Skip to content

Text Chats

Principle · All principles
Evidence · 23 claims (12 for, 10 mixed, 1 against) · 20 studies (7 qualitative, 6 causal, 2 review, 2 design, 1 quant-synthesis, 1 associational, 1 theoretical), q2–q4 · 1 of 20 report an effect size · 21 claims rest on one study

Conditional relationship

This page covers live or near-live written exchange among learners, or between learners and a teacher, used for a learning purpose: a task conversation, a question channel during a presentation, a short peer response. For learners who need practice formulating ideas in writing for a present audience (including adult and adolescent second-language learners working on a meaning-focused task), or who are less ready to speak aloud in a group, a focused text chat, on a task with a defined outcome, with the content prepared beforehand and someone keeping the pace legible, is expected to produce written turns in which learners negotiate meaning, notice and correct their own errors, and ask questions they would not ask aloud. The expected gains are in participation and interaction during the task, and in short-form written communication; whether they carry to delayed-retention, skill-performance in speaking, or course achievement is a separate question the wiki does not settle.

The relationship is conditional on the task (closed tasks with one outcome drew more negotiation than open ones in one study), on preparation (familiarity with the task content raised negotiation in another), on the mode chosen (in the one study comparing modes, text chat turns were short, about 9 words, and carried a smaller share of negotiated turns than videoconferencing or a discussion board, but chat, unlike the board, produced corrective feedback), and on pace and moderation. No claim in the wiki compares learning with a text chat against the same task without one, or against speaking. The nearest evidence is one descriptive study comparing three computer-mediated modes with the same French biology students, one small quasi-experiment varying preparation before chat tasks, one correlation between chat activity and test scores in a course run with AI agents, and one teacher's redesign of a chat-based feedback task.

How the converted siblings sit beside this: Peer Discussion owns the commit-then-discuss relationship for conceptual questions; a chat can be its medium, and its model applies unchanged. Social Presence owns being seen and heard as a person in an online course, including when to add live contact with a chat alongside. Community of Inquiry owns the movement of an online discussion towards resolution. Digital Learning holds that a tool helps only through the method it makes affordable. This page owns the choice and running of the written live channel itself.

Default design, while the relationship is untested

The earlier page's guidance, made concrete. Each step is a design proposal with its evidence status beside it; the durations and group sizes are untested proposals unless a claim is linked.

  1. Name the job the chat does in this lesson. One of: a task conversation (learners must reach an outcome together), a question and example channel during a presentation, a short peer response to work just shown, or brainstorming before writing. Do not open a chat "for discussion" with no outcome. From the earlier page; untested.
  2. Prefer a closed task for task conversations. Give a task with one outcome the group must agree on (choose, rank, solve, decide), rather than an open exchange of opinions. In one study of 12 groups of four second-language biology students, negotiation routines were about 50% more numerous in closed than in open tasks, across chat, board and video (Closed meaning-focused tasks foster more negotiation of meaning and negative feedback than open tasks in CMC [+M]); a second study, whose mode the claim page does not state, found more confirmation checks and clarification requests in unfocused tasks (the row on novices below).
  3. Prepare the content before the chat. Give learners 5–10 minutes with the material the task draws on (a short subtitled video, a text, a worked case) before the chat opens. In one quasi-experiment with 10 non-native-speaker dyads doing four 50-minute tasks in a course chat, dyads given subtitled videos beforehand produced a higher share of negotiated turns than dyads without them (Non-native speaker dyads provided with subtitled videos before tasks produce a significantly higher percentage of negotiated turns in online TBLL than dyads without videos [+M]). The 5–10 minutes is an untested proposal.
  4. Keep groups small and the segment short. Two to four learners per chat (the studies above used dyads and groups of four), 10–20 minutes per task segment, one task per segment; with a whole class in one chat, use it only as a question channel (step 6). Group size from the studies' own groupings; durations untested.
  5. Ask for a specific written move, not "comments". Instead of "give feedback in the chat", ask learners to type what they heard, one question they still have, or one example; make posting anonymous or ungraded where fear of being seen wrong is the barrier. In one blended language class, open chat feedback on a speaking task drew four students; asking them to type notes on what they heard drew more specific and varied posts, and an anonymous two-column whiteboard later engaged both cohorts (In a single-teacher case study, redesigning a peer-feedback task produced more specific and varied student notes, which the authors read as a more engaged class [+M]; one teacher's classes, no comparison group).
  6. For a question channel during a presentation, assign a monitor and stop for it. One person (a co-teacher, teaching assistant or rotating learner) reads the chat, groups questions, and raises them at planned pauses every 10–15 minutes; the presenter answers there rather than reading the chat while speaking. Untested proposal, from the earlier page's backchannel example and its warning that fast chats need moderation; the reason to respond to what the chat shows is carried from one-to-one tutoring (Contingent scaffolding improves learning more than fixed or absent support. [~M]; most of its evidence is one-to-one, not chat).
  7. Set and show norms before the first chat. Turn-taking (reply to the previous point before starting a new one), one idea per message, how to signal a question, respect and tone. In one case study of a synchronous classroom chat, the discussion followed rules of tool use, institutional rules and rhetorical routines of opening, topical discussion and closing, and participants took recurring roles such as technological leader and socio-emotional facilitator (SCMD activity systems exhibit three categories of rules and four types of participant roles [+M]; descriptive, one course); naming those roles is an untested proposal.
  8. Leave correction mostly to the learner, and prompt it. In chat, almost all corrective moves were learners correcting their own messages unprompted (112 of 119 in one study), read by its author as avoiding loss of face (Unprompted self-correction dominates negative feedback in NNS-NNS CMC interaction, interpreted as a face-saving preference [+M]). After the chat, give each learner the log and ask them to mark and fix two of their own messages against a stated criterion; the review step is an untested proposal, with the reason carried from self-monitoring research (Self-regulated learning models treat self-monitoring as a core process through which learners notice gaps and adjust strategy [~M]; reviews of self-regulation, not chat studies).
  9. End with a product outside the chat. Close each chat segment with a short summary, decision or list one learner posts for the group, and use it in the next activity, so the exchange feeds composition or a decision. Untested; from the earlier page's caution against surface participation.

