Here is a sentence with no grammatical error in it. A student writes to a professor: I want you to write me a recommendation letter by Friday. Every word is spelled correctly. The tense is right. The word order is right. And the professor, reading it, decides the student is presumptuous, and remembers it.
Now here is the same sentence with four grammatical mistakes: I am wanting that you will be so kind to write for me a letter of recommendation, if it is possible for Friday. This one is broken in at least four places, and the professor thinks: how hard-working, to write in a second language.
The two sentences are identical in intention. They differ by roughly forty words of padding, and they produce opposite social verdicts. That gap is the subject of a research literature that language learners almost never hear about, and it is the reason some people with B2 grammar get treated better than people with C1 grammar.
The construct is called pragmatic failure. The name and the framework come from Jenny Thomas, in a 1983 paper in Applied Linguistics that is still the standard reference. Her definition is deliberately narrow:
"I have given the term 'pragmatic failure' to the inability to understand 'what is meant by what is said'."
Note what is excluded. Not mishearing. Not misparsing. Not not knowing the word. Pragmatic failure begins where the sentence is understood and the intention is missed - or where the intention is clear and the social meaning of expressing it that way is wrong.
Two Failures That Look Identical and Are Not
Thomas split the term in two, and the split is the most useful thing in the paper, because the two halves have completely different fixes.
Pragmalinguistic failure: the wrong tool for the job
Pragmalinguistic failure happens when the linguistic form you choose carries a different pragmatic force in the target language than it did in yours. It is a mapping problem. The words are right; the job they do is not.
Thomas's cleanest example comes from Russian classrooms teaching English. A teacher, running a reading activity, asks a student: X, would you like to read? In British English that is a highly conventionalized polite directive. It is not a question about preference; it is an instruction wrapped in a question mark. Russian learners answered it as a question about preference. They said no, I wouldn't - not to be rude, but because nobody had asked them whether they wanted to.
The same paper gives the of course problem. Russian speakers transferred konechno, which functions as an enthusiastic yes, into English, where of course carries an implicature that the question was stupid. Thomas puts the result at "at best peremptory and at worst insulting."
And there is a harder version, reported by Anna Wierzbicka, where the form does not exist at all: ability questions such as can you return the videos? are not conventionalized as requests in Polish. So a Polish speaker asking for a favor in English may reach for a form that sounds, to them, indirect and therefore polite, and the English listener hears a question about capability and answers it literally. Both parties are being reasonable in their own system.
Sociopragmatic failure: the wrong judgement about the room
Sociopragmatic failure is worse, and Thomas says so explicitly. It is not a mapping error but a values error. She locates it in "cross-cultural mismatches in the assessment of social distance, of what constitutes an imposition, of when an attempt at a 'face-threatening act' should be abandoned, and in evaluating relative power, rights, and obligations."
This is why the two are not equally fixable. A pragmalinguistic error can be taught in an afternoon, as Thomas argued - it is conventionalized usage, close enough to grammar to be drilled. A sociopragmatic error involves the learner's model of how people ought to treat each other, and correcting it means telling an adult that their instincts about politeness are wrong. Thomas's advice to teachers was to point these out and discuss them, never to correct them prescriptively.
Anyone who has watched a learner be told that their culture's way of showing respect is rude will recognize the problem.
The Evidence That Grammar Does Not Carry Pragmatics
The claim that you can be grammatically excellent and pragmatically incompetent is not an anecdote. It has been measured, twice, in the same institutional setting.
Kathleen Bardovi-Harlig and Beverly Hartford recorded 32 academic advising sessions between faculty advisors and both native and highly proficient non-native graduate students at Indiana University. Their 1990 paper in Language Learning examined how students made suggestions and rejections - speech acts that challenge the advisor's status, because the student is proposing a course of action to the person whose job is to decide. Native speakers managed this with status-preserving strategies: hedges, deferential framing, reasons that positioned the suggestion as information rather than instruction. The non-native students, at a proficiency level where vocabulary and syntax were not the issue, frequently did not. Their 1993 follow-up tracked the same students longitudinally and found the same pattern.
"Because of the advanced level of the NNSs, lack of success is not attributable to lack of linguistic competence but to lack of context-specific pragmatic competence involving the use, kind, and number of status-preserving strategies."
Read that as a learner and it stings, because it means the thing you have been working on for years is not the thing that determines whether people take you seriously in a meeting.
Then there is the study that should be printed on the wall of every language school. In 1998, Bardovi-Harlig and Zoltan Dornyei tested 543 learners and 53 teachers using a videotape of 20 scenarios, each containing either a grammatical error or a pragmatic error. Participants rated how serious each one was. There was a primary sample in Hungary and the United States, plus a secondary sample of 112 EFL speakers in Italy.
