Little words
« previous post | next post »
Molly Ireland, Richard Slatcher, Paul Eastwick, Lauren Scissors, Eli Finkel, and James Pennebaker, "Language Style Matching Predicts Relationship Initiation and Stability", Psychological Science 12/13/2010.
Previous relationship research has largely ignored the importance of similarity in how people talk with one another. Using natural language samples, we investigated whether similarity in dyads’ use of function words, called language style matching (LSM), predicts outcomes for romantic relationships. In Study 1, greater LSM in transcripts of 40 speed dates predicted increased likelihood of mutual romantic interest (odds ratio = 3.05). Overall, 33.3% of pairs with LSM above the median mutually desired future contact, compared with 9.1% of pairs with LSM at or below the median. In Study 2, LSM in 86 couples’ instant messages positively predicted relationship stability at a 3-month follow-up (odds ratio = 1.95). Specifically, 76.7% of couples with LSM greater than the median were still dating at the follow-up, compared with 53.5% of couples with LSM at or below the median. LSM appears to reflect implicit interpersonal processes central to romantic relationships.
This is interesting stuff, in my opinion. But so far, it's gotten relatively little uptake in the popular press. And in one of the few — rather obscure — pieces to mention this research, it was spun as a guide to dating success.
Alasdair Wilkins, "Using very short words can make people fall in love with you", io9 ("We come from the future") 5/23/2011:
Generally speaking, romantic relationships form around common interests, compatible personality traits, or similar beliefs and values. But none of that compares to the love connection that can be formed by using basic words like "he" or "and" together.
Pronouns ("he", "them", "you"), along with articles ("the", "a", "an") and conjunctions ("and", "but", "or") are among the most basic and most widely used words in the English language. Collectively referred to as "function words", these words are so simple and well understood that most people don't actively process them when understanding sentences – instead, our brains process them quickly and unconsciously.
That's why these function words can have surprising power in forging relationships, at least according to research by University of Texas psychologist James Pennebaker.
There are some interesting examples of science-journalism spin here. First, Mr. Wilkins tells us that function-word matching "forges" an incomparably stronger "love connection" than "common interests, compatible personality traits, or similar beliefs and values", none of which were in fact compared quantitatively to Linguistic Style Matching in the cited research. And second, and more important, correlation is turned into causation. What Ireland et al. actually say is this:
As is the case for most research on verbal coordination, our data are correlational. Although Pickering and Garrod (2004) contend that language alignment causes mutual understanding, other researchers contend that individuals’ goals to be liked and understood cause behavioral coordination (Brennan & Hanna, 2009; Lakin, Chartrand, & Arkin, 2008). It is likely that LSM and its underlying psychological processes are bidirectionally linked. Specifically, we suspect that style matching and relationship engagement reciprocally increase one another and jointly facilitate positive relationship outcomes (Niederhoffer & Pennebaker, 2002).
There's a more accurate take on the research here — Melissa Wenner Moyer, "The Language of Love: Word Usage Predicts Romantic Attraction", Scientific American 5/23/2011:
What distinguishes a fling that ends in tears from long-term love? Past research suggests that the most successful couples share common interests, values and personality traits. Now new research published in Psychological Science proposes that the simplest words lovebirds use to speak to each other also make a difference—both in determining how attracted they are and how likely they are to stay together. […]
The big question is whether individuals feel more aligned to others who already talk the way they do or whether they adapt their language to match that of individuals they really like. Pennebaker admits that both are possible, but he believes the latter is the driving force: language, he says, predicts relationship success because it reflects how well couples listen to each other. What is Pennebaker’s advice for living happily ever after with a loved one, then? “Pay closer attention to the other person,” he says.
Update — since there seems to be some confusion about what "Linguistic Structure Matching" actually is, in operational terms, here are the details. First, they counted each participant's use of words in nine categories:
Second, they turned these counts into proportions, dividing by the total number of words of all types produced by the participant in question.
Third, they calculated a separate LSM score for each of the nine categories, using the following equation:
As they explain,
In this formula, preps1 is the percentage of prepositions used by the first person, and preps2 is the percentage used by the second. In the denominator, 0.0001 is added to prevent empty sets. The nine category-level LSM scores were averaged to yield a composite LSM score bounded by 0 and 1; higher numbers represent greater stylistic similarity between two speakers. One LSM score was calculated for each speed date sets.
Chris said,
June 7, 2011 @ 11:23 am
Pennebaker pops up in the news often thanks to his Linguistic Inquiry and Word Count (LIWC) software. but since it's proprietary, it's closed, as far as I know. I've long wanted to peak inside and see how it works.
[(myl) Actually, it's pretty well documented. Take a look at the documentation, and try the online version, and see what questions you still have. Probably the most obscure parts would be the lists of things like "positive emotion words", which (as far as I know) are not available.]
