Language Log

« previous post | next post »

"Are you a different person when you speak a different language?" That's the headline of the press release, released from embargo on June 17, describing David Luna, Torsten Ringberg & Laura Peracchio, "One Individual, Two Identities", Journal of Consumer Research, August 2008.

The press headlines (not many, so far) echo the same idea: "How Switching Language Can Change Your Personality" (Reuters and New Scientist, published at ABC News); "Switching languages could cause you to switch personalities" (Discovery Reports, Canada); "Change in language alters personality" (IT Examiner, India — subhead "Oh, fickle woman"); "People switch personality with language" (Times of India);  "For bilinguals, a distinct personality for each language" (Agence France Presse).

The Times of India took this language = personality concept as the basis for an editorial, "Why not adopt American English?":

Many Indians consider American English infra dig. But it's time we got over this distaste. A recent survey has found that people unconsciously switch their personality when they change languages.

Since American English is by far the most dominant language today, anyone who wants to be a confident player in a globalised world has to speak the American lingo.

But in fact, as the press release and most of the articles explain, it's only bicultural individuals who were found to change their personality when changing languages (where "bicultural" means not identifying strongly with the dominant culture of either language). And it was only certain bilingual individuals who were studied: Hispanic-American women living in Milwaukee, Wisconsin. And it was only certain aspects of their personality that were measured: degree of self-sufficiency vs. other-dependence, along with some related gender-role associations. And (as the press release and the articles don't tell us) those aspects of their personality didn't change all that much.

Here's the paper's abstract:

Bicultural bilingual individuals have incorporated two cultures within themselves and speak the languages of those cultures. When cued by a particular language, these individuals activate distinct sets of culture-specific concepts, or mental frames, which include aspects of their identities. Three studies show that language-triggered frame switching (i.e., switching from one set of mental frames to another) occurs only with biculturals, not with bilinguals who are not bicultural. The studies uncover frame switching at the within-individual level, and they include both qualitative and experimental evidence. They also provide a methodology to identify the relative activation strength of specific mental frames in different languages.

Perhaps the most interesting thing was the direction of the shift:

Traditionally, popular stereotypes of the Hispanic and the Anglo cultures might lead to the assumption that the Spanish language would cue bicultural females to register as other-dependent and that the English language would cue them to register as self-sufficient (Martínez 1995; Vega 1990). However, some researchers suggest a recent and significant shift within the cultural fabric of Anglo and Hispanic female subcultures (see Martínez 1995; Vega 1990). Hispanic females have begun to express more self-sufficiency and Anglo females more other-dependence. Thus, while remnants of a machismo-based interaction are still present in various Hispanic social subgroups, these are increasingly considered a facade (Webster 1994). In fact, a large ethnographic study on gender and participatory democracy among urban Hispanics in California found that an overwhelming majority (84%) of Hispanic women support feminist goals (Takash 1993). Hispanic women are frequently and actively involved at the grassroots level, fighting for equal rights (Hardy-Fanta 1993; Takash 1993). It is noteworthy that independence and assertiveness seem to be dominant values emerging in the discourse of some female Hispanic groups (Farr 2005). In contrast, the social trend among Anglo women appears to have reverted toward traditionalism (e.g., moms as homemakers; Chandler 1999; Luttrell 1989; Ovadia 2001).

But in all three studies, the shift was small in quantitative terms — about 4-12% of the quantity measured.

I'll describe the studies, briefly, in order. In Study 1,

Fourteen female informants participated in two language-specific sessions set apart by 6 months. Only bicultural female informants were included, to eliminate potential differences based on gender and to enable us to focus on any language-triggered differences in our findings.

