Monday 28 July 2014

(stereo) type your questions into Google

click to see this better!

While searching for information on the web, Paul Baker and Amanda Potts noticed that Google’s auto completion algorithm was inadvertently reproducing stereotypes. Troubled by this phenomenon, they set out to investigate which social groups elicited more stereotyping questions than others and how these differed in nature.

In 2010 Google added auto-completion algorithms to offer a list of suggestions when users type words into the search box. The predicted text shown in a drop-down list has either been entered by previous users or appears on the web. While this process can save time, it also has some unintended consequences. For example, when one types ‘why do gay’ into the search box, the information in the picture above appears.

These suggestions are stereotypical; they ascribe characteristics to people on the basis of their group membership, reducing them to certain (often exaggerated) traits.

To carry out their study, the researchers first created a list of identity groups to investigate. They found that the terms which produced the most questions were related to ethnicity, gender, sexuality and religion. The groups eventually chosen were: black, Asian, white, Muslim, Jewish, Christian, men, women, gay, lesbian and straight. In each category, a selection of similar terms was used, so for example the male category included men, boys and guys. The researchers also chose people as a control group to show how humans are characterised when they are not associated with particular identities.

Next, they paired these terms with question forms in order to elicit auto-suggestions. The question forms included items such as why do, how do, where do as well as questions beginning with do, should and are. Each question was entered individually and the top suggestions were recorded.
Some of the elicited questions did not refer to social groups and were excluded. Finally, 2690 queries were analysed and it was found that the groups which elicited the most queries were Jewish, Black  and Muslim with over 300 results each, whereas People only elicited 70 and Lesbian a mere 41 queries.
The questions were then divided into the following categories:

Each question was also rated with regards to evaluation, and was classed as positive, negative or neutral. While the majority of the questions were classed as neutral, most groups tended to have more of their questions categorised as negative than positive. Surprisingly, the control category people elicited proportionally the most negative questions, which tended to be about why people engage in hurtful behaviours such as bullying and self-harming.

The relatively high proportion of negative questions for three groups was particularly concerning. For black people, these involved constructions of them as lazy, criminal, cheating, under-achieving and suffering from various conditions such as fibroids. Gay people were constructed as contracting AIDS, going to hell, not deserving equal rights or talking like girls. The negative questions for males positioned them as catching thrush, under-achieving and treating females poorly.

Conversely, all ethnic groups also elicited many positive questions. Black people were constructed as stronger, more attractive and virile. Asians were viewed as smart, slim and attractive, while white people were viewed as attractive and ‘ruling’ other groups.

In general, race and gender searches elicited questions concerned with the level of interest of one group for another. Indeed, top results for both genders featured the opposite sex. However, sexual fulfilment and references to orgasm appeared more frequently for the female questions.

The straight category included fewer questions (perhaps as this is seen as the ‘norm’). These tended to be about whether straight men enjoyed stereotypically gay entertainment such as TV show Glee or singer Cher, and whether straight men could ‘turn’ gay or have homosexual thoughts.

Whereas the Gay results included questions about whether gay people should be allowed to marry, adopt, join the military or give blood, Lesbian questions included negative stereotypes such as acting/looking like men, questions about sexual and emotional behaviour towards men and the mechanics of lesbian relationships.

While the researchers do not claim that people who are exposed to such questions will be influenced by the stereotypes they encounter, it is important to acknowledge that these do exist. In addition, this paper raises the moral question of whether content-providers should ‘protect’ their users and remove offensive auto-suggestions (and indeed who decides what is inappropriate?), or should they simply reflect the phenomena that people are interested in?

Baker and Potts warn that auto-complete suggestions could perpetuate stereotypes, and that not all people would view them with a critical eye. They therefore recommend that there should be a facility to flag certain auto-completion suggestions as problematic, and that Google should consider removing those that are consistently flagged.

