Thinking in a foreign language distorts allocation of cognitive effort: Evidence from reasoning

Highlights

• Bilinguals show decreased accuracy when reasoning in a foreign language.

• This can be attributed to differences in sensitivity to logic and not to belief bias.

• We found weakened conflict detection for FL reasoners.

• We propose a model in which FL reasoners show weakened logical intuitions.

Abstract

Bilinguals, in their foreign language, are spared from several decision-making biases. We examined this “Foreign Language Effect” in the context of logical reasoning, in which reasoners are required to track the logical status of a syllogism, ignoring its believability. Across three experiments, we found the reverse Foreign Language Effect; foreign language reasoners are less able to evaluate the logical structure of syllogisms, but no less biased by their believability. One path to succeeding in reasoning tasks is always engaging in reflective processing. A more efficient strategy is metacognitively tracking whether belief-based intuitions conflict with logic-based intuitions and only reflecting when such conflict is present. We provide evidence that foreign language reasoners are less accurate because they struggle to detect belief-logic conflict, and in turn fail to engage in reflective processing when necessary to override the incorrect, intuitive response. We propose that foreign language reasoners are less able to detect belief-logic conflict either due to weakened intuitions or due to a more conservative threshold for the detection of conflict between multiple competing intuitions. Data for the experiments can be accessed publicly at https://osf.io/phbuq/

Introduction

Research on judgement and decision making suggests that bilinguals, when making decisions in their foreign language (FL), are protected from several of the biases they would otherwise be affected by in their native language (NL). For example, FL decision-makers tend to be less affected by the framing of the decision (Keysar, Hayakawa, & An, 2012), loss aversion (Costa, Foucart, Arnon, Aparici, & Apesteguia, 2014), hot hand fallacy (Gao, Zika, Rogers, & Thierry, 2015), and superstitious beliefs (Hadjichristidis, Geipel, & Surian, 2019). This change in susceptibility to decision-making biases is dubbed the Foreign Language Effect (FLE).

To this point, two main explanations have been proposed as to what mechanism is responsible for the FLE. The first of these explanations proposes that using a foreign language attenuates the emotional resonance elicited by a problem, as the foreign language is usually acquired in a low-emotion, classroom context. Therefore, FL decision-makers might not engage emotional processing as easily as NL decision-makers and in turn be less susceptible to emotionally-based biases. The second explanation proposes that using a foreign language promotes deliberation, as processing in a foreign language is less fluent. In this case, processing difficulties serve as a signal that more deliberative processes are needed and thus, the role of intuition is reduced (Costa, Vives, & Corey, 2017; Hayakawa, Costa, Foucart, & Keysar, 2016). Obviously, these mechanisms are not mutually exclusive, and the FLE could be a product of both mechanisms.

In a direct test of the emotional explanation, the FLE in moral judgments was not mediated by emotionality (Geipel, Hadjichristidis, & Surian, 2015), and emotionality was similar across groups that demonstrated multiple types of common decision-making fallacies (Miozzo et al., 2020). The alternative, cognitive explanation also fails to fully account for the effect: no language effects were observed in the cognitive reflection tests (Białek, Paruzel-Czachura, & Gawronski, 2019; Costa, Foucart, Amon, et al., 2014; Mækelæ & Pfuhl, 2019), nor in the deliberative use of statistical information in gambling decisions (Muda, Walker, Pieńkosz, Fugelsang, & Białek, 2020), or in intertemporal choice (Białek, Domurat, Paruzel-Czachura, & Muda, 2020). There is some hint that cognitive reflection, a trait of how individuals allocate their cognitive effort, interacts with the FLE. Specifically, the effects of cognitive reflection are only visible when deciding in one's native but not in one's foreign language (Białek et al., 2019, Białek et al., 2020).

Finally, recent research has complicated the understanding of the FLE, finding no effects in linguistically similar languages (Dylman & Champoux-Larsson, 2020; Mækelæ & Pfuhl, 2019), but observing a FLE in Italian local dialects (Miozzo et al., 2020). A report shows no effects of language in highly acculturated individuals (Cavar & Tytus, 2018, but see Białek & Fugelsang, 2019 for a commentary on why the evidence for this claim is weak). Hence, some linguistic and cultural factors certainly play a role in driving the FLE. Considering all the above, we lack a good explanation for the FLE and its cognitive mechanisms.

