Communication accommodation theory, measured at scale
Communication accommodation theory says people drift towards each other's way of speaking. For forty years it was only shown in small studies. Then someone ran it over 215,000 real conversations.
Communication accommodation theory is the observation that people in a conversation converge on each other’s way of speaking, in word choice, sentence shape, length, pitch and gesture. A 2011 study tested it on 215,000 real conversations and found it holds, but that the pull between two people is usually one sided.
The theory is about forty years older than the evidence that most people now cite for it. What makes the large study worth reading is not the confirmation. It is the complication it found underneath.
What the theory claims
The theory accounts for the general observation that participants in conversations tend to converge to one another’s communicative behaviour, coordinating across several dimensions at once: choice of words, syntax, utterance length, pitch and gestures. Howard Giles is the name most associated with it, and a 2011 study by Cristian Danescu-Niculescu-Mizil, Michael Gamon and Susan Dumais cites his 2008 account of the theory as the reference point.
Two properties matter before any measurement. The coordination happens without the speakers noticing. And it does not have to happen on every dimension at once: converging on one feature does not rule out diverging on another. The same paper cites a 1988 study in which women conversing with men converged on how often they paused while diverging on laughter.
The part that is easy to miss is what counts as style here. Style means the components of language unrelated to content, how things are said rather than what is said. The paper notes that style words such as articles and prepositions make up only about 0.05 per cent of the English vocabulary, yet an estimated 55 per cent of all the words people actually use.
Why it took forty years to test properly
The 2011 paper puts it bluntly: in its almost forty years of existence the theory had been supported exclusively through small scale or controlled laboratory studies. A room, a handful of people, a recording.
The obstacle was never interest. It was data. Measuring accommodation between two people requires many conversations between the same two people, not many conversations in general. The authors note that the largest conversational corpus then available had an average of only 4.3 conversations per user, which is nowhere near enough to model how a particular pair speaks to each other.
How the large test was built
The researchers assembled a second dataset designed for the problem. They kept only pairs where both people had started a conversation at least twice, collected the public posting history of those users, and reconstructed the exchanges between them. That produced 15 million messages from 7,800 users, making up 215,000 conversations between 2,200 pairs with a reciprocal relationship.
The density is what makes it work. On average each pair had 98 conversations, 2.7 turns per conversation and 270 days of contact. The median pair had 60 conversations over 257 days, and the most prolific pair had 1,744.
Style was measured with the Linguistic Inquiry and Word Count method, which sorts word use into more than sixty categories such as articles, auxiliary verbs and positive emotion words. The authors then built a probabilistic framework specifically so that accommodation could be separated from homophily, which is the much duller explanation that similar people talk to each other in the first place.
The finding that complicates the theory
Accommodation held up. The paper reports this as the first time linguistic style accommodation had been examined and verified in a large scale, real world setting.
The complication is symmetry. The authors found that imbalance in stylistic influence between two people is preponderant, and that whether accommodation is mutual depends on which feature you look at. People were more likely to accommodate symmetrically on first person singular pronouns and asymmetrically on prepositions. The paper describes this as the first time such complexity in the phenomenon had been revealed.
In plain terms: in most pairs, one person is doing more of the moving, and which person that is can change depending on the feature. The drift is real, and it is rarely an even trade.
The limits the authors state
The conversations studied were public posts on one platform, capped at 140 characters, not face to face and not in real time. The authors raise these as genuine challenges to the theory rather than footnotes, and they say it was not clear in advance whether accommodation would survive the constraints. The message history collected was also capped at the 3,200 most recent posts per user by the platform’s own interface.
So the result travels further than a laboratory study and not as far as every conversation you have ever had. A long private thread with one person is a different object from a short public reply, and the paper does not claim otherwise.
What it means for reading your own conversations
The useful consequence is that the way you write in one conversation is partly a property of the other person. If influence is usually lopsided, then a record of how you speak to one friend is not a clean record of how you speak.
That is the same caution from a different direction as what a chat analysis cannot tell you about someone, and it is why what your texts sound like to everyone else changes depending on who is reading them. If you are weighing whether to hand a conversation to a tool at all, whether it is safe to upload your chat history sets out the data path first.
Politeness carries the same problem. Why a please can read as rude covers the strand of the literature where a courtesy marker is read against the relationship it appears in.
Portraitor reads one named participant in a conversation you choose, and its privacy policy describes the result as inferences about that participant, each carrying an estimate of confidence. A portrait drawn from one thread is a portrait of someone inside that thread. Analysing a WhatsApp chat with AI covers what that involves, and Portraitor shows the flow itself.
The idea that a speaker shapes their words for the person they are addressing has its own name and its own origin, set out in audience design.
Frequently asked questions
Frequently asked questions
What is communication accommodation theory?
The theory that people in conversation converge on each other’s communicative behaviour, coordinating word choice, syntax, utterance length, pitch and gesture. It is most associated with Howard Giles.
Is accommodation deliberate?
No. The coordination occurs without the speakers noticing, which is one of the properties of the phenomenon described in the literature.
Does accommodation happen on everything at once?
No. Converging on one dimension does not exclude diverging on another. One cited study found women conversing with men converged on pause frequency while diverging on laughter.
Is accommodation mutual between two people?
Usually not. The 2011 large scale study found that imbalance in stylistic influence between two people is preponderant, and that symmetry depends on which feature is measured.
How large was the study that tested it?
It used 15 million messages from 7,800 users, giving 215,000 conversations between 2,200 pairs with reciprocal relationships, with 98 conversations per pair on average.
Does this apply to private messaging?
Not automatically. The study used short public posts capped at 140 characters, which the authors name as a limit. A long private thread is a different setting.
A narrower measure of the same effect counts the small function words two people share rather than style as a whole. Language style matching is that measure, and it has its own literature.
Two people talking are not two independent writers. Any reading of one of them carries a little of the other.
One thread, one person, stated limits
Portraitor reads the participant you name and returns inferences that each carry an estimate of confidence. Masking happens in your browser first.
See how Portraitor works →- Danescu-Niculescu-Mizil, C., Gamon, M., & Dumais, S. (2011). Mark my words! Linguistic style accommodation in social media. arXiv:1105.0673.
- Giles, H. (2008). Communication accommodation theory. In Engaging theories in interpersonal communication: multiple perspectives. Sage Publications. Cited in Danescu-Niculescu-Mizil et al. (2011) above.