These were "first- or second-generation Hispanic Americans, 24-59 years of age, and living in a Midwestern city", presumably Milwaukee (where two of the authors are based). Although the subjects are called "bicultural", the authors also write that

We measured the degree of their identification with Hispanic culture with a five-item, seven-point scale: “I do not identify strongly with my ethnic group” (reverse coded); “I enjoy celebrating Hispanic/Latino cultural events”; “I think it's important to support activities that maintain our cultural heritage”; “If I had children, I would make sure they learn their cultural tradition”; and “In terms of your affiliation to the Anglo/Latino culture, how do you view yourself?” (very Latino/very Anglo; reverse coded). Higher scores denoted a higher identification with Hispanic culture. Scores ranged in the high end of the scale, between 5 and 7.

As we'll see, this is oddly at variance with the way they define "bicultural" in their other two studies. Anyhow, these 14 subjects completed "a reduced version of the Bem inventory", which is

… a 20-item scale including 10 “other-dependent” characteristics and 10 “self-sufficient” characteristics. Informants were asked to report whether those characteristics applied to them on a seven-point scale (1 = never or almost never true, and 7 = always or almost always true). Scores were added to form self-sufficiency and other-dependence indexes for each individual. Other-dependent (self-sufficient) characteristics suggest a more (less) traditional perception of a woman's role in the world. Composite scores were obtained by subtracting the other-dependent from the self-sufficient scores.

They don't tell us how these indices were calculated — were they just the sum of the 10 answers in each category, which is the literal interpretation of what they say? Or did they use the average? or some proportional measure?  This will obviously make a big difference in the interpretation of the result, which was that the M-score (difference between "self-sufficient" scores and "other-dependent" scores) was -0.19 when the women were interviewed in Spanish, but -0.54 when the women were interviewed in English, for an average difference of 0.35. (The difference was statistically significant, they tell us — but it's also important to know what its magnitude was.)

If we take them literally ("scores were added to form self-sufficiency and other-dependence indices"), then the sum for each index could have been as high as 70, and the difference as small as 0.5% of the total. But I believe that the usual way to report BSRI results is on a seven-point Likert scale, and on that basis, the average difference between the two conditions was 0.35 out of a possible 7, or 5% of the scale range. Actual average values tend to be around 5, so on that basis the effect of the language of the interview was to shift average Bem scores by about 7%.

We have to keep in mind that these are first- or second-generation Hispanic-American women, being interviewed either in English or in Spanish in a university setting, by "three male interviewers of similar age (25-35) … two of whom were biculturals". It seems to me that there are many things about this intercultural setting, besides the cultural frames associated with a language in some general sense, that might evoke feelings of self-sufficiency or other-dependence to different degrees for women of different cultural backgrounds.

The paper also provides some very striking quotations illustrating differences in the subjects' evaluation of evocative ads. Thus a woman sitting alone on a hill overlooking a lagoon was described by "Esperanza" in Spanish like this (only the English translation is given):

I think that she is strong. A woman likes to go out and travel. Women who are more feminine like to go to a nice hotel and have their nails done, and all that. There are other women who like to travel to places where they will have nice drinks, dance, and all those things, and then there are other women who like the fresh air, outside, camping-they like to go to other countries like Africa. She is a strong woman who does not like titles, who does not like to be told what a woman should be. She wants to be outside, likes to sleep outdoors, so she looks strong.

And then six months later, the same woman's reaction in English was this:

The woman in this ad is uptight and not really relaxed. She is hardworking, trying to survive. She needs to see what's next. She went there to sort out her problems.

I would be more persuaded by these quotations if the authors had published the complete interview transcripts, or at least done some blind coding of the totality of responses. Without that, it's hard to be sure that they are not cherry-picking examples to validate the point that they want to make.

