Language and economic behaviour: Future tense use causes less temp discounting

Previous studies have found cross-cultural correlations between linguistic obligations for talking about future events and economic decisions like saving money. The hypothesis is that a grammatical obligation to use the future tense (e.g. will) causes speakers to perceive future rewards as temporally distal and therefore less valuable (“temporal discounting”). However, no studies have tested whether speakers actually temporally discount as a function of the extent to which they use the future tense. We present two studies which use a novel language-elicitation paradigm to do this, involving speakers of English (which obliges the future tense) and Dutch (which does not). We used mediation analysis to test how language-level differences in the grammatical obligation to use the future tense impact economic decisions via individual language use habits. However, we found that English speakers who habitually make greater use of the future tense actually discount less, not more. These results suggest obligatory future tense use is not responsible for previously-reported cross-cultural correlations. Instead, we suggest that a better explanation involves modal notions of certainty (the probability of an event occurring) rather than temporal distance (when an event will occur). Future tenses express high certainty, which makes the correct prediction that obligatory tense marking should cause less discounting. In contrast, the cross-cultural differences may be driven by variation in other aspects of future time reference, such as low-certainty modal terminology (e.g. may, might).

The temporal hypothesis

Most research into this question has advanced a theoretical account we refer to as the temporal hypothesis [7]. This hypothesis is based on two observations. First, some languages oblige speakers to use the future tense for FTR, while others do not. For instance:

1. (1)
   1. English
      It will rain tomorrow.
   2. Dutch
      Morgen   regent   het.
      tomorrow   rain:PRS * it
      ‘Tomorrow it rains.’
      *The verb regenen ‘to rain’ is in the present tense (“PRS”), i.e. regent ‘It rains’. We use the Leipzig rules for interlinear glosses [8].

English speakers must use will, be going to, or shall in example (1a), whereas Dutch speakers are free to use the present tense (1b). Languages like English are referred to as “strong-FTR”, while languages like Dutch are referred to as “weak-FTR”. We refer to the distinction between weak- and strong-FTR as “FTR status.”

Specifically, weak-FTR languages are those languages which do not oblige the future tense in prediction-based contexts. Critically, example (1) is a prediction, and it is ungrammatical in English predictions to use the uninflected present tense, e.g. (*)The Cardinals win on Sunday. Weak-FTR languages are defined as those which permit such constructions for future predictions, and strong-FTR languages are any language which is not weak-FTR [7]. FTR strength is defined in terms of future predictions because it is common for languages, even strong-FTR languages, to permit the present tense in other common FTR contexts. The most important of these are schedules and intentions [9,10]. In fact, use of the present tense to refer to known future schedule is cross-linguistically universal, at least in European languages [10]. Taking such constructions into consideration might risk obviating any attempt at classifying language by FTR obligatorization. English is a good example: When referring to a schedule, the present tense is perfectly acceptable, e.g. The Bears play at 7pm. The same can be said of intentions, though in a more limited set of contexts, in particular near-term well-known future events, e.g. I’m away next weekend. As such, when we write of languages “obliging” the future tense, we follow Chen [7] in only referring to prediction-based FTR.

The second observation is that people tend to devalue (or “temporally discount”) outcomes as the time until they will occur grows longer [11,12]. For instance, given the choice of receiving $10 now or $10 in a month, most would choose the former because they would have temporally discounted the latter as a function of the delay. However, people have different time preferences: Give a choice between $10 now and $20 in a month and some will take $10 now, while others will choose to wait for $20 [13]. Temporal discounting is therefore usually measured by giving participants a series of binary choices like those above. Participants who temporal discount less tend to choose “future-oriented” options to wait for future rewards. Participants who discount more tend to choose “present-oriented” immediate rewards.