Dose and progression across sessions (untested proposals): over a unit of four to six sessions, start with closed tasks, prepared content and a teacher or assistant monitoring every group; in middle sessions, rotate the monitor and summariser roles among learners and move to tasks with two acceptable outcomes; in late sessions, use open tasks with less preparation and learner-led moderation. Move a group to the next step when its last chat reached an outcome every member contributed to, with the norms kept without reminder. Where the goal is spoken interaction, move the same tasks from chat to audio or video once learners complete them in chat with few clarification requests, because chat turns stay short (row below).

When a condition is not met: if no one can monitor a fast chat, switch to a slower channel (a shared document or a board with a 24-hour reply window) and lose live correction; if typing is the barrier, allow dictation or voice notes in the same channel.

Fitting the design to a situation

The three facts that most change the decision:

  • What the written exchange is for. Negotiating meaning in a second language, asking questions during a talk, giving peer response, or reasoning at length each need a different task and mode; chat turns are short, so extended reasoning belongs elsewhere. If unstated, ask: what should a learner have done or produced by the end of the chat, and would a spoken or asynchronous exchange do it better?
  • Who can watch the chat, and how many threads it carries. A dyad or a group of four can run itself with a closed task; thirty people in one live chat need a monitor. If unstated, ask: who reads the chat while it runs, how many learners post into one stream, and when will someone respond?
  • Learners' typing, language and access. Keyboard speed, proficiency in the language of the chat, device and bandwidth decide who can keep pace. If unstated, ask: can every learner type short replies in this language at conversation speed, on the device they will use?
If the brief says… Then change… Basis
Learners are second-language learners and the goal is negotiating meaning or accuracy Use small-group chat on closed tasks with content prepared beforehand; keep the chat synchronous rather than a board where correction is wanted; review logs for self-correction afterwards Discussion board interactions generated no negative feedback at all, suggesting synchrony is required to foster corrective feedback [+M] (12 groups of French biology students, descriptive mode comparison; carried to other languages untested)
The goal is spoken interaction or oral proficiency Use chat as a rehearsal stage only: run the task in chat first, then the same task by audio or video, where turns are longer and more of them are negotiated Videoconferencing was conducive to more negotiation of meaning than text chat and discussion board, measured as proportion of negotiated turns [~M] (one study, proportions of turns, no learning outcome)
The goal is extended argument, explanation or syntax practice Do not rely on chat; use a board, shared document or spoken discussion, and bring only short questions to the chat Content-related problems trigger most negotiation routines, and syntactic triggers are absent from text chat where turns are short [~M] (chat turns about 9 words against 55 and 130; one study)
Learners are novice in the language or the topic, or the task is open Give a closed task with one outcome and 5–10 minutes of prepared content first; supply sentence starters for the first session Closed meaning-focused tasks foster more negotiation of meaning and negative feedback than open tasks in CMC [+M]; subtitled videos before chat tasks [+M]; Unfocused tasks elicit significantly more confirmation checks and clarification requests than focused tasks [~M] (36 learners, mode not stated on the claim page: a qualification on "closed is always better")
One teacher with a large class, or a lecture or presentation Use one class chat as a question and example channel with a named monitor and planned pauses every 10–15 minutes; for task conversations, split into groups of two to four with their own chats untested proposal; contingency at class scale is not shown (Contingent scaffolding improves learning more than fixed or absent support. [~M])
Mixed mode: some learners in the room, some remote; or some on phones only Run one shared chat for everyone, in-room learners included, so remote learners' questions sit beside the room's; the monitor reads remote posts first at each pause; keep messages short enough to type on a phone and allow voice notes untested proposal; see also Social Presence for mixed groups