In the present work, we use a Dual-Process Theory framework to test the relative claims made by each of these mechanisms. In a Dual-Process Theory framework, it is assumed that cognition consists of two complementary processes: Type I processes are autonomous and do not require access to working memory, and Type II processes require access to working memory to cognitively decouple¹ and run mental simulations (Evans & Stanovich, 2013). The most prominent model of Dual-Process Theory proposes that, by default, people tend to use only Type I processes and engage in Type II reasoning when necessary (e.g., when they detect a conflict between two or more intuitive responses,² or if no intuitive response is readily available; Bago & De Neys, 2017; De Neys, 2014; De Neys & Pennycook, 2019; Handley & Trippas, 2015; Pennycook, Fugelsang, & Koehler, 2015; Šrol & De Neys, 2020; Stanovich, 2018; Thompson, Turner, & Pennycook, 2011; Trippas, Thompson, & Handley, 2017; Trippas & Handley, 2017). When reflecting, reasoners either try to investigate the logical status of the premises so that they can arrive at a valid conclusion, or they try to rationalize the most promising intuition (Pennycook et al., 2015). From this framework, one could postulate several possible mechanisms responsible for better performance on decision-making tasks. Perhaps individuals perform better when thinking in their FL because they, simply by encountering the problem in their FL, are forced to engage Type II processing. Alternatively, foreign-language reasoning could prompt conflict, either because reasoning in a foreign language produces the experience of conflict by default, or because reasoners produce no convincing intuitions, and the only route left is to solve the problem reflectively.

In this work, we investigate the foreign-language debiasing effect and its potential mechanisms. We focused on reasoning because the Dual-Process Theory originated and is most strongly understood in the domain of reasoning (De Neys, 2006; De Neys & Białek, 2017; Evans & Stanovich, 2013; Thompson & Johnson, 2014). In a typical reasoning task, participants are presented with logical syllogisms (e.g., “All A are B. All B are C. Therefore, all A are C”).³ Participants are asked to evaluate the conclusion of the syllogism, indicating whether the conclusion does (or does not) follow logically and necessarily from the premises, while ignoring whether the participant believes the premises and conclusions are true in the real world. This paradigm allows researchers to examine the extent to which participants base their responses on the logical structure of the syllogism (as syllogisms differ in terms of logical validity) and the extent to which this process is affected by feelings of belief or disbelief (as syllogisms can differ in terms of the believability of their conclusions⁴). In order to detect the effects of believability, syllogisms are counterbalanced in terms of their objective properties such that sometimes belief and logic cue the same response (congruent trials) and sometimes they cue opposite responses (incongruent trials). For example, a syllogism can be valid and believable (congruent trial, where responding along either dimension prompts a ‘yes’ response) or valid and unbelievable (incongruent trial, where responding along the logical dimension cues a ‘yes’ response, but responding along the belief dimension cues a ‘no’ response). Whether and when participants, when solving syllogisms, correctly classify them as congruent or incongruent is an empirical question we pursue in this investigation.

Evaluating conclusions on the basis of belief is assumed to be intuitive and a product of Type I processing, while evaluation of the logical structure of the syllogism, although perhaps accomplished via logical intuition, is usually assumed to require Type II engagement (Evans & Stanovich, 2013; Pennycook et al., 2015).⁵ Since one always first engages Type I processing and logic-based intuitions are usually weaker than belief-based intuitions, individuals typically consider the belief-based intuition first, and responding in line with logic then requires one to override the belief-based response with Type II processing. In other words, a response in line with believability suggests a failure to override a Type I response with a Type II response (Handley & Trippas, 2015).