The result split cleanly by environment. EFL learners and their teachers - learners studying English in a country where it is not spoken - consistently ranked the grammatical errors as more serious. ESL learners and their teachers, living where the language is spoken, ranked the pragmatic errors as more serious.
That inversion is the whole argument in one number. The language is the same. The grammar is the same. What differs is the feedback environment. In Hungary, a grammatical mistake gets a red pen. In Indiana, a pragmatic mistake gets you a reputation. Learners calibrate to whichever punishment arrives first, and most classroom learners have never once been punished for the second kind.
How Much Directness a Language Tolerates
If sociopragmatic norms were arbitrary, none of this would be learnable. They are not arbitrary; they are distributed, and the distributions have been measured.
The Cross-Cultural Speech Act Realization Project, published as Blum-Kulka, House and Kasper (1989), collected requests and apologies across seven languages and varieties using a discourse completion test, and coded every request into three bands: direct, conventionally indirect (could you...), and hints. The distribution of direct requests is the part worth memorizing.
Share of requests made with the most direct strategy
Australian English: 10%, with 82% conventionally indirect. Canadian French: 24% direct. Hebrew: 33% direct. Argentine Spanish: 49% direct, 58% conventionally indirect.
So a Hebrew speaker and an Australian speaker can both be perfectly polite in their own language and land five times apart on a scale that neither of them knows exists. An Argentine speaker using the directness that is normal at home in Sydney is not being aggressive. They are being, by their own calibration, ordinary, and the Australian hears something else entirely.
The forms themselves differ too, not just the frequencies. Juliane House and Gabriele Kasper found that for a low-imposition request between equals, English speakers' preferred strategy was the query preparatory - can you close the window? - while German speakers' preferred strategy was the obligation statement - you should close the window. German is not ruder. It conventionalizes a different form, and a German learner of English who translates their own most polite option produces a sentence that reads as an order.
Blum-Kulka also found that the strategy rated most indirect is not the strategy rated most polite in either Hebrew or English. Hints are the most indirect and the least preferred; the query preparatory wins on politeness. Politeness is not a dial you turn toward "more indirect." It is a convention with a shape, and the shape differs by language.
The Bulge: Why the Middle Is the Danger Zone
Nessa Wolfson's contribution to this literature is a warning against a natural assumption - that politeness increases steadily with social distance, so strangers get the most elaborate forms and intimates the least.
What she found, across compliments, apologies, invitations and requests in American English, is a curve rather than a line. Elaborate politeness formulas cluster at the two ends of a specific band: between status-equal strangers, and between status-equal intimates. In the middle - the ambiguous relationships, where you are neither stranger nor friend, and where status is unclear - the formulas thin out and people improvise. She called it the bulge.
This matters for learners because the middle is exactly where most of their interactions happen. A new colleague. A landlord. A classmate's parent. These are the situations with no script, and they are also the situations where a wrong move is most likely to be read as character rather than as error. Wolfson also found that a tiny set of syntactic patterns covers roughly 80% of compliments in American English - which is good news, because it means the script exists, it is short, and nobody teaches it.
The Formula Trap: Words That Do Two Jobs
The hardest failures to notice are the ones built into a single word, because the learner has already learned a translation and the translation is the trap.
Japanese sumimasen is the standard example, and the research on it is unusually detailed. Ki Kimura recorded ten hours of everyday Japanese conversation and found 44 instances of sumimasen, sorting them into five functions: request marker, attention-getter, closing marker, regret marker, and gratitude marker. Risako Ide then analyzed 51 instances observed ethnographically in a Tokyo ophthalmology clinic and identified seven functions, including sincere apology, quasi-thanks, request marker, leave-taking device, and a ritual reciprocal acknowledgement exchanged in public space.
A learner who files sumimasen under "sorry" will use arigatou for thanks. But the two are not interchangeable: sumimasen is the more formal and more polite gratitude form in many contexts, it tends to suit accepting an offer rather than refusing one, and it is inappropriate as a response to congratulations, condolences or compliments, where arigatou is fine. The card that says sumimasen = sorry is not incomplete. It is actively misleading, and it will keep being misleading for years because nothing about it produces a visible error.
The same pattern shows up in every language with honorifics or address systems. Korean and Japanese sentence-final forms and honorific verbs make a grammatically flawless sentence socially wrong if the register is off by one level. French, German and Spanish tu/du versus vous/Sie decisions encode a judgment about the relationship, and learners routinely make that judgment using their own culture's rules. None of these produce a correction. They produce a slight change in how the other person talks to you afterward.