Chandra said,
June 7, 2011 @ 12:38 pm
I knew someone was going to have written some kind of stupid dating-guide type article even before I got to that part of the post. Researchers should start affixing giant warning stickers to their papers: MISUSE OF THIS INFORMATION MAY BE HAZARDOUS TO YOUR CREDIBILITY.
jmmcd said,
June 7, 2011 @ 1:08 pm
> First, Mr. Wilkins tells us that function-word matching "forges" an incomparably stronger "love connection" than "common interests, compatible personality traits, or similar beliefs and values", none of which were in fact compared quantitatively to Linguistic Style Matching in the cited research. And second, and more important, correlation is turned into causation.
Third (and most important :), it's not about using zillions of short words. It's about *matching* your partner's use of them.
GeorgeW said,
June 7, 2011 @ 1:46 pm
Frankly, I was not aware that the percentage of function words varies significantly from speaker to speaker. Does this correlate somehow with intelligence, social class, interests, etc.?
I am trying to think of how to speak with fewer or more function words.
[(myl) I suspect that you may find that the world is full of things that you were not aware of — I certainly often encounter unsuspected facts. Read some of James Pennebaker's collected works to see some of the many ways that function-word distributions vary across speakers and writers, and what sorts of the things the variations can "mean".]
Chandra said,
June 7, 2011 @ 1:59 pm
@GeorgeW – My understanding is that it's not simply a matter of quantity, but as much (or more) to do with usage. So, for example, some people will use one function word in a certain phrase or context where others would use a different one.
[(myl) There are surely also contextual differences, but this paper is strictly about relative frequency of use, independent of context.]
diana j said,
June 7, 2011 @ 2:02 pm
I find it very surprising that even in written communication, the idea is that shorter/function words are more attractive to potential partners…
[(myl) No, the idea is if two people's relative frequency of function-word uses are more similar, then they're more likely to be attracted to each other and more likely to stay together. And the similarity in question might be higher rates of "little word" use, or lower rates of "little word" use.]
GeorgeW said,
June 7, 2011 @ 2:08 pm
@Chandra: I must have missed something. According to the study, "LICW calculates the percentage of total words in a test that fall into nine basic-level function-word categories . . ."
[(myl) That's not the relevant independent variable in the study, which instead was LSM ("Language Style Matching"), a number reflecting the average similarity between two people in their percentages in the nine categories.]
m.m. said,
June 7, 2011 @ 2:53 pm
So, my adopting parts of my boyfriends lect is a good thing in the long run then?
John Lawler said,
June 7, 2011 @ 3:16 pm
@m.m: Yes — that seems to be a good way to describe it — provided you want the relationship to last. If you're looking to end it, do the opposite.
I strongly suspect that the function-word counts are not central, but rather something that was easy enough to measure. What seems to be going on is more a matter of matching styles, which is something couples do in many ways. I'd guess that syntactic structures and phonological rules of all sorts would also show a correlation, but they'd be much harder to measure.
The Ridger said,
June 7, 2011 @ 3:43 pm
It seems pretty clear: if you say "for you and me", "give it to me", and "you and I should go to a movie", while your partner says "for you and I", "give it to myself", and "me and you should go …", you're doomed.
[(myl) This may be true, but if so, the cited study would not discover it, since the two sets of utterances you cite would be scored identically by the metric used.]
Keith M Ellis said,
June 7, 2011 @ 3:49 pm
I really am not eager to defend Wilkins, as crappy science journalism is something I despise; but I think you're not being fair/accurate on both counts.
On the former, you characterize what he wrote as asserting
What he actually wrote was
Granted, "none of that compares to [this]" is often used as "this is very much stronger than [that]". But I think it's possible that Wilkins simply meant that the two are genuinely incomparable.
That is to say, he may not be asserting something about the relative importance of the two in love, but rather that the discovery of the LSM correlation to relationship longevity is something quite different and unexpected relative to the other, well-known factors.
With regard to the latter criticism, that he's confusing correlation with causation, I think that he's correctly inferring a purported bidirectional causal relationship from the words of Pennebaker; particularly so with regard to his source, the SA article:
…where "latter" refers to "they adapt their language to match that of individuals they really like" (emphasis mine).
The study itself, from what I understand, doesn't have anything whatsoever to say about a possible causal relationship. This is Pennebaker offering his unsupported opinion to SA, which Wilkins then repeats in even more forceful terms, building a theme around it.
There's three levels of progression here, resulting in bad science journalism. But it began with Pennebaker offering something pretty hand-wavy to Scientific American, who included this somewhat provocative claim, and which was then subsequently repeated and amplified by another journalist two-steps removed from the research itself.