The most dominant view in the field is that Type II responses are triggered by the detection of a conflict between multiple intuitions. More specifically, when assessing a syllogism, people automatically produce a Type I intuition related to the believability of the syllogism and a competing Type I intuition about the logical status of the syllogism. The belief-based, but not the logic-based intuitions are affective (Klauer & Singmann, 2013; Morsanyi & Handley, 2012). The belief-based intuition simply reflects one's assessment of the believability of the conclusion and premises. The logic-based intuition reflects one's assessment of the logical validity of a conclusion, and can reflect innate or learned logical rules, or some logical heuristic (e.g., the atmosphere heuristic: promoting affirmative conclusions if the premises are affirmative or promoting particular conclusions if any of the premises is particular rather than universal; Woodworth & Sells, 1935). This logical intuition has to be distinguished from effortful processing of Type II, which fleshes out the logical structure of the premises and only then computes the validity of the conclusion. Reasoners have access to the output of both Type I and Type II processes, but have access to the content of only the Type II processing. In other words, for intuitive (i.e., Type I) judgments, reasoners will have an intuitive sense of whether a conclusion is valid or invalid, but for reflective (i.e., Type II) judgments, they also know why they believe so. Type I logical intuitions and Type II logical processing are not identical in every way. However, problems typically used in reasoning research have been simplified so that both logics (Type I and Type II) must agree. As such, belief-logic congruence necessarily refers to the agreement between belief and both types (Type I and Type II) of logic.

If either the belief-based or logical intuition is much stronger than the other, no conflict will be detected and the output will simply be the dominant intuitive response (usually the belief-based one; De Neys, 2006; De Neys, 2014). If, however, these intuitions are of approximately equal strength, conflict will be detected, triggering Type II processing. In such cases, individuals will either attempt to override a Type I output with a Type II output or simply rationalize the most promising intuition (Handley & Trippas, 2015; Pennycook et al., 2015). To respond to each syllogism correctly, one strategy is to focus only on the structure of the syllogism, ignoring the believability of the conclusion. A more efficient system would intuitively estimate whether believability will cue a correct response (which is the case for congruent trials), or would not (which is the case for incongruent trials), and only engage in reflection for the latter type of syllogism. This minimizes the number of trials on which one must engage reflective, Type II processing without a proportional decline in accuracy (Stanovich, 2018).

One consequence of strong intuitions about believability is that logical validity judgments are biased by the believability of the syllogism (mostly cued by the believability of its conclusion, but also by its premises, see Solcz, 2011). This effect is labelled belief bias. This appears as a response bias: an overall tendency to endorse believable as opposed to unbelievable syllogisms as valid, regardless of their logical status (Dube, Rotello, & Heit, 2010). Because believability is usually assessed intuitively, belief bias is currently best thought of as an effect on Type I processing and not on Type II processing (Dube et al., 2010; Trippas et al., 2018; Trippas, Handley, & Verde, 2013).

The potential benefit of reasoning in one's foreign language is for two, non-exclusive reasons: people reflect more or are less biased by the believability of a conclusion. If FL reasoners experience an increase in reflective processing but not a decrease in intuitive processing, we should see in a syllogism task that they are more sensitive to the logical structure of the syllogism, but no more biased by its believability. If FL reasoners experience weaker affective resonance to the believability but are no more reflective, we should see in a syllogism task that they are no more sensitive to the logical structure of the syllogism, but less biased by its believability.

In three similarly designed experiments, we presented participants with 32 syllogisms, either in their native language or in their foreign language. Half of these syllogisms were valid, and half invalid. Half of each type had believable conclusions, and the other half unbelievable conclusions. As such, half of the syllogisms were congruent (i.e., logic and belief cue the same response) and half were incongruent (i.e., logic and belief cue different responses).

In all three experiments, we observed a decline in reasoning accuracy in one's FL. While still observing a FLE, ours is opposite to the direction typically reported. This decline in accuracy was driven by a decrease in the sensitivity to the logical structure of the syllogism, but not by greater reliance on the believability of the conclusion. This decrease in reasoning accuracy is caused by lower sensitivity to logical structure driven either by a lack of deliberation, or by distorted deliberation. Further analysis of our data revealed that NL reasoners were slower and less confident when they answered incongruent syllogisms incorrectly compared to answering congruent syllogisms correctly. Both of these responses are consistent with believability of a conclusion. This decrease in confidence suggests the detection of a logic-belief conflict, so that participants, even when not following logic, were still intuitively affected by it when responding. This conflict detection allows them to engage in reflection only when such conflict is strong enough, and results in reasonably high accuracy of reasoning. FL reasoners showed fewer signs of such conflict detection and accordingly failed to engage in deliberation when required. Altogether, we conclude that thinking in a foreign language disrupts conflict detection so that FL reasoners are less aware of when reflection is required. In turn, FL participants allocate their cognitive effort less accurately.