The temporal hypothesis predicts that weak-FTR speakers should temporally discount less than strong-FTR speakers, and therefore tend to be more future oriented. Two mechanisms are suggested to explain this [7]. Firstly, habitual use of the present tense for FTR in weak-FTR languages is suggested to “collapse” future time with present time, causing weak-FTR speakers to perceive future events as more temporally proximal than do strong-FTR speakers. The reasoning is that the present tense encodes notions of immediacy; for instance Chen [7] points out that it is a common literary practice to use the present tense to heighten the salience of past action. Conversely the future tense is suggested to encode notions of temporal distance, presumably because it is used to refer to temporally distal future events [7]. Obligatory use of the future tense is therefore hypothesized to cause speakers to construe future events are more subjectively distal than speakers of languages whose grammar permits the present tense to be used for FTR. This entails relatively increased discounting. We refer to this as the “distance mechanism” [7]. The other mechanism is that languages which have a future tense, “slice” time up into three segments– past, present, future— whereas languages which do not have a present tense only encode two segments— past, present + future. This is suggested to cause speakers of languages with a future tense to have more precise beliefs about the temporal location of future events, which would also lead to increased discounting [7]. We refer to this as the precision mechanism. Both mechanisms assume the future tense encodes temporal notions (distance, precision) and both predict that obligatory future tense use in strong-FTR languages will cause speakers to devalue future events (discount) relative to weak-FTR speakers. Together, we refer to these mechanisms as the temporal hypothesis.

Empirical support for the temporal hypothesis

In the contribution which first proposed and tested the temporal hypothesis, [7] classed N = 129 languages in terms of FTR status. He then used FTR status to predict a set of behaviours which are plausibly downstream of temporal discounting. He found weak-FTR speakers were more likely to have saved money each year, retired with more assets, were less likely to have smoked, and more likely to practice safe sex. Additionally, they were healthier, as indexed by obesity, peak blood flow, grip strength, and physical exercise levels. Together, these findings indicate weak-FTR speakers prioritise future outcomes to a greater extent than strong-FTR speakers, suggesting they temporally discount less.

A number of studies have applied [7] approach to new outcomes. German (weak-FTR) speaking households were found to save more, smoke less, and resort to credit less than French (strong-FTR) speaking households in Switzerland [14]. Additionally, firms headquartered in weak-FTR-speaking countries engaged in more future-oriented behaviour [15], a pattern of results which has replicated across a variety of contexts [16–18]. Weak-FTR-speaking countries engaged in more future-oriented management of their economies [19,20]. Mandarin (weak-FTR) speakers had more pro-environmental attitudes than Korean speakers (strong-FTR) [21]; bilinguals expressed higher support for a fictitious pro-environmental “green tax” when they were tested in Estonian (weak-FTR) than Russian (strong-FTR) [22]; weak-FTR speakers (across a number of languages) were more likely to adopt environmentally responsible behaviours [23]; and weak-FTR-speaking countries enacted more pro-environmental policies and exhibited stronger commitment to global environmental cooperation [23]. Such decisions are compatible with an intertemporal interpretation, because climate change mitigation efforts tend to incur short-term costs but confer long-term benefits. In perhaps the strongest evidence to date, [24] assigned bilingual participants who spoke both a weak and strong-FTR language from 12 language pairs to complete an intertemporal choice task; they found speakers addressed in weak-FTR languages discounted less, as predicted by the temporal account, and that the variance in future-distance ratings was greater for weak-FTR speakers— the first empirical support for the precision mechanism.