Learners have low confidence speaking aloud, or are new to the group Open with anonymous or ungraded posts of a specific kind (what you heard, one question), then named posts in small groups once the norms hold In a single-teacher case study, redesigning a peer-feedback task produced more specific and varied student notes, which the authors read as a more engaged class [+M] (one teacher's blended language classes; no comparison group)
Learners have low literacy or slow typing, or a disability affecting typing Allow dictation (Speech-to-text) or voice notes in the chat; give more time per turn, a dyad rather than a group of four, and pre-written phrases they can paste untested proposal; from the earlier page's access and typing constraint
child learners, or adolescent learners in school Keep chats teacher-visible and within a class platform; dyads or groups of three; 10 minutes per segment; one closed task untested proposal; the chat studies here are with university students or adult learners
adult learners in a work setting Use the team's own chat tool and a real decision from their work as the closed task; agree what is kept from the log untested proposal
Material is private or sensitive (health, personal history, workplace conflict) Do not run it in a group chat; use a one-to-one channel the learner controls, or an anonymous question box; say who can read the log and how long it is kept untested proposal
Facilitator time is scarce (a volunteer, a busy shift, one tutor checking in briefly) Give each group a closed task with a written outcome to post, a learner monitor, and one check-in per segment by the tutor reading only the posted outcomes untested proposal
Single session only, or a fixed short time budget (for example 30 minutes) 5 minutes norms and task, 5–10 minutes preparation, 10–15 minutes chat in groups of two to four, 5 minutes for each group to post its outcome; do not add a quiz or review the brief did not ask for untested proposal
The brief names its own routine for chat (its protocol, prompts, rubric or categories) Keep it; change only what the evidence here says matters: make the task closed, prepare the content, or name a monitor untested proposal
The brief wants a whole-task performance kept whole (a negotiation, a case conference, a client conversation) Run the whole task in chat or by voice as one piece; use the log afterwards for review, not to split the task into separate chat items untested proposal
The chat partners include AI agents or an AI tutor Treat chat activity as a signal of engagement, not of learning; check learning with an unaided task In-class chat engagement significantly predicts higher test scores in MAIC courses [~M] (correlational, controlling for earlier module scores; does not show chat caused the scores)
Stakes are high, or chat posts will be graded Grade the posted outcome or summary, not message count or speed; never judge proficiency from chat alone, where turns are short and self-corrected untested proposal

Six of the seventeen rows rest on a claim: five on chat or computer-mediated-communication studies and one on a correlational course run with AI agents. Each is a single study, and none measures learning after the chat. The large-class row cites a claim only for its limit; the other ten are untested proposals.

Observation, state and explanation

What can be observed is the log: who posted, how often, message length, replies to others, questions asked, clarification requests and confirmation checks, self-corrections, and whether the group reached the task's outcome. "Engaged", "understands" or "too shy to speak" are inferred states. A busy chat shows activity, not learning; a quiet learner may be reading closely.

Response Plausible explanations Observation that could change the interpretation
A learner posts nothing or only "ok", "yes" Typing or language speed cannot keep pace; does not know what kind of post is wanted; fears being seen wrong; reading and following Give a specific, anonymous post type (one question, one thing heard) and a dyad; see whether posts appear. Ask afterwards what they would have said.
Many messages, no outcome reached Open task with no defined end; threads competing; no one summarising Re-run with a closed task and a summariser role; count whether an outcome is posted.
Messages fast and short, few questions or clarifications Chat's pace rewards short turns; learners avoid showing non-understanding; the task does not need negotiation Compare the same task by video or with a familiarisation stage first; count clarification requests.
Errors left uncorrected in the log Learners do not notice them; face concerns stop peer correction; the task rewards speed over accuracy Give the log back and ask learners to mark two of their own errors; see how many they find unaided.
Remote or phone-only learners post less than in-room learners Typing on phone; posts not read by the room; latency Have the monitor read remote posts first at each pause; compare their share of posts across sessions.

