Tikalon Header Blog Logo

Verbal Cues and Stereotypes

December 8, 2016

Recently, in response to an email message from a male acquaintance, I replied using the word "cute." His message contained a funny, photoshopped image; and, while it made me laugh, it didn't demand more than this one word acknowledgement. Just a few hours after that, I read a press release entitled, "Real men don't say 'cute'."[1] I decided to restore my manhood by using a modern tool that's mightier than a sword, a desktop computer, to write this article.

Cute little dog

Now that's cute!

Scientists are most familiar with quantitative analysis, but even they can admire the quality of "cuteness."

(Via Wikimedia Commons.)


It's generally easy to identify an educated person by his/her vocabulary. The original purpose of the vocabulary portion of the SAT test and other such tests was to assess how much a student has read as a measure of his intelligence. A diligent student will read many books, and he will find the definition of words he doesn't understand. This portion of the test was subverted over the years into a word memorization activity. Interestingly, a high score from successful memorization is also a good predictor of academic success, since the activity requires both intelligence and perseverance.

Aside from the level of education, what else might be gleaned from a person's speech? I've often been able to determine whether an author is male or female from their use of language; but, is such stereotyping more accurate than not? An international team of researchers, including one from the interestingly named, "Ubiquitous Knowledge Processing Lab," looked into this problem. Team members were from the University of Pennsylvania (Philadelphia, Pennsylvania), the Technische Universität Darmstadt (Darmstadt, Germany), and the University of Melbourne (Victoria, Australia).[2]

The research team, composed of social psychologists and computer scientists, used publicly available Twitter messages (tweets) to analyze the accuracy of stereotypes.[1] They used the artificial intelligence technique of natural language processing (NLP) to show that the spectrum of stereotype accuracy ranges from plausible to wrong.[1] NLP strives to create an automatic understanding of written language, and examples of NLP include the Siri intelligent personal assistant, and applications such as spell checking and predictive text.[1]

Participants in this psychological study were asked to categorize authors solely on the contents of their tweets. They identified the actual characteristics of 3,000 Twitter users and divided them into the following categories: male/female; liberal/conservative; younger/older; and no college degree/college degree/advanced degree.[2,3] A hundred of each Twitter user's tweets, randomly selected from all tweets authored in the course of a year, were placed on the Amazon Mechanical Turk crowd-sourcing website, where Internet users were paid two cents every time they matched a tweet to a category.[2,3]

The NLP methodology went beyond the way stereotype research had been done in the past, since it started with the behaviors and asked about identity, rather than start with a group and ask about behaviors.[1] In this way, NLP allowed insight into a person's stereotypes without asking them to explicitly state them. Says lead author, Jordan Carpenter,
"This is a novel way around the problem that people often resist openly stating their stereotypes, either because they want to present themselves as unbiased or because they're not consciously aware of all the stereotypes they use."[1]

Stereotype words - Female

Words written by men but characterized as "female words."

Color indicates the relative word frequency, while size indicates the strength of the correlation.

(Social Psychological and Personality Science, 2016, image.)


When the guesses were about age, gender, or politics, they were better than chance.[3] Guesses for education level, however, were worse than chance, possibly because there were three choices in that category rather than two. Gender guesses were 75% correct,[3] affirming my ability to accurately guess whether a writer was male or female. It seems that people think that highly-educated people never use the f-word, and this contributes to the errors in guessing education (I don't use the f-word, except when talking to myself). Says Carpenter,
"...People had a decent idea that people who didn't go to college are more likely to swear than people with PhDs, but they thought PhDs never swear, which is untrue."[1]

Stereotype words - Male

Words written by women but characterized as "male words."

Color indicates the relative word frequency, while size indicates the strength of the correlation.

(Social Psychological and Personality Science, 2016, image.)


While it is true that men author more technology related tweets than women, stereotyping resulted in the misconception that all technology tweeters were men.[1] Also, feminine-sounding people were assumed to be liberal, while masculine-sounding people were thought to be conservative.[1] It's interesting to think of how this information might be applied. Would articles about the validity of global warming convince conservatives if they were written in a highly masculine tone?

References:

  1. Real Men Don't Say "Cute," Society for Personality and Social Psychology Press Release, November 15, 2016.
  2. Jordan Carpenter, Daniel Preotiuc-Pietro, Lucie Flekova, Salvatore Giorgi, Courtney Hagan, Margaret L. Kern, Anneke E. K. Buffone, Lyle Ungar, and Martin E. P. Seligman, "Real Men Don’t Say "Cute" - Using Automatic Language Analysis to Isolate Inaccurate Aspects of Stereotypes," Social Psychological and Personality Science, Advanced Online Publication, November 15, 2016, doi: 10.1177/1948550616671998.
  3. Karl Leif Bates, "Would You Expect a 'Real Man' to Tweet 'Cute' or Not?" Duke Research Blog, November 16th, 2016.

Permanent Link to this article

Linked Keywords: Photo manipulation; photoshopped; laughter; laugh; press release; tool; The pen is mightier than the sword; desktop computer; scientist; quantification; quantitative; analysis; quality; cuteness; Wikimedia Commons; education; educated; vocabulary; SAT test; reading; read; intelligence; diligence; diligent; student; memorization; prediction; predictor; academic achievement; academic success; perseverance; author; male; female; language; stereotype; stereotyping; accuracy; research; researcher; Ubiquitous Knowledge Processing Lab; University of Pennsylvania (Philadelphia, Pennsylvania); Technische Universität Darmstadt (Darmstadt, Germany); University of Melbourne (Victoria, Australia); social psychology; social psychologist; computer scientist; Twitter messages (tweets); analysis; analyze; artificial intelligence; natural language processing (NLP); Siri; intelligent personal assistant; spell checker; spell checking; predictive text; human subject research; participant; psychology; psychological; author; Twitter; social liberalism; liberal; conservatism; conservative; undergraduate education; college degree; postgraduate education; advanced degree; randomness; random; Amazon Mechanical Turk; crowdsourcing; crowd-sourcing; Internet; penny; cent; methodology; behaviors; Jordan Carpenter; bias; unbiased; frequency; correlation; Social Psychological and Personality Science; hypothesis; guess; age; gender; politics; chance; f-word; intrapersonal communication; talking to myself; profanity; swear; Doctor of Philosophy; PhD; technology; information; global warming.