Problems with the temporal hypothesis

However, there are number of results which suggest more research is needed to understand the relationship between FTR grammaticisation in intertemporal decision making. Several studies have found future marking predicts pro-environmental behaviour (the opposite to the direction predicted by the temporal hypothesis). For instance, Germans given a pro-environmental choice task in the future tense (as compared with the present) were more likely to say they would invest money to plant trees [25]; across individual and national data, speakers of languages with inflectional FTR marking were more likely to perceive climate change as serious and support mitigation action [26]; and countries which spoke language with mandatory inflectional FTR marking had better climate change mitigation performance as measured by CO₂ per capita, energy per capita, and energy per GPD [26]. These conflicting results suggest the temporal mechanisms proposed to relate FTR marking to future-orientation are not fully understood. This conclusion is supported by a study which found no support for the distance mechanism; in fact, weak-FTR speakers construed future events are more distal, the opposite to predicted direction [24]. This replicates our previous research, which found Dutch (weak-FTR) speakers rated future outcomes as more distal than English (strong-FTR) speakers [27, pp. 98–99]. This reverses the direction predicted by the temporal hypothesis. Taken together, these results suggest that the future tense does not encode temporal distance, cf. [7]. The other mechanism in the temporal hypothesis is that the future tense encodes temporal precision. This has found recent support, when variances were found to be greater when bilingual participants speaking weak-FTR language were asked to locate perceived distance using a slider [24]. However, if people speaking weak-FTR languages did represent future events less precisely, rewards marked by the future tense should be construed as less valuable, e.g. (“you get $10 in a week”) versus (“you will get $10 in a week”). However, no such pattern emerges. At least eight independent experiments have failed to find significant differences as a function of whether intertemporal choices were present tense or future tense [28–30]. This suggests the future tense choices are not construed as more temporally distal, or more temporally precise, since either mechanism would lead to increased discounting. Our previous research supports this conclusion: When participants used a slider to estimate the temporal distance encoded by a simple expression, no difference emerged as a function of whether the expression was framed in the future or present tense [, pp. 98–99]. Taken together these results suggest that the future tense does not encode notions involving temporal distance or precision, relative to present tense FTR, cf. [7].

If future tenses do not encode temporal distance relative to the present tense, what do they encode? In a previous study we investigated whether the future tense in English and Dutch encodes modal notions of high certainty [31]. Participants used a slider to rate future marked sentences between high and low-certainty. In both languages, participants rated future tenses as encoding high-certainty (%70-%80) [31]. This is perhaps un-surprising. Many contemporary scholars argue that will is not a tense at all, but rather part of English’s modal verb system [32–35]; similar arguments have been made about zullen [36]. Other approaches suggest will’s semantics are a mixture of temporal and modal [37,38]. While some emphasize will’s role in marking predictions as evidence it is a weak modal of low-certainty [35], others suggest it encodes modal strength (high-certainty) [33,39]. To understand these debates, it is necessary to expand the discussion beyond tenses strictly to the modal verb systems of which will and zullen are constituents. Paradigmatic analyses are a useful tool in this regard. Paradigmatic analyses attempt to understand the semantic contribution of a linguistic unit by contrasting it with the other units by which it may be replaced in a given context [40]. For instance, the future tense in English and Dutch can be formed with either a modal verb (English will/shall; Dutch zullen ‘will’) or a “go-based”, construction (English be going to/gonna; Dutch gaan ‘be going to’). These are constituents of FTR systems comprising other modals which— at least among those which are permitted in prediction-based FTR— encode modal weakening relative to the future tense markers (English could, may, might, should, would; Dutch kunnen ‘may’) [41]. This modal weakening was present in our previous study: When participants rated the certainty encoded by modally weak verbs like may, it was approximately %30– %40 [31]. A paradigmatic analysis therefore suggests that will and zullen encode high-certainty notions relative to the other modals, e.g. De beren zullen vanavond winnen ‘The Bears will win tonight’ expresses high certainty relative to De Beren kunnen vanavond winnen ‘The Bears might win tonight’. This interpretation was supported by our previous study [31]. For these reasons we conclude that a plausible interpretation of FTR semantics is that future tenses encode high-certainty FTR.

Hypotheses

We hypothesize that obligatory use of the future tense would cause strong-FTR speakers to construe future outcomes as more certain (less risky). We further hypothesise this would cause them to discount less not more. cf. [7]. Recent accounts treat modal expressions as scalar operators which map transparently onto notions of probability. Rather than Boolean quantification over possible worlds, modal semantics involve encoding the likelihood of events on a one-dimensional scale between high (p = 1) and low (p = .5) certainty [42–44] (p = .5 for low-certainty because p = 0 indicates high certainty in the non-occurrence of some event). Evidence suggests scalar accounts capture modal semantics better than notions of Boolean quantification, since the latter yields incorrect predictions in some linguistic contexts [42]. Thus, future tenses plausibly encode high certainty, this maps onto probability, and people discount as a function of probability as well as time [13,45,46]. Imagine someone has a belief F about some future probability p. The obligation to use high-certainty future tenses would lead to high-probability construals of future events in strong-FTR speakers , where indicates we are strictly referring to future probabilities. Since people discount as a function of probability as well as time [13,45,46], this would lead directly to decreased temporal discounting.