Can It Be Taught? Yes, and the Effect Sizes Are Large
If pragmatics were an unteachable residue of upbringing, this article would end here. It is not.
The first meta-analysis was Jeon and Kaya (2006), which screened 34 studies and included 13. Since then: Plonsky and Zhuang (2019) synthesized 50 studies and found instruction effective, with explicit teaching beating implicit, instruction with feedback beating instruction without, and instruction offering practice opportunities beating instruction without them. Yousefi and Nassaji (2019) pooled 44 studies from 2006 to 2016 and found an overall d = 1.101, with explicit instruction at 1.213 against implicit at 0.878, and computer-mediated instruction at 1.172 - larger than face-to-face.
The most recent synthesis, published in Applied Linguistics in 2023, pooled 29 studies involving 1,898 learners and produced g = 1.656 (95% CI 1.387-1.924). Three moderator findings from it are worth extracting:
- Duration matters more than in other areas of language teaching. Treatments under three hours produced the smallest effects. Treatments over 18 hours produced the largest. The authors note this inverts the general SLA finding that short interventions often outperform long ones - pragmatic development appears to need more time than grammar does.
- Foreign-language settings gained more than second-language settings. The learners with least access to the target culture benefited most from being taught about it, which is precisely the opposite of the assumption that you have to be there.
- Speech acts dominated the research - over 90% of studies - which means requests, apologies, refusals and compliments are the best-evidenced targets, and everything else is extrapolation.
One caution the meta-analyses repeat: outcome measures inflate effects. Written discourse completion tasks produce the largest gains, free oral production the smallest. If you test yourself by writing what you would say, you will overestimate yourself. That is the same illusion of competence that ruins self-testing everywhere else, wearing a politeness hat.
A Protocol for Studying Pragmatics with Cards
Pragmatics is not vocabulary, so it does not go on a card the way dog does. But a card is just a prompt for retrieval, and the thing to retrieve here is a frame plus its conditions. Six steps that follow from the research:
1. Collect frames, not words
One card, one complete utterance in a situation: "Would you mind if I opened the window?" - asking a colleague you have just met, in an office, in winter. The situation is not decoration. Thomas's whole point is that the same string can be correct in one context and wrong in another, so a card without a context trains the wrong unit.
2. Put the social variables on the back
Back of the card: who you are, who they are, relative status, size of the imposition, and what happens if you get it wrong. This is the schema that sociopragmatic competence runs on. Naming the variables out loud is what turns a phrase into a rule you can transfer.
3. Build "one scene, three versions" cards
Take a single intention - asking someone to repeat themselves - and produce three versions: to a close friend, to a stranger, to your boss. Then a fourth card with the reverse: given the three situations, which version? Directness gradients are exactly what learners flatten, and CCSARP's numbers say the flattening is not small.
4. Anchor every formula to audio
Tone carries a large part of politeness, and a written frame teaches none of it. If your app has text-to-speech, use it; if you have a native speaker, record them. What you are storing is not the words but the delivery, and the delivery is where "it sounded rude" usually lives.
5. Run the two-day test
A pragmatic card is only learned when it has survived contact. Within 48 hours of adding a frame, use it once in the wild - even in a service interaction, even with a language partner. Then log one line: what you said, what happened, and whether the response matched the script. Misfires are the highest-value data in this entire method, because self-assessment cannot detect them. The Bardovi-Harlig and Dornyei result is that learners systematically mis-rank their own errors; reality does not.
6. Recruit a tone check
Find one person who will tell you that your sentence sounded off. Not a teacher marking grammar - a reader with native instincts and the willingness to be blunt. Every advanced learner in the archive of self-taught linguists eventually needed this, and the ones who never found it kept producing sentences that were correct and wrong at the same time.
What Fluency Actually Buys You
The comfortable story about language learning is that accuracy is the mountain and fluency is the summit. The pragmatics literature says accuracy is one foothill, and that a great many learners reach the top of it and then wonder why nobody wants to talk to them.
The good news is that the invisible competence is unusually trainable. It responds to explicit instruction more strongly than most grammar does, it benefits from short focused interventions, and it gains most in exactly the setting where learners assume it is impossible - the foreign-language classroom with no native speakers in it.
What it does not respond to is more vocabulary. You can add two thousand words to a deck and still hand your professor an order in the shape of a request, because the missing piece was never lexical. It was the knowledge that I want you to and would you mind are the same proposition and two different people.
The test is not whether a native speaker understands you. It is whether they would have said it that way. Those are separate skills, they are acquired by different means, and only one of them is on the exam.