These rows are proposals, untested with learners as diagnostics. In one case of chat-based dynamic assessment of two learners, both moved towards responding to less explicit help across sessions, with backsliding, and they differed from each other at the same proficiency level (Interactionist DA mediation in SCMC reveals learners' microgenetic movement through internalization levels within the ZPD [+M]; Learners at the same proficiency level differ in responsiveness to mediation, with Sarah needing less explicit help than Dena [~M]): read a learner's chat responses to help across sessions, not from one exchange.

Evidence and qualifications

No claim in the wiki compares learning with a text chat against the same task without one, against speaking, or against an asynchronous board on a learning outcome. The claims below describe what happens in chats and test design features within them.

Learners, tasks and measures differ (French university students in English, intermediate language learners, a large AI-agent course), and four of the chat-specific claims come from one study, so do not rank them. None measures delayed learning.

The evidence by question

Further evidence, not yet read against this model

Claims bearing on text chat that are not part of the model above, with how far each bears on this page.

Objective and learner-valued goal

The designer's objective is usually stated as participation, interaction practice or a product of the exchange (a decision, a summary, a set of questions answered). Elicit separately what learners want from the chat: to get a question answered without being seen asking it, to practise a language without speaking in front of others, to finish the task quickly, or to stay in touch with a group they rarely see. A learner who values not being seen wrong may post only anonymously or only correct themselves; one who values speed may post fragments that reach an outcome without the language practice the designer intended. Record both, and do not read a lively log as the capability either side wanted.

For example, an adult learning English for work may want to follow and answer messages in their team's chat at work, while the course objective is negotiating meaning in a task. One closed task run in the team's own tool, with a real decision, serves both; a classroom chat on an invented topic serves only the second.

What would revise this model?

The expectation should weaken if comparisons of the same task with and without a chat (or chat against speaking) show no difference in learning, or in the participation of learners who speak least, or if negotiation in chat does not predict later language gains. If open tasks routinely produce as much negotiation as closed ones in chat, drop step 2; if prepared content does not replicate beyond small samples, drop step 3 to a suggestion. If anonymous posting raises participation but lowers the quality of posts, the low-confidence row needs a limit. If studies with children show chat working differently from university students, the age rows need their own evidence. Do not protect the model by calling every quiet chat "a moderation problem" after the fact.

Participation in the chat, the share of negotiated turns, self-correction, unaided writing or speaking afterwards, and course achievement are separate claims. The present evidence establishes no group size, segment length, or best mode for any of them.

Claims

Examples

Key Sources

  • Heafner, T., & Plaisance, M. (2014). Exploring synchronous text chat in remotely delivered early field experiences. Journal of Interactive Learning Research, 25(3), 327-352.
  • Lin, H. (2015). A meta-synthesis of empirical research on the effectiveness of computer-mediated communication in SLA. Language Learning & Technology, 19(2), 85-117. doi:10.64152/10125/44419
  • Payne, J. S., & Whitney, P. J. (2002). Developing L2 oral proficiency through synchronous CMC: Output, working memory, and interlanguage development. CALICO Journal, 20(1), 7-32. https://doi.org/10.1558/cj.v20i1.7-32
  • Tare, M., Golonka, E. M., Vatz, K., Bonilla, C. L., Crooks, C., & Strong, R. (2014). Effects of interactive chat versus independent writing on L2 learning. Language Learning & Technology, 18(3), 208-227. doi:10.64152/10125/44391
Page metadata
Field Value
type principle
id text-chats
title Text Chats
description For learners who need practice writing for a live audience or a lower-threshold way to ask and respond, a focused text chat on a closed task, with content prepared beforehand and someone keeping the pace legible, is expected to produce negotiation of meaning, self-correction and wider participation, but no claim in the wiki tests chat against the same task without it on a learning outcome.
status review
canonical True
generated by claude/unspecified on 2026-10-15
trust unverified — no one has confirmed this page's content
sources 1 — payne-2002

Frontmatter is the page's machine-readable record — see the Schema & Guide.