The Interaction Hypothesis: Why Conversation Drives Language Learning
The interaction hypothesis is the second-language acquisition (SLA) account of why conversation drives language learning. Input becomes usable not just when it is simplified, but when two speakers negotiate meaning to keep a conversation going.1 Michael Long proposed it to explain a gap that input alone leaves open. It is the clearest theory available for deciding whether a tutor or a language exchange is worth your time, and what to do once you are in the session.2
Overview
The interaction hypothesis holds that comprehensible input is necessary for acquisition, and that conversation is the main place where input becomes comprehensible.12 It connects two older theories and turns them into a single, conversation-centred mechanism.
It matters for self-study because it identifies the active ingredient in talking with another person. That ingredient is not talk time; it is the repair work speakers do when communication threatens to break down.3
Where it came from
The applied linguist Michael H. Long first formulated the interaction hypothesis in 1981, in the paper "Input, Interaction, and Second-Language Acquisition."1 The 1981 version argued that modified input and modified interaction together facilitate acquisition more efficiently than the alternatives. It also argued that the adjustments speakers and listeners make to keep a conversation going are what render input comprehensible.1
Long built directly on Stephen Krashen's input hypothesis. He took from it the premise that comprehensible input (input at the learner's level plus a little beyond, "i + 1") is central to acquisition.4 Long's distinctive move was to locate the source of that comprehensibility not in pre-simplified texts, but in the give-and-take of conversation.1
Long substantially revised the hypothesis in 1996, in "The Role of the Linguistic Environment in Second Language Acquisition."2 The revision shifts emphasis toward cognitive mechanisms. Environmental contributions to acquisition are mediated by selective attention (noticing) and the learner's developing processing capacity. These resources come together most usefully, though not exclusively, during negotiation for meaning.2
The 1996 version is the one usually meant by "the interaction hypothesis" in current SLA writing. It foregrounds corrective feedback and noticing in a way the 1981 version did not.2
The hypothesis bridges Krashen's input hypothesis, where input drives acquisition,4 and Merrill Swain's output hypothesis, where producing language (especially when pushed) also drives acquisition.5 Long's account incorporates both: conversation supplies comprehensible input and demands output. The negotiation between the two parties is where acquisition is hypothesized to be promoted.26
The two (then three) pillars
In its common summary, the hypothesis rests on two linked propositions. First, comprehensible input is required for second-language acquisition. Second, input is made comprehensible to the learner through negotiation for meaning in conversation. In other words, speakers make interactional adjustments when communication threatens to break down.12
A third pillar is associated with Teresa Pica and colleagues. Communication tasks that require an exchange of information, with participants in equal, two-way roles, generate more negotiation of meaning than tasks where one party simply holds all the information.37 Task design therefore affects how much acquisition-relevant interaction actually happens.
Pica, Young, and Doughty found experimentally that interactionally modified input, the kind learners could negotiate, produced better comprehension than input that was only pre-simplified.7 That result supports the claim that the negotiation process itself, not just the simpler language, is doing the work.
What negotiation of meaning actually is
Negotiation of meaning is the work two speakers do to make a message understood when they sense a comprehension problem. Pica frames it as serving three functions in learning: modifying input toward comprehensibility, prompting modified output from the learner, and providing feedback focused on form.3
The repair moves
The recurring conversational moves that carry out this work are often called interactional moves. They are commonly grouped into four types. Each is described below with an English gloss of what the move sounds like. No verbatim Japanese transcript is reproduced because none was available from a citable source.
A clarification request is the listener signalling that the message was not understood and asking the speaker to rephrase or explain. In English it surfaces as "What do you mean?" or "Sorry?"38
A confirmation check is the listener echoing back what they think they heard, usually with rising intonation, to invite a yes or no answer. In English: "You mean ___?"38
A comprehension check runs the other direction: the speaker checks that the listener has followed. In English: "Do you know what I mean?" or "OK?"38
A recast is the more proficient speaker reformulating all or part of the learner's erroneous utterance into target-like form while preserving the meaning. The speaker does this without explicitly flagging the error. Long treats recasts as a key channel of implicit negative feedback.28
These moves have been documented among learners of Japanese as a foreign language. Iwashita recorded learner-to-learner Japanese interaction and analyzed how often interactional moves occurred and how much modified output they elicited.9 She found that mixed-proficiency pairs produced more interactional moves than same-level pairs. But a higher count of moves did not by itself yield the most modified output.9
Why repair drives acquisition
Long's 1996 account proposes that negotiation is acquisition-relevant because it brings together the resources learning needs at the same moment. It draws the learner's selective attention to a gap (noticing), supplies situated comprehensible input calibrated to what the learner failed to understand, and can elicit pushed output as the learner reformulates.2
The "noticing" link is what the 1996 revision adds over 1981. Negotiation is held to encourage noticing of the gap between what the learner produced and the target form, which is a precondition for that form to be acquired.2
The pushed-output half of the mechanism is Swain's contribution: producing language, not just receiving it. When a learner is pushed to make an utterance comprehensible to the listener, the effort of producing more precise, better-formed language is itself hypothesized to promote development. Comprehension alone is not thought to do the same work.56
The diagram below traces how one repair cycle is hypothesized to feed these three resources.
These are the hypothesized mechanisms. The claim that they cause acquisition is supported but not proven in full, as the evidence section below sets out.
How tutors and language exchanges instantiate it
The mapping from theory to platforms here is reasoned, not measured. This article does not assert an efficacy statistic for any specific platform.
