Most learners do best with plenty of input first and some output from early on. Stephen Krashen argues that understood input drives acquisition and that output is too rare to matter much. Merrill Swain counters that speaking and writing make you notice gaps that listening and reading let you skip. Both sides have evidence.
What is the output hypothesis?
Merrill Swain proposed it in 1985 after studying Canadian French immersion pupils. Their achievement tests, taken in French, matched those of pupils taught in English, which showed the French they had heard for years was comprehensible. Yet their speech and writing still had many grammatical errors. Swain's explanation was that they were never pushed: they could get their meaning across without getting the grammar right, so they never had to process it closely.
She later described three things that producing language does which input alone may not.
| Function | What happens | A learner's example |
|---|---|---|
| Noticing | Trying to say something shows you the gap between what you want to say and what you can say. | You start to describe your weekend and realise you don't know the past tense of "to go". |
| Hypothesis testing | You try a form, and the listener's reaction tells you whether it works. | You guess a phrase, get a puzzled look, and rephrase it. |
| Metalinguistic reflection | Talking or thinking about the language itself helps you control what you know. | Afterwards you ask a friend why the phrase didn't land. |
Swain's own claim was modest. As Krashen summarises it, she never said output builds all or even most of a learner's competence, only that under some conditions it helps in ways input doesn't.
Why does Krashen think output matters less?
In a 1998 paper Krashen made three objections: output, and especially output that a learner has to repair, is too scarce to build much competence; people reach high levels without it; and there was no direct evidence that it causes acquisition. His scarcity figures come from studies of real conversations and classrooms.
| Study | What was counted | Result |
|---|---|---|
| Pica (1988) | Ten one-hour conversations between low-level ESL learners and native-speaker teachers | 44 occasions when learners modified what they said (about four an hour), of which 13 involved grammar (about one an hour) |
| Ellis, Tanaka & Yamazaki (1994) | 42 learners allowed to interact with a native speaker during a task | Only 7 negotiated meaning; the rest just listened, and the 7 had no clear advantage in comprehension or vocabulary |
| Lyster & Ranta (1997) | 18.3 hours of French immersion lessons, 3,268 pupil turns | 73 clarification requests from teachers, followed by 20 pupil corrections, about one an hour |
| Swain & Lapkin (1995) | Grade 8 immersion pupils writing and editing a short essay while thinking aloud | 190 moments of noticing a language problem, 10.6 per pupil, with no evidence any led to improvement |
He adds that being pushed to speak is unpleasant for many learners: in Young's 1990 survey, students put talking at the top of their list of anxiety-provoking classroom activities. Krashen's conclusion is that more comprehensible input is a better bet than more output. Our page on comprehensible input covers that side of the argument.
What does later research say?
The fairest summary is that output helps with producing language and input helps with understanding it. A 2013 meta-analysis by Shintani, Li and Ellis compared comprehension-based teaching with production-based teaching of grammar. According to a published summary of it, both had large effects; comprehension-based teaching did better on immediate tests of understanding, while production-based teaching did better on delayed tests of production.
Research on conversation points the same way. Michael Long's interaction hypothesis (1981) gave a role to negotiating meaning with a partner. Larsen-Freeman and Long later concluded that interaction isn't necessary for acquisition but helps in certain circumstances, and critics note that beginners may lack the language to negotiate at all.
The pro-output studies are often small. Nobuyoshi and Ellis (1993) found two of three Japanese learners improved their past-tense accuracy after being pushed to correct it, one from 31% to 89%. Krashen points out that only that one gain was statistically significant. So the case for output is real but thinner than the case for input, and mostly about accuracy and fluency in speaking.
So when should you start speaking?
Start producing early, but in small, low-pressure doses, and keep input as the main course. If your goal is to read and understand, input does most of the work. If your goal is to talk, you need to practise talking, because understanding a sentence and building one are different skills.
This line from a Spanish A1 romance puts Swain's "gap" in two sentences:
En su cabeza tiene una conversación perfecta. En la vida real solo dice: "Hola".
Stories are also a good supply of lines to borrow. In the Contos library (October 2026), 54% of the paragraphs in the free chapters of the 13 languages written in Latin or Cyrillic script contain dialogue. A practical sequence:
- Read first. Follow a chapter at your level until most of it is easy.
- Borrow a line. Pick one line of dialogue you could use yourself, and say it or write it from memory.
- Retell. Summarise the chapter in three sentences without looking. Swain counted writing as output too, and you'll find your gaps quickly.
- Talk when you can. A patient partner gives you the feedback Swain describes. If that makes you anxious, Krashen's evidence suggests you're not alone, and private practice is a reasonable start.
Read a chapter at your level, then retell it in three sentences without looking.
Read free in your browserFrequently asked questions
Can you learn to speak a language just by reading?
You can build the knowledge speaking draws on, and Krashen cites people who reached high levels from input alone. But Swain's immersion pupils understood French well and still spoke it inaccurately after years of input. If speaking is the goal, plan to practise it.
Does writing count as output?
Yes. Swain and Lapkin's 1995 study was of pupils writing and editing an essay, and they noticed language problems about ten times each. Writing is slower and less stressful than conversation, so it's an easy way to start producing language.
How much input do I need before I start speaking?
Research doesn't give a threshold. Krashen would say speaking emerges once you've understood enough; Swain would say early output helps you notice what you're missing. A workable middle path is to start short retellings once you can follow beginner stories with few look-ups.
Why is speaking so much harder than understanding?
Understanding lets you lean on context and guess, so you can skip grammar you haven't mastered. Speaking makes you choose every word and ending yourself. That's the core of Swain's argument, and why comprehension usually runs well ahead of production.