This is probably not so unlike most bad science journalism. Although I suspect it's less likely the researcher himself/herself offering sensationalist spin on otherwise respectable results, but rather originates with someone from the institution's PR department who wrote a press release about the research.
Evan said,
June 7, 2011 @ 3:52 pm
I don't know about "making people fall in love with you" but it does seem like it could be a useful signal for dating site matching algorithms.
Keith M Ellis said,
June 7, 2011 @ 3:57 pm
Oh, wait, it's the study itself which makes a bidirectional causal claim, isn't it?
…referring to Pennebaker directly.
[(myl) That's not a "causal claim", it's a "causal hypothesis".]
Rubrick said,
June 7, 2011 @ 4:25 pm
Ah! Now I get it! I see that the key, if I'm to get a lot of hot sex, is for me to use all one-, two-, and three-letter…
Damn.
[(myl) Amount of sex was not a dependent variable in either study, I'm afraid. And even if you think that perceived attractiveness and length of involvement are appropriate proxies, it doesn't matter whether you use many or few function words; what matters is whether you and your partner are similar in this respect.]
Xmun said,
June 7, 2011 @ 4:27 pm
My wife pronounces "extraordinary" as extra-ordinary. I say "extr'ordinary". We've managed to stay together for forty-odd years despite this and several other differences in our speech habits. She's the only person I know who says "slippy" instead of "slippery". Word-formation experts, what do you call these shortenings of ours? Are they examples of syncope?
Chandra said,
June 7, 2011 @ 4:50 pm
@GeorgeW – Admittedly I'm no expert on the matter, but it seems that they're measuring whether compatible pairs use similar types of function words, rather than overall quantity (can anybody who understands mathematico-science speak better than me confirm this?).
[(myl) The exact measure that they used is described in an update to the post. It's not very complicated — basically fifth-grade math. Nothing but adding, subtracting, and dividing.]
They also state: "Among function words, personal pronouns appear to be particularly relevant to relationships. Married couples who use "we" more often and "you" less often have lower divorce rates and report greater marital satisfaction (Seider, Hirschberger, Nelson, & Levenson, 2009; Simmons, Gordon, & Chambless, 2005)." Usage is clearly the issue here (i.e. measuring which of two possible pronouns is preferred in certain contexts, rather than overall quantity of pronouns used).
[(myl) In those two studies, context was not taken into account, but the relative frequency of "we" and "you" apparently mattered. In the (Ireland et al.) study under discussion in this post, counts of "we" and "you" were lumped together with all other personal pronouns in the first of their nine function-word categories, and so no differences in we/you choices could have any effect on their measures.]
m.m. said,
June 7, 2011 @ 6:39 pm
Funny; I'm playfully vehement against being infected by his use of [ɐ] for /ʌ/, but he'll catch me using [ɐ] instead of [ɜ] sometimes and will point it out, but I always get him for beign poor-pour merged [where im Cure-fir merged], and for using [ɑ̟] for his back vowels where I'm mostly a [ɒ] user.
Were doomed xD
Relaxed pronunciation ?
Brett R said,
June 7, 2011 @ 8:30 pm
I don't know if the particular lexical categories they use have a bearing on their argument, but they're an odd mix of traditional and modern. The CGEL would disagree with:
-all their impersonal pronouns: `it' is personal (3rd person), `that' is a determinative, and `anything' is a compound determinative.
-`because' as a "conjunction'; it's a preposition
-`lots' as a quantifier; it's a noun
Also, `be' and `was' are usually auxiliary, but not always.
J. Goard said,
June 7, 2011 @ 9:51 pm
This work is so cool!
I think I'm gonna get distracted from my current project for a day or two…
Jerry Friedman said,
June 7, 2011 @ 10:14 pm
@Brett R: I think you're onto something. If the data had been reanalyzed with the CGEL's lexical categories, would the correlation with positive relationship outcomes have been better or worse? Clearly, the lexical categorization with the best correlation is objectively the best. In fact, if we redefine "positive relationship outcomes" as the number of children, we have an evolutionary criterion for evaluating grammars!
I'm joking, but for some reason I enjoy this idea of extra-linguistic tests of grammatical schemes—brain scans, for instance. Is such a thing even possible?
Rubrick said,
June 8, 2011 @ 1:08 am
@myl: I can't figure out whether my joke was too subtle and you didn't get it, or your reply is too subtle and I don't get it.
[(myl) Both…]
chris said,
June 8, 2011 @ 8:01 am
The big question is whether individuals feel more aligned to others who already talk the way they do or whether they adapt their language to match that of individuals they really like.
My first reaction was that LSM is detecting people who speak similar dialects, which serves as a marker for common background. Not terribly interesting, but IMO something that needs to be ruled out before advancing to more complicated hypotheses like "language . . . predicts relationship success because it reflects how well couples listen to each other".
