Language Style Matching: How the Smallest Words Predict Who Dates and Who Stays Together

By Noam Shemla · The Studies

TL;DR Two people who are in tune start to use the small words of language (pronouns, articles, prepositions) at similar rates, without noticing. This paper measured that “language style matching” in two ways: in 40 recorded speed dates, pairs who matched more were likelier to both want a second date; and in 86 couples’ instant messages, pairs who matched more were likelier to still be together three months later.

• Paper: Ireland, M. E., Slatcher, R. B., Eastwick, P. W., Scissors, L. E., Finkel, E. J., & Pennebaker, J. W. (2011). Language style matching predicts relationship initiation and stability. Psychological Science, 22(1), 39–44.
• Study 1: 40 transcribed four-minute speed dates (80 undergraduates, mean age 19.6) from events at Northwestern University.
• Study 2: 86 dating couples (mean age 19.0, together about 1.3 years on average) who shared ten days of their instant messages.
• Measure: Language style matching (LSM): how similarly two people use nine categories of function words, scored from 0 to 1 by the LIWC text-analysis program.
• Outcomes: A mutual “yes” to meeting again after the speed date; still dating three months later.
• Method: Logistic regression on standardised LSM, with perceived similarity (Study 1) or relationship satisfaction (Study 2) as controls.
• Headline: Each standard deviation of style matching tripled the odds of a mutual match and doubled the odds that a couple was still together.

Most of what we notice in a conversation is its content: what the other person said, what they meant, whether we agreed. Underneath that runs a layer we almost never notice: the small words. The, a, in, about, I, it, but, not, very. They make up a large share of everything we say, they carry almost no meaning on their own, and they are produced so fast that nobody chooses them consciously.

That makes them a useful window. James Pennebaker's lab at the University of Texas had shown that when two people are engaged with each other, their use of these function words starts to converge, and that the convergence is invisible to the speakers and to trained observers alike.12 This paper asked whether it could tell anything about romance.3

How style matching is measured

The measure is simple enough to write down in full. For each of nine categories of function words, take the share of each person's words that fall in that category, and compare the two:

The nine categories come from LIWC (Linguistic Inquiry and Word Count), the text-analysis program Pennebaker's group built:

The nine kinds of function word the score compares

The categories are LIWC 2007’s; the examples are ours.

The paper gives a feel for it with two real speed dates, one that ended in a mutual yes (overall LSM .77) and one that did not (.54). It reports the percentages behind three of the categories, which is enough to watch the formula work:

The formula on two real dates: three of the nine categories

Nothing in those numbers is about what the two people talked about. The matched pair could have been discussing anything. What they shared was the rhythm of their grammar.

Study 1: forty speed dates

The first study used speed-dating events at Northwestern University, where 187 students each went on up to twelve four-minute dates, all recorded. Forty dates were transcribed, chosen to cover a wide range of how well the dates went, with no person in more than one and nobody who already knew their date. On average each person said about 429 words.

Within a day, each person said privately whether they wanted to see their date again. A mutual yes counted as a match.

In the regression, each standard-deviation increase in style matching multiplied the odds of a match by about three (odds ratio 3.05). Two checks make the result more interesting than it first looks. First, the pairs' own sense of how similar they were, which is one of the strongest known predictors of attraction, was uncorrelated with their style matching (r = .06): the matching was picking up something people did not report. Second, style matching still predicted a match once perceived similarity was in the model, and once talkativeness was (with word count controlled, the odds ratio rose to 5.70).

Study 2: eighty-six couples

The second study turned to couples already dating, recruited for an earlier experiment on expressive writing. Each couple handed over ten days of the instant messages they sent each other, about a thousand words per person per period, and three months later the researchers asked whether they were still together.

Each standard deviation of style matching roughly doubled the odds of still being together (odds ratio 1.95). As in the speed dates, matching was unrelated to how satisfied the couples said they were (r = .11), and it predicted staying together over and above satisfaction.

Both studies in one picture

The odds ratios are easier to judge with their uncertainty drawn in, and with the self-report measures beside them for scale.

What this study cannot tell you

• Forty dates is a small sample. The speed-date odds ratio could plausibly be anywhere from about 1 to about 9, and with perceived similarity controlled its p-value was .051, just above the conventional line. At the time of writing we could not find a large replication of the speed-date result, so treat it as promising rather than settled.
• It is correlational. Matching might be a sign of interest rather than a cause of it. Nobody was asked to match their partner's style.
• It is a narrow population. Undergraduates, heterosexual pairs, one campus, and in Study 2 text messages rather than speech.
• It does not say matching can be faked. The authors describe style matching as something that happens without awareness. Deliberately copying someone's pronouns is a different act from falling into step with them, and nothing here tests it.

It is also worth setting beside the Stanford speed-dating study, which measured a related but different thing, turn-by-turn mimicry of function words. For women it went with clicking on its own, but once traits and the rest of the conversation were in the model it no longer predicted how much either sex felt they had clicked. The two measures and the two outcomes differ, which is one more reason to hold this finding lightly.

What a speaker can take from it

Style matching is best read as a gauge rather than a lever. It rises when attention is shared: when you are actually following the other person, your grammar drifts towards theirs on its own. The useful lesson is not to imitate anyone, it is that attention shows up in the smallest details of language, whether or not anyone is consciously listening for it.

This article summarises a published paper. The percentages, odds ratios and correlations are the authors’. The confidence intervals and the three category scores in the worked example are ours, computed from numbers the paper reports.

Measured

• Function-word rates in speed-date transcripts and instant messages, counted by LIWC
• A mutual yes after each speed date; still dating at three months

Inferred, not measured

• That style matching reflects shared attention is the authors’ interpretation; the studies are correlational

References

1. Niederhoffer, K. G., & Pennebaker, J. W. (2002). Linguistic style matching in social interaction. Journal of Language and Social Psychology, 21(4), 337–360. https://doi.org/10.1177/026192702237953
2. Gonzales, A. L., Hancock, J. T., & Pennebaker, J. W. (2010). Language style matching as a predictor of social dynamics in small groups. Communication Research, 37(1), 3–19. https://doi.org/10.1177/0093650209351468
3. Ireland, M. E., Slatcher, R. B., Eastwick, P. W., Scissors, L. E., Finkel, E. J., & Pennebaker, J. W. (2011). Language style matching predicts relationship initiation and stability. Psychological Science, 22(1), 39–44. https://doi.org/10.1177/0956797610392928