Paid tutors: structured negotiation
A paid one-to-one tutor deliberately and repeatedly uses the interactional moves the hypothesis identifies as acquisition-relevant: running comprehension checks, issuing clarification requests, and recasting learner errors into target-like Japanese.12 Because the tutor is the more proficient speaker, they can also scaffold input at roughly "i + 1" and supply corrective feedback on demand.4
This is negotiation of meaning under controlled conditions. The asymmetry that makes a tutor an expert also lets them target the learner's actual gaps rather than leaving repair to chance.23 The trade-off is that the interaction is arranged rather than spontaneous.
Language exchanges: reciprocal but noisier
A reciprocal language exchange, where each partner is a native speaker of the other's target language, generates authentic, two-way, information-exchange interaction. That is the symmetrical-role condition Pica associates with the most negotiation of meaning.37
The cost is corrective expertise. An untrained exchange partner negotiates meaning naturally but is far less likely than a trained tutor to deliver reliable, well-targeted recasts. Some errors pass uncorrected, and feedback quality varies by partner.23
Iwashita's Japanese-learner data speaks directly to this noisier condition. Among learners, more interactional moves did not translate into the most modified output. That is consistent with the caution that learner-to-learner exchange produces real negotiation, but does not guarantee the output gains an expert can elicit.9
What input-only study cannot give you
Passive listening and reading supply input. For Krashen, comprehensible input is the engine of acquisition.4 But input-only study structurally cannot supply two of the three things negotiation provides: it elicits no pushed output, because the learner produces nothing, and it offers no real-time, contingent repair targeted at the learner's own gaps.256
This is not a "you must speak from day one" claim. The defensible version is that input alone leaves the output and feedback channels unused. Adding interaction opens them.2 The hypothesis does not require that interaction begin immediately.
Nuance and usage contexts
What the evidence supports and complicates
Mackey ran a pretest-posttest study with adult English as a second language (ESL) learners (N = 34). Learners who took part in negotiated interaction targeting question forms developed more advanced question structures than learners who received the same input without interaction.10 That is experimental support that interaction can facilitate development, not merely comprehension.
Mackey and Goo's meta-analysis of interaction studies found a clear overall benefit for interaction. Averaging across vocabulary and grammar outcomes, they report an effect size of roughly d = 1.09, a large effect. Effects were larger in delayed post-tests than immediate ones, and somewhat larger in foreign-language than second-language settings.11
The standard SLA caution still applies. The strongest, most consistent evidence is for vocabulary and for some morphosyntax. Long's own 1996 statement is that negative feedback obtained during negotiation may facilitate development at least for vocabulary, morphology, and language-specific syntax. It may also matter more for some L1-to-L2 contrasts than others.211
The mechanism is not automatic. Casual conversation does not always contain negotiation. If speakers simplify too aggressively, the input can lose the very features the learner needs to notice. Beginners may also lack the second-language (L2) resources to negotiate effectively, so negotiation can be less useful at the very bottom than in the intermediate range.2812
Not every recast is registered as a correction. Because a recast preserves meaning and does not flag the error, learners often interpret it as a continuation of the conversation rather than as feedback. Its corrective value therefore depends on whether it is noticed.28
When interaction matters most
The sources support diminishing returns at the zero baseline and rising value once a learner has enough language to negotiate from. Mackey and Goo's robust gains come from learners who already had a footing to interact from. The beginner-limitation point implies interaction yields most once a learner has accumulated enough input to bring something to the exchange.118
This maps onto the plateaued intermediate: the learner who consumes input but rarely interacts. No source supports any claim that ties interaction to a fixed timeline to fluency, and none is made here.
Good to know
"Interaction" is not the same as "speaking practice"
The active ingredient the hypothesis identifies is negotiation of meaning and form-focused feedback, not raw talk time. Pica frames negotiation specifically as input modification, output modification, and feedback on form. None of these is guaranteed simply by talking a lot.3
Iwashita's finding that more interactional moves did not produce the most modified output is the concrete caution.9 An hour of conversation is not automatically an hour of acquisition input. The value tracks how much negotiation, noticing, and feedback the hour contained, not its length.
Recasts are easy to miss
Because a recast reformulates the learner's utterance into correct form without announcing that an error occurred, learners often process it as ordinary uptake rather than correction. Its benefit is contingent on noticing.28 A recast is socially gentle, which is part of why it can slip past unregistered.
In a tutored session, the practical fix is to ask the tutor to make corrections explicit, or to confirm the corrected form back. That converts implicit feedback into noticed feedback.2
The "you must speak from day one" misreading
Interaction theory does not mandate immediate speaking for all learners. Input remains necessary. Interaction adds the output and repair channels, and is least effective at the absolute-beginner stage where learners cannot yet negotiate.28 When to start speaking is the practical question this leaves open.
The sources note that negotiation may be less effective for beginners who lack the L2 knowledge to negotiate. That directly contradicts an obligatory day-one-speaking reading.28
Krashen versus Long in one line
Krashen holds that comprehensible input is the sufficient condition for acquisition.4 Long holds that comprehensible input is necessary, but that interaction (negotiation of meaning) is largely what makes input comprehensible in the first place. The 1996 revision adds that noticing and corrective feedback during negotiation are central.12
The distinction beginners blur is "exposure" versus "negotiated exposure." A compact way to hold it: input plus repair, not input alone.
See also
- Why You Can Read Japanese But Can't Speak It: Closing the Output Gap
- Using AI for Japanese Conversation Practice: What an LLM Can and Cannot Do
- Online vs. In-Person Japanese Tutoring: Which Actually Works Better
- The Case for Shadowing Before Conversation
- The Comprehension Threshold: How Easy Should Japanese Input Be?