It’s not what you say, but how you say it. By analyzing online arguments, Cornell researchers have identified how language and interaction with the other party contribute to winning an argument.

Successful arguers start early, stay calm and go into detail.

As a spinoff, the research also offers linguistic clues that will show in advance whether your opponent’s mind can be changed.

An old Monty Python sketch has come to life: On “ChangeMyView,” a subcommunity of the Reddit social network, you can post an idea and invite others to argue against it (and you don’t have to pay for this argument).

“The Internet seems to be a comfortable place to put forward your ideas,” said Chenhao Tan, a graduate student in information science, explaining why people post to such a site. The site, moderated for civility, offers a mountain of data for linguistic analysis. If a particular challenger succeeds in changing the original poster’s (OP) mind, the OP awards that person a “Delta” (the Greek letter used in mathematics to represent a change). This allowed researchers to compare winners and losers.

Tan and colleagues collected posts from January 2013, when the community was created, to August 2015, involving 211,000 subscribers. They compared posts from challengers who received Deltas with those who didn’t when arguing on the same topic, and ran a computer analysis of the way the arguments proceeded and the language used. They reported their results at the 26th annual World Wide Web Conference in Montreal, April 11-15. Co-authors of the paper are graduate student Vlad Niculae; Lillian Lee, professor of computer science; and Cristian Danescu-Niculescu-Mizil, assistant professor of information science.

For starters, they found it pays to get there first. On average, out of 10 challengers, the first two are 10 times more likely to succeed than the 10th. And it pays to hang in there, at least for a while: Challengers who engaged in several back-and-forth exchanges with the OP were more likely to succeed, but if it went on to more than five rounds they had no chance of success. Overall, the winning arguments were longer and contained more sentences and more paragraphs.

During that exchange, they noted, successful challengers more often mimicked the linguistic style of the OP, as measured by comparing “stopwords” that are independent of content, like “about,” “only” or “often” along with prepositions, conjunctions and personal pronouns. These challengers also used more words relating to the content of the argument that were different from the content words used by the OP.

Then they compared the kinds of words challengers used in categories such as:

• calm (“librarian,” “forest”) vs. excited (“terrorism, “erection”);
• concrete (“hamburger”) vs. abstract (“justice”);
• dominant (“completion”) vs. weak (“dementia”); and
• pleasant (“sunshine, “love”) vs. unpleasant (“leukemia,” “murder”).

Information scientists who work with natural language have compiled standard dictionaries of words in such categories that the computer can use to classify the words it finds in a post.

The analysis showed that it was good to use calmer words, and successful arguments were “less happy.” Out of all this, the researchers concluded that dissimilarity with the wording of the original post was the best predictor of success. The study does not prove that any of these elements directly cause a change of mind, Lee pointed out; it merely identifies features that are common to successful arguments.

Finally, the researchers analyzed the text of the very first post presenting the topic, to see if it could predict the OP’s openness to change. The most easily persuaded posters wrote longer posts with lists and other supporting material, used more dominant words and more first-person-singular pronouns. The person who is less sure will argue harder, and is more likely to hold on to what “we” believe than what “I” believe.

To learn more about the mechanism of persuasion, the researchers said, they might extend their analysis into less orderly arguments. It might also be useful, they added, to learn how the rules of ChangeMyView could be used to build other better-behaved communities.

The research was supported by the National Science Foundation, Google and Facebook.

Secrets of successful arguers

A few examples of what has worked to change someone’s mind:

• Timing: The first person to reply to a statement has a greater chance of swinging the opponent’s view than someone who joins the debate later on. 
• Alternative terminology: Use words that are different from those used in the original proposal.
• Using “calm” language to make a point is more effective than swearing or using aggressive terms.
• Longer replies tend to be more persuasive. 
• Evidence: Using numbers, statistics and examples to back up opinions make you sound more convincing. Use “e.g.,”  “for instance” or “i.e” before presenting these arguments to strengthen their persuasiveness. 
• Links: Quotes and quotation marks play little role in trying to convince someone online, but linking to examples and outside sources does.
• Hedge your bets as in “It could be the case that….” Although this sounds like it might signal a weaker argument, it may make an argument easier to accept by softening its tone.
• Pick your battles:  Study the language used by your opponent to decide whether or not it's worth engaging in an argument. Personal pronouns such as “I” suggest a person is more open-minded to persuasion, but the use of the words “we” and “us” suggests they will be more stubborn. Stubborn people use more emotive and decisive words including “certain”, “nothing” and “best.” 
• Know when to give up: After four or five times back and forth, the chances of swaying someone's view significantly drops.