Tuesday, November 19, 2013

Reduce Bias In Analysis By Using A Second Language

In a world where the majority of analysts are bi- if not multi-lingual, the question of how language affects both the analytic process and analytic product is an important one. Emotion, language processing and cognitive biases aside, the intriguing question remains: Would you make the same decision in English as you would in, say, Chinese? Most analysts would likely answer yes to this question, but recent research led by Boaz Keysar out of the University of Chicago suggests otherwise[1].

Thinking of learning another language? This infographic is a good
suggestion of which language you might want to consider
tackling first.
The study, published in Psychological Science, concludes that “people are not as loss averse in a foreign language as they are in their native tongue.” Being less loss averse, that is more willing to take on risk, might sound like a dangerous characteristic to possess from an intel analyst’s perspective. In this case, however, being less risk averse means that people more systematically assessed the problem and came to a more rational conclusion. At the root of this finding is the conclusion that “people rely more on systematic processes…when making decisions in a foreign language.” Regardless of how accepting of risk we are as analysts, the ability to make decisions driven more by rational thought and less by emotion is a capability to which every analyst likely aspires.

In three studies, Keysar showed that while participants made different decisions based on how the problem was framed[2] (as more or less risky), they made the same decision for both risk conditions when using their foreign language. The three groups of participants had English as a first language and Japanese as a second, Korean as a first language and English as a second or English as a first language and French as a second, indicating that this effect is replicable within and across language family boundaries.

So why, then, do we make more rational, less biased decisions in our second language than in our first? It largely has to do with the lack of “emotional resonance” that we derive from foreign language text. Literature on second language acquisition unanimously agrees that people perceive messages delivered in their second language as less emotional (and consequently less impactful) than messages delivered in their first language; this concept applies to everything from political opinion to curse words[3].

How we perceive emotion then ties directly to our internal cognitive processes. According to Daniel Kahneman, the most widely respected authority on these internal processes, we have two broad systems of thinking – System 1 and System 2 thinking[4]. System 1 is automatic (and often times uncontrollable) while System 2 is more deliberate and rational. Think of System 1 as the mechanism driving impulse buys and split second decisions, whereas System 2 is more like making a grocery list in advance. Cognitive biases, or internal heuristics (shortcuts) that influence both our analytic process and analytic product, originate in System 1 thinking. Examples of cognitive biases particularly relevant to intelligence analysis are confirmation bias, anchoring bias and the framing effect (addressed directly in Keysar’s article).

Cognitive biases originate in System 1 thinking along with our gut instincts, emotional reactions and a less credible substantiation for intelligence analysis, intuition. Consequently, it makes sense to pursue analysis derived from System 2 processes as it will likely be less biased, more rational and more systematically attained. The argument here is that conducting analysis within the domain of a second, third or fourth language will lead to an increased reliance on System 2 processes, thereby reducing bias and ultimately resulting in more systematically-derived analysis. The results of Keysar’s study, while still relatively new, support this perspective.

In practice, with bilingualism now practically a pre-requisite for analysis work, the benefit of this argument to intelligence analysts is obvious (coupled with the other known benefits of bilingualism).[5] The traditional view is that an analyst is at an automatic disadvantage when operating in a non-native linguistic domain to conduct analysis, fearing the loss of meaning and context. The argument in this article, however, sheds new light on the quality of the analytic product obtained in a non-native language. Would you make the same decision in English as you would in, say, Chinese? The answer is that you might not, and your Chinese decision just might be more impartial.

[1] Keysar, B., & Hayakawa, S.L. (2012). The Foreign-Language Effect: Thinking in a Foreign Tongue Reduces Decision Biases. Psychological Science, 23, 661-668.
[2] Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263-292. The bias phenomenon Keysar’s article claims to neutralize is what Kahneman and Tversky call The Framing Effect, and is one of the many known cognitive biases to affect intelligence analysis.
[3] Emotion and Lying in a Non-Native Language (2009). International Journal of Psychophysiology, 71, 193-204. Puntoni, S., Langhe, B. D., & Van Osselaer, S. M.J. (2009). Bilingualism and the Emotional Intensity of Advertising Language. Journal of Consumer Research, 35(6), 1012-1025.
[4] Kahneman, D., & Frederick, S. (2002). Representativeness Revisited: Attribute Substitution in Intuitive Judgment. Heuristics and Biases: The Psychology of Intuitive Judgment, 49-81.
[5] Bialystock, E. (2011). Reshaping the mind: The benefits of bilingualism. Canadian journal of experimental psychology, 65(4), 229-235. Though there are many studies that demonstrate the known benefits of bilingualism, this is a recent article that reviews many of these previous articles.  


douglas said...

Do you suppose that in at least some cases, you might risk losing the emotionally connected element that may be important in reading human subjects? If you are attempting to analyze information in the interest of predicting their actions- they aren't going to be thinking in a foreign language- and are therefore less likely to act rationally.

Melonie Richey said...

To your question, I think this would be one of the major downfalls of working in a foreign language for a HUMINT analyst (if I am interpreting your question correctly). Certainly there would be cultural expressions, gestures, pauses etc. that a non-native speaker of a language might not pick up on in a native speaker.

More broadly, the question this vein of research inspires for me is how does analysis (forecasting accuracy) differ for the same analyst working on the same question when having received information in only their L1 as opposed to having received information in only their L2. In other words, does language matter in an analyst's forecast and if it does, how do we measure how much?

This study effectively shows that working in an L2 reduces bias, but what does that mean for forecasting accuracy?