This means Chen’s (2013) temporal account and our modal account have different predictions about how obligatory future tense use should impact temporal discounting. Specifically, we argue that the future tense encodes high-certainty and that therefore strong-FTR speakers will construe future events in high-certainty terms (i.e. ) and that this will lead to decreased discounting. We refer to this as the “high-certainty” hypothesis. Conversely, [7] argues that the future tense encodes temporal distance or temporal precision. In the first case, strong-FTR speakers construe future events as more distant (i.e. ), in the second, strong-FTR speakers construe future events with more temporal precision (i.e. ). Either of these differences in belief lead to increased discounting [, conditional on some assumptions]. Resolving which of these is the case is an empirical question: Does increased use of the future tense in strong-FTR languages lead to increased discounting [7] or does it lead to decreased discounting? Studies 1 & 2 attempt to answer this.

Methods

Results

English speakers used more future and fewer present tense constructions, Table 1, which supports the classification of English as strong- and Dutch as weak-FTR. Future tense use was negatively correlated with discounting in English but not Dutch, Table 2. This suggests the English future tense encodes high certainty rather than temporal distance.

Discussion

The principal result of Study 1 is that obligatory future tense use in our strong-FTR language (English) resulted in less rather than more discounting. This suggests the English future tense encodes high certainty, not temporal distance or precision, cf. [7]. This supports the high-certainty and not the temporal account.

Methods

Results

Replicating Study 2, future tense use was negatively correlated with discounting in English but not in Dutch, Table 2. This suggests that higher future tense use in English would again result in less discounting, as the modal hypothesis predicts.

Discussion

In Study 2— as in Study 1— higher use of the future tense in English resulted in less, not more, temporal discounting. This suggests that higher English future tense use is not causally implicated in higher English discounting, as is hypothesised by the temporal account. This entails that other factors are responsible. A plausible account is that time preferences are affected by constraints in English that oblige low-certainty modal FTR— e.g. could, may, or might. We address this idea below.

What is responsible for observed discounting in strong-FTR speakers?

To the extent they may generalise, the negative indirect effects of future tense use undermines the temporal mechanisms hypothesised in [7]. What is to be made of findings that FTR status is a consistent predictor of intertemporal behaviour [7,14–22,30,60–67]? On the one hand, correlational research involving linguistic predictors is notorious for producing spurious relationships due to non-independence between languages resulting from historical processes of cultural inheritance which see causally unrelated bundles of traits being passed on from antecedent to descendant cultures [68]. When linguistic history controls are added, results have been mixed [69], or have failed to replicate [63]. Reported correlations may be spurious. On the other hand, FTR grammaticisation could be impacting temporal discounting, just not via the tense-based temporal mechanisms hypothesised in Chen [7]. In this case, new theoretical accounts would need to be developed to explain the observed effects of FTR status.

One possibility is that strong-FTR languages oblige the use of low-certainty modal constructions for FTR. English is a good example. When English speakers make predictions, it is not obligatory to use the future tense, cf. [7]. What is obligatory is a modal verb. For instance, in example (1), it is acceptable to say any of It can/could/may/might/shall/should/will/would/is going to rain tomorrow. With the exception of the future tense constructions (shall, be going to), the other modals generally encode modally weak notions of low certainty. In a previous study using a similar FTR-elicitation task, we found English speakers used more low-certainty FTR constructions than Dutch speakers [, pp. 86–95]. This indicates that in English strong-FTR grammar obliges speakers to use low-certainty modals to talk about the future. Could this cause them to construe future events as more risky?

If this were the case, it could explain effects of FTR status. A crucial point is that risk may be an overarching confounding factor in many outcomes involving real-world behaviour assumed to be impacted by time preferences. Most real-world intertemporal decisions involve some degree of risk. For instance, someone deciding whether to smoke a cigarette has to balance the present pleasure against future costs as a function of delay and probability, i.e. when and if she will develop smoking-related health issues. Many of the outcomes predicted by FTR status involve real-world intertemporal decision making involving risk. If strong-FTR speakers construe future outcomes as riskier, effects of FTR status might result from a difference in risk preferences, i.e. (which is the opposite direction to the high-certainty hypothesis we have outlined above). If the obligation to use low-certainty language about the future resulted in such beliefs, strong-FTR speakers would discount more not less than weak-FTR speakers. This is, in fact, what has repeatedly been found, [7,14–22,30,60–67]. Since we have demonstrated that Chen’s (2013) proposed mechanisms are not supported — at least in English and Dutch— we posit that this “modal” account may be responsible for observed findings.