Baker, Paul & Amanda Potts (2013) ‘Why do white people have thin lips?’ Google and the perpetuation of stereotypes via auto-complete search forms, Critical Discourse Studies,10:2, 187-204


This summary was written by Danniella Samos

Tuesday 8 July 2014

Accent on accent

image of Language Diversity from Tobias Mikkelson

"It is impossible for an Englishman to open his mouth without making some other Englishman hate or despise him" wrote George Bernard Shaw in his preface to Pygmalion. He was, of course, referring to the way people evaluate accents and make (usually negative) judgements about speakers.

When we talk about accent, it is important to remember that this relates only to pronunciation and intonation rather than grammar or vocabulary. Thus, two people speaking the same language, who use the same grammar and word choices will give different cues about their social and regional origins, ethnic group membership or class. While we, as listeners, naturally pick up these cues about people’s ethnic, socioeconomic and geographical background, experimental research has shown that listeners can also make judgements on others’ intelligence, warmth and even height just by listening to recorded accented speech.

Although Shaw was referring primarily to British English accents in a class-oriented society, many studies from the UK, USA and Australia from the past six decades all show that foreign accented speech is negatively evaluated by native speakers of a language. People who view their own group or culture as the centre of everything, and who scale and rate all other groups with reference to it, can be said to be ethnocentric. Ethnocentric people tend to strongly identify with people in their own group and are biased against outsiders.

James Neuliep and Kendall Speten-Hansen hypothesised that there would be significant negative correlations between ethnocentrism and the way speakers with non-native accents are socially perceived. To test this, they recruited 93 male and female undergraduate students and randomly assigned 46 of them to an experimental group and 47 to a control group. All participants were native speakers of English.

In the experiment, participants in both groups completed a Generalized Ethnocentrism scale test to see how ethnocentric they were. This included rating on a 5-point scale items such as ‘‘My culture should be the role model for the world’’ and ‘‘I have little respect for the values and customs of other cultures’’.

Then, both groups watched a video of the same male speaker talking for 12 minutes about a non-controversial topic – the benefits of exercise. The videos were identical in every way except for the accent of the speaker. In the film viewed by the experimental group, the speaker had a non-native accent, and in order to try and reduce stereotypical judgments, this accent was left ambiguous, with no detectable regional, ethnic, or national associations. The speaker viewed by the control group spoke with a standard American accent. In both videos, the speaker wore the same clothes, spoke in the same location, at the same pace, and used the same number of gestures.

After viewing the video, participants were asked to complete tasks designed to assess their perception of how attractive the speaker was, and how credible and like themselves he was.
To rate speaker credibility, participants used 7-point semantic scales including scales asking about expertise (for example, a scale between “Qualified” and “Unqualified”) and character (for example, scales such as ‘‘Reliable-Unreliable’’ and ‘‘Honest-Dishonest’’).

Speaker attractiveness was assessed using 7-point scales testing for social, physical, and task attraction. The latter refers to the perception that someone is competent, trained, and qualified to perform a job. To assess attractiveness, participants had to rate items such as ‘‘I think he could be a friend of mine’’, ‘‘I find him physically attractive’’ and ‘‘I have confidence in his ability to get the job done’’.

Perceived homophily (how like the participant the speaker was judged to be) was also assessed by a 7-point semantic scale, where participants were asked to rate items such as ‘‘The Speaker is: Like me-Unlike me’’ and ‘‘The Speaker is: Of similar status to me-of different status to me’’.

As predicted, for the experimental group ethnocentrism was negatively and significantly correlated with perceptions of the speaker’s physical, social, and task attractiveness, his credibility, and perceived homophily. Moreover, this bias is not binary but continuous; the more ethnocentric a participant was, the lower their ratings of the non-native accented speaker’s attractiveness, credibility, and homophily. However, when presented with a speaker with a standard American accent, ethnocentrism played little to no role in the way that the speaker was socially perceived.

The findings are important because previous research has shown that credibility, attractiveness and homophily are three of the most significant social perceptions we make about others, affecting the interpretation of what a person says.  What a non-native speaker is interpreted as saying, then, can depend on the ethnocentricity of their interlocutor.
Neuliep, James W. & Kendall M. Speten-Hansen (2013) The influence of ethnocentrism on social perceptions of nonnative accents. Language & Communication 33:167–176

This summary was written by Danniella Samos