An English teacher with a four-person B1 online class watches two university students share one laptop and one 120-word assignment, and the quicker writer takes the keyboard, drafts six sentences, asks for approval with a quick "OK", then continues after the partner agrees. Both names appear at the top, yet only one learner has made a lexical or grammatical decision. Good pair work depends on shared control. Proficiency affects what each learner can offer, while the interaction determines whose knowledge enters the task and who gets to test an idea. Teachers should design for mutual contribution instead of assuming that a high-low pairing automatically produces peer teaching.
A pair can contain one worker
Pair work is usually counted by seating: two people, one task. Learning researchers listen for a stricter kind of evidence. A language-related episode occurs when learners discuss the language they are producing, question a choice, correct a form or search for a better expression. "Use despite here," one learner proposes. The partner checks whether it needs a noun after it and asks to read the whole sentence once. That exchange gives both partners something to retrieve later.
Compare a polite takeover. One learner holds the keyboard and supplies every answer; the other agrees often enough to keep the interaction pleasant. The final paragraph may score well. The quiet partner has had few chances to retrieve vocabulary, explain a choice or notice a gap. Product quality and learning opportunity have separated.
That distinction becomes visible only in the talk.
Yuko Watanabe and Merrill Swain's 2007 study of adult ESL learners followed four core participants as each worked with partners of higher and lower proficiency. Pairs wrote together, compared their text with a reformulated version, and then wrote individually, with collaborative interaction patterns associated with more language-related episodes and higher post-test performance regardless of the partner's level. Four focal learners and one three-stage writing design provide close evidence, with limited reach beyond comparable adult tasks. The result supports watching how partners engage, while leaving open how stable one pair's pattern remains across topics and weeks.
Level labels predict less than expected
| Study | Pairing evidence | Main limit |
|---|---|---|
| Watanabe and Swain, 2007 | Four focal adult ESL learners | One writing sequence |
| Storch and Aldosari, 2013 | 15 proficiency-grouped pairs | One male college cohort |
| Li and Liu, 2022 | 31 Chinese EFL pairs | One collaborative writing task |
Neomy Storch and Ali Aldosari recorded 30 male college learners in Saudi Arabia. They completed a short composition. Their 2013 study formed five high-high pairs, five low-low pairs and five mixed pairs. The researchers examined attention to language, use of English and the relationship that emerged inside each pair. Their conclusion was conditional: the best proficiency arrangement depended on the task's aim, and the dyadic relationship could matter more than the level combination.
A larger Chinese EFL study found a similar pattern. Ningning Li and Yingliang Liu recorded 62 learners in 31 pairs during collaborative writing: 12 high-high, 12 high-low and seven low-low. Across 1,019 language-related episodes, collaborative and expert-novice relationships produced more discussion of vocabulary, grammar and discourse than dominant patterns. Proficiency grouping did not yield differences in the total or successful resolution of those episodes.
Level labels are only a starting point.
This 2022 study used one Chinese educational setting and one writing task. It also treated language-related episodes as an indicator of learning rather than tracking every learner for a semester. Its value lies in showing a mechanism teachers can observe: requests, explanations, challenges and uptake.
Build two decisions into the task
Telling learners to "work together" leaves control available to the faster hand or louder voice, especially when both can see the same page and the first acceptable answer ends the search. A better task gives each person information, responsibility or a decision the other person needs. Choc's article on information gaps in speaking activities covers one route. Collaborative writing needs its own distribution of control because access to the document alone says nothing about who may challenge, revise or decide, especially once a plausible first draft makes further discussion feel optional.
One simple constraint is alternating ownership by sentence. Learner A proposes the first sentence, learner B must ask one genuine language question or offer a revision, and B makes the final choice. Roles reverse for the next sentence. The exchange should remain about meaning and form. Ceremonial comments such as "looks good" satisfy the rule on paper and teach very little.
Another design separates resources. One partner receives the source text and the other receives the audience brief. They have to explain their information before drafting, so the written product cannot begin until both voices have supplied content that the other person lacks, and neither learner can complete the brief by quietly copying the other learner's work. For a workplace email, one may hold the delivery delay and dates while the other holds the relationship, purpose and requested action. The resulting disagreement has content: include September 17, soften you failed to send, keep the request visible.
Both must leave a mark.
I prefer a shared document with the cursor passed back and forth over one learner serving as permanent typist. Typing speed has an odd way of becoming authority. On Microsoft Teams, the teacher can ask each learner to mark one phrase they proposed and one choice their partner changed. Version history adds another check because it records whose cursor inserted a sentence, which revision survived discussion, and whether the quieter partner ever moved beyond comments into authorship. That history shows participation without asking either learner to perform collaboration for the teacher.
Teach collaboration as observable moves
Some pairs already negotiate well. Others need language for disagreement, clarification and invitation. Give them short moves tied to the current task.
- Ask why. Then request the verb's connection with the previous sentence and ask the partner to point to the link in the text.
- State a different reading and invite the partner to choose after both reasons are clear.
These prompts make participation visible and keep the dispute attached to the sentence on screen. During monitoring, a teacher can note one proposal, one question and one accepted change beside each name, then compare that small record with the final individual task once the activity has ended. A displayed bank of four phrases is enough. A long politeness sheet becomes another text to process, and learners may spend their attention selecting a polite frame while the disputed grammar slips past untouched. The teacher needs evidence of reasoning, uptake and revision.
Next, the teacher listens for balance and mutuality. Balance concerns control. Mutuality concerns engagement with the partner's idea. A learner can speak for half the time while ignoring every proposal. Another can speak less, ask two precise questions, and change the final text. Raw talk time misses that difference.
Count decisions.
Intervention should target the pattern. With a dominant-passive pair, take the keyboard away briefly and ask the quieter learner to choose between two wordings. With a dominant-dominant pair, require each person to state the other's reason before selecting a version. An expert-novice pair may be working well if the more proficient learner invites predictions and lets the novice finish decisions. Equal level and equal value are separate matters, and a skilled teacher keeps the useful asymmetry of knowledge while changing any routine that turns one learner into a silent audience for the other learner's performance.
Check what each learner carries away
A jointly written answer can hide individual gaps. End with a 73-word solo version, a one-minute explanation, or two sentences that reuse the disputed form. Watanabe and Swain included individual writing after collaboration for exactly this reason: shared performance and individual learning need separate evidence.
Separate the evidence.
Follow up individually. If one learner can explain the pair's choices but cannot reproduce them, schedule retrieval. No influenced choice means redesigning the roles. If both independently use the repaired phrase, keep the task structure and change the partner later.
For the next pair activity, watch the cursor for 47 seconds. Write down who types, who asks the first language question and who decides after disagreement. Those three observations tell you more about the learning opportunity than the two placement-test labels beside their names.