Limitations

There are several limitations to the present study which should be discussed. First, in Study 2, the participants were recruited from different platforms (English from Prolific, Dutch from Qualtrics). This was done for practical reasons (the Dutch population on Prolific is limited and due to funding constraints, we were unable to use Qualtrics, which is more costly, for both samples). It is possible differences in recruiting methods drove results. We consider this to be unlikely since we found the same effects in Study 1, which recruited participants from different populations than Study 2 (English from Oxford students and Prolific, and Dutch from the MPI). Replicating findings across different recruitment platforms suggests participant sampling methods were not a determinant of outcomes in either study. Second, data in Study 2 were collected at different times for the different samples (Dutch: 2018; English: 2020). It could be the case that global events drove observed findings (e.g., COVID-19, inflation). We suggest this is unlikely on the grounds that Study 2 replicated Study 1, which recruited both language samples contemporaneously in 2017. Additionally, whilst exhibiting some state-like characteristics, delay discounting has strong trait-like characteristics and appears to be relatively stable across time, context, and outcome [70]. Third, the values in the intertemporal choice task may not be representative of real world situations, and necessarily do not represent a full-spectrum comparison of English and Dutch speakers across all reasonable reward values. This is important since reward value does change discounting function curves [71]. More research would be needed with larger reward values to better understand the range of values for which the findings we report may replicate.

The Dutch sample should be discussed. Dutch speakers may use English frequently (especially the kind of population who participate in online surveys). If speaking a language with obligatory FTR marking affects intertemporal decision making, this could have made Dutch-English bilinguals more “English like” in their discounting behaviour. However, even if this were the case, this suggests the mediation analyses we report are conservative, since the Dutch sample would presumably discount more than the Dutch population. Additionally, Dutch does have a future tense (just not an obligatory one). Readers may notice that present:future tense ratios were less stable study to study, and exhibited higher variance, in Dutch than English, Table 1. This observation may give rise to reasonable concerns regarding Dutch’s status as a weak-FTR language. However, we point out that greater variance and less stability is exactly what you would expect in speakers of a language with a more weakly gramaticised future, where utterances are guided by idiosyncratic differences between speakers rather than strong grammatical constraints. Nonetheless, Dutch does have a future tense, which may make it difficult to extrapolate from Dutch to languages (like Finnish or Chinese) which do not. Similarly, English does not have an bound morphological future tense (like French or Spanish), which may make it difficult to extrapolate from English to other strong-FTR languages. In fact, both Dutch and English are closely related Western Germanic languages in which the future tense comprises part of a modal verb system. Our conclusion that the future tense may encode the modal notion of high-certainty should be seen in this context. More research is needed to understand how general the effects we report may be. However, we suggest building from our approach will be useful for future research. We have closely investigated the mechanisms that underpin relationships between FTR marking and discounting in a small sample of 2 languages. This has afforded a more granular understanding of the FTR systems in question, but necessarily sacrifices generalisability. However, we suggest that more close examinations of language-wise contrasts in FTR marking and intertemporal choice in small samples of languages will better advance future research than broad cross-linguistic correlational research which necessarily paints cross-linguistic differences in broad strokes.

Implications for the linguistic relativity hypothesis

The present results support the linguistic relativity hypothesis. We found that grammatical differences between languages caused differences in usage between speaking populations (English speakers used more future tense constructions). This in turn was associated with a difference in non-linguistic cognition: decreased discounting in English speakers relative to Dutch. A difference between populations does not necessarily constitute evidence for relativity, as factors other than language might be responsible. However, the mediation analyses were able to statistically quantify the relationship between certain grammatical constructions (the future tense) and discounting. This suggests language is causally implicated in observed differences in discounting, supporting the linguistic relativity hypothesis.