In Study 2, the subjects were twenty eight "Hispanic female students at a large urban university", which was presumably the University of Wisconsin-Milwaukee. Their average age was 22, they were fluent in both Spanish and English. This time, the subjects were in the middle of the scale in terms of cultural affinities:

Level of acculturation was measured by an adapted version of Mendoza's Cultural Life Style Inventory (CLSI; 1989). This 10-item scale measures to what degree respondents have a greater affinity for either Anglo or Hispanic culture. The scale has five possible answers to a series of questions. For instance, one question asks to what extent respondents watch Spanish-language or English-language TV (1 = only Spanish; 2 = mostly Spanish; 3 = Spanish and English about equally; 4 = mostly English; 5 = only English). […] The mean acculturation score was M = 3.04, with a standard deviation of SD = .54; thus, respondents scored around the midpoint, suggesting that they were indeed biculturals.

The dependent measure was reaction time differences in an "Implicit Association Test" (IAT), where the subjects were asked to press a key to classify words or phrases, and the classificatory questions are disjunctions that pair two concepts in two different ways. In this case, on each trial the subject is asked to classify a word or phrase as "masculine" vs. "feminine", or else as "self-sufficient" vs. "other-dependent".  There are versions of this task. In the traditional task the category “masculine" appears in the same screen location as "self-sufficient", while the category "feminine" appears in the same location as "other-dependent". In the non-traditional task, the association of locations is reversed. (You can take a test of this kind yourself at the Project Implicit web site.)

Here are the words and phrases that they used, in the English-language version:

The basic reaction times for presentation in the different languages are not comparable, since the items have different lengths (a total of 135 syllables in Spanish vs. 86 in English) and presumably different word frequencies and so on. However, in a within-subjects design, it's fair to compare the difference between the "traditional" and "nontraditional" tasks when the items are presented in English, with the same difference for the Spanish-language version.

In English, the average response time to the traditional task was  M=858.50 milliseconds, and the average response time to the nontraditional task was M=1,025.83. In Spanish, the average response time to the traditional task was 866.82, and the average response time to the nontraditional task was 993.60 . Therefore, the difference in response times was 167.33 in English and 126.78 in Spanish … These findings provide evidence that the associations masculine/other-dependent and feminine/self-sufficient are stronger in Spanish than in English, relative to the associations masculine/self-sufficient and feminine/other-dependent.

Thus this difference of average differences was 40.55 milliseconds, or about 4.3% of the average reaction time of 936 milliseconds. And again, the effect was statistically significant, but not very large in magnitude — though you could argue that the "implicit association" effect was reduced by about 24%.

In Study 3, the authors wanted to compare bilingual women who are also bicultural, with bilingual women who are not. So they divided 93 bilingual women, average age 22, into three groups, based on the same CLSI self-assessment test used in study 2:

We assigned individuals who had average scores between 2 and 4 to the bicultural group, and individuals whose average score was smaller than 2 or greater than 4 to the monocultural Hispanic or the monocultural Anglo group, respectively.

Despite the average age, not all of the women were students: "Respondents held a variety of occupations, ranging from office managers to students. There was also a variety of national origins represented in our sample (e.g., Mexican, Puerto Rican)." (They don't tell us how many subjects were in each of the three groups.)

The sessions "began with an unrelated task to help activate the appropriate frame", and measured three things: the Bem masculinity-femininity score; a (self-reported) measure of assertiveness; and assignment of the Bem-inventory items to the masculine vs. feminine categories.

In the case of the Bem scores, the bicultural group scored 0.16 in the Spanish session, vs. -0.48 in the English session. The monocultural Anglos (-.40 vs. -0.7) and the monocultural Hispanics (-.80 vs. -.35) showed differences that were not statistically significant.

In the Bem categorization task, the subjects were asked to classify each word or phrase as either "masculine" or "feminine" or "both". Essentially all the subjects made essentially all the traditional associations, and a smaller number of non-traditional associations. But the bicultural group also made slightly more non-traditional associations in the Spanish-language version of the test — a difference of 0.57 vs. 0.43:

The monocultural groups had differences in the opposite direction that were not statistically significant: Anglos 0.53 vs. 0.58; Hispanics 0.52 vs. 0.60.

