYP terminology. He says (fn. 3): “Dahl deﬁnes “futureless” languages as those which do not require “the obligatory use [of grammaticalized future-time reference] in (main clause) prediction-based contexts”.” In fact, I do not use the term “futureless language”, I only speak of the “futureless area” in northern Europe; the definition he quotes is given as a criterion for the grammaticalization of a marker. In the same footnote he says “I adopt Thieroﬀ’s more neutral language of “weak-FTR” for “futureless” languages”. But Thieroff 2000 does not use the expression “weak-FTR”; rather he talks of “languages with a weakly grammaticalized future”.
Update: a missing and somewhat crucial negation has been added to the above.