In particular, the fact that the high LSM is already observable in a speed date suggests the couples are bringing it with them. Pennebaker's theory would suggest gradual convergence in linguistic style over the course of a relationship.
I wish more people would ask "is there a completely boring/unremarkable explanation for this, or could it be a consequence of a phenomenon that is already well known to exist?" before spinning more elaborate theories.
[(myl) Accommodation and priming are both pretty rapid processes. And even without any adjustments based on the partner's contributions, it's plausible that a measure like LSM can differentiate symmetrical from asymmetrical interactions, which is not a matter of dialect but a matter of style and attitude.
I'd be intuitively skeptical of the view that LSM mainly registers dialect as opposed to style differences, and similarly skeptical of the view that people are in general more romantically attracted to others who share their own dialect. So if your explanation is correct, it would be neither "a phenomenon already known to exist" nor an idea that is "boring/unremarkable".]
Chris said,
June 8, 2011 @ 12:33 pm
I'm sympathetic to Pennebaker (honest, I am…), but I've been asking Twitter linguists to run their feeds through his online tool Analyze Words which uses LWIC to analyze anyone's twitter feed and frankly, not pretty. Apparently we're all highly depressed yet oddly In-the-moment. Go check it out and judge for yourself.
Rubrick said,
June 9, 2011 @ 4:31 am
@myl: And I even screwed it up, with "that". Maybe I should find a new line of non-work.
Lane said,
June 9, 2011 @ 1:32 pm
What's the hypothesis for why function-words correlate so well with personality-clicking? I'd think that whether or not my conversation partner likes to use "rebarbative" and "chiasmus" in roughly the same proportions that I do would be a better predictor of how we'll get along than how often we use prepositions…
[(myl) To a first approximation, the reason for using function words for this application is the same reason that Jamie Pennebaker used function-word profiles for all sorts of other applications in social and clinical psychology. In individual English texts — conversations, narratives, essays, poems, suicide notes, whatever — almost the only words that are common enough to support statistical analysis are function words. (You can also analyze the relative frequency of word-classes, or the distribution of projections of words into a low-dimensional subspace like "latent semantic analysis", etc.)
Contrary to what I would have guessed, it has turned out that function-word distributions are somewhat correlated with all sorts of interesting things. (In fact, by the standards of social psychology, the correlations are quite robust. But note that in absolute terms, the connections are still rather weak ones.)
It's often possible to construct convincing post-hoc explanations for these relationships, and after some experience, to guess in advance what the connections will probably be.
In this case, there are several plausible ideas, it seems to me. Perhaps LSM is a proxy for communicative symmetry, i.e. the degree to which the give-and-take between two people involves balanced discourse roles. Perhaps it's a proxy for more general linguistic accommodation.]
Barbara Phillips Long said,
June 12, 2011 @ 12:37 am
What I have learned from editing is that other writers don't choose the same prepositions I would in various situations. Here are some possible variations:
Here's some gravy to put over your potatoes.
Here's some gravy to put on your potatoes.
Here's some gravy for your potatoes.
Sit down at the table.
Sit down on the chair [at the table].
Sit yourself down.
Sit down over here.
Could having similar patterns of preposition use mean that less processing time is involved in understanding the speech of the other person, and if processing is easier, does that either reduce stress or allow greater processing from other sensory inputs or provide some other benefit to building a relationship I haven't thought of?
When some people read, they can infer things that are missing in the text. With matching patterns in speech, does someone hear "gravy" and "potatoes" and think "[a] can go over [b]" until they realize they heard "for," at which point the hearer backtracks and makes sure "[a] for [b]" is equivalent to "[a] over [b]"?
[(myl) This could well be true, and It would be interesting to check it. But in the experiment under discussion, the matching in question was at a rather coarse level, namely in terms of LIWC's nine categories of function words:
In that tally, "over" and "on" (for example) are both counted in the category "preposition".
The historical reason that LIWC does it that way is that it's the only way to get large enough counts in small documents for the relative frequencies to be statistically stable. So to test your idea, you'd have to do a more focused kind of experiment, where you check the particular choices made in designated circumstances.]
re/interpret: how does your work read? | re/search said,
July 20, 2011 @ 2:29 pm
[…] Liberman recently wrote at the Language Log about a case of misinterpreted science. Liberman begins by describing a study showing that how […]
John Lawler said,
August 29, 2011 @ 7:45 pm
Late-breaking news:
SciAm article on this topic here.
This Week’s Language Blog Roundup | Wordnik said,
June 5, 2013 @ 11:11 am
[…] “that’s mighty white of you” (which surprisingly did not always have to do with race); how language style matching may predict relationship attraction and stability; and the Navy SEALs of snowclones. Language Log also took on speech-based lie detection of […]