Likewise in the assertiveness assessment, members of the bicultural group scored themselves as about 12% more assertive when interviewed in Spanish (mean of 5.55 out of 7) than when interviewed in English (mean 4.91 out of 7); while the monolinguals showed no difference by language (Anglos 4.92 vs. 4.94; Hispanics 4.61 vs. 4.84).

Overall, it seems to the me that the study makes a persuasive case for its plausible conclusion, even though the effects seem to be fairly small in quantitative terms. However, I worry about a few loose ends. I've already mentioned my concern that the studies might be telling us more about the reactions (of a particular category of people) to (a particular type of) socially-charged interview situation, than about the effects of general cultural associations between words and concepts. Would the results of this experiment have been the same if the subjects had been interviewed in Mexico City or in San Juan? Or in a place where neither English nor Spanish is the dominant language? And would we get the same sort of results — personality-test differences for bicultural bilinguals but not for monocultural bilinguals — if we tested speakers of other language pairs in other situations — say speakers of French and German in Switzerland, or speakers of Mandarin and Cantonese in Hong Kong?

I'll mention three other things as well.

One is the issue of cultural diversity among Hispanics. Some subjects in this came from a Mexican background, while others had roots in Puerto Rican, and perhaps in other countries as well. These are by no means the same cultures. Indeed, among the Mexicans alone there could be a range of radically different cultural backgrounds. And these differences might well turn out to correlate with the monocultural/bicultural groupings — in which case, the different effects of language might not have anything to do with biculturality, but rather with specific cultural origins.

The second issue has to do with what it means to be bilingual: that term can cover a very wide range of relative abilities; and even apparent native fluency can mask basic differences in linguistic affinity. Anne Cutler et al. "Limits on bilingualism", Nature 340 229-230, 1989 "tested speakers who acquired two languages, French and English, in early childhood, still spoke both languages regularly, and were accepted by other native speakers of each language as native speakers". Nevertheless, they found that these people could be divided (by self-assessment) into "French-dominant" and "English-dominant" groups, whose performance on certain phonetic perception tasks was remarkably different. It's not clear that Luna et al.'s subjects were bilingual to the same degree as Cutler et al.'s subjects — it wouldn't be a surprise to find that the language abilities of their three culturally-defined groups were in fact different, as measured by much cruder methods than Cutler et al. used. But Cutler's results suggest that it's sometimes necessary to probe deeper than simple fluency. A similar conclusion is suggested by the fact that mere ability to read a second language easily is not enough to generate a Stroop effect as strong as in one's first language. This is not to say that cultural affiliation is not playing a role here, just that one can't dismiss the purely linguistic aspects simply by saying that everyone involved is in some sense "bilingual".

The third issue is the possibility that the gender of the Spanish version of the words on the Bem list might matter. Luna et al. don't give us this translated list, or discuss the grammatical gender of the words on it. But  as previously discussed here — "Sapir/Whorf: sex (pro) and space (anti)", 11/19/2003 — Lira Boroditsky has shown that speakers of European languages can be affected to a surprising degree by the (apparently arbitrary) grammatical gender of words. In her experiments, "Spanish and German speakers generated adjectives that were rated more masculine for items whose names were grammatically masculine in their native language than for items whose names were grammatically feminine". Thus

…the word for "key" is masculine in German and feminine is Spanish. German speakers described keys as hard, heavy, jaged, metal, serrated, and useful, while Spanish speakers said they were golden, intricate, little, lovely, shiny, and tiny.

But English, of course, doesn't have grammatical gender except in singular pronouns. Depending on the gender of the items in the Spanish version of the Bem list, this might create an effect that depends on grammatical gender rather than on personality characteristics. And it might actually be only the "bicultural" speakers who are actually bilingual enough to be affected by this difference.

[hat tip: Omri Ceren]

June 28, 2008 @ 1:16 pm · Filed by Mark Liberman under Language and advertising, Language and culture

Permalink