All articles

Grade the help a learner needs

24 October 2026

Choc Education


Two learners can both score 6 out of 10 on a grammar check and need quite different lessons next. One may repair an answer after hearing, “Check the time phrase.” The other may still need a worked example and a choice between two forms. A useful assessment should record that difference. The amount and type of help required can reveal more than the raw total.

This idea comes from dynamic assessment, which combines a task with support given while the learner is working. The aim is diagnostic: find the lightest prompt that lets the learner continue, then see whether the same help works on a new item. For a teacher planning remediation, “correct after a cue” and “correct after a model” are different results.

A score freezes one moment

An ordinary quiz samples independent performance under fixed conditions. That is appropriate for many decisions, including certification and end-of-course reporting. It is less informative when a teacher must decide what to teach on Tuesday.

Poehner and Wang's 2021 research timeline describes second-language dynamic assessment as attention to both completed abilities and abilities that are still emerging. Support may take the form of a reminder, a leading question, a hint, feedback, or a model. Responsiveness to those moves becomes part of the evidence.

Consider a hypothetical B1 learner in Taichung who writes, “I live here since 2022.” A teacher could reveal the answer immediately. A graduated sequence produces a finer record:

  1. Ask the learner to reread the sentence and mark the time expression.
  2. Ask which form connects a past starting point with the present.
  3. Offer a choice between “have lived” and “lived.”
  4. Show a parallel example with “since 2019.”
  5. Supply the corrected sentence and ask for a fresh example.

If step 2 is enough, the learner probably has the concept and failed to retrieve it. If step 5 is needed, another explanation and several interpretation tasks may be warranted. The original error looks identical on paper.

Prompts need an order

Random help is hard to interpret. “Try again,” a translation, a gesture, and a complete answer do not carry equal weight. Before the check, the teacher needs a short ladder that moves from implicit to explicit assistance.

The early classroom work by Aljaafreh and Lantolf used a regulatory scale for feedback in second-language writing. Later work has adapted graduated mediation to grammar, reading, listening, and computer-delivered tasks. The exact number of levels varies. Thirteen levels may suit close research analysis. Four or five are easier to use beside a class register.

A compact code can be enough: I for independent, C for cue, Q for guiding question, M for model. Add the language target and the transfer item. The record “present perfect, Q, transfer C” tells the next teacher much more than 1/2.

Keep the prompts tied to one construct. A reading inference item should not quietly become a vocabulary test because the hint introduces three unfamiliar words. A pronunciation check should separate hearing a contrast from producing it. Choc's article on testing the same word in two ways makes a related point: one response format samples only one route to knowledge.

The next item carries the evidence

A learner may copy a model accurately without understanding the choice behind it. Give a transfer item after mediation, with one detail changed. After repairing “I have lived here since 2022,” ask for a sentence about working at a company from 2023 until now. Remove the prompt if possible.

This is the part I would protect when time is tight. A long ladder without transfer can document compliance. A short ladder plus one fresh item tests whether control has begun to move to the learner.

The distinction is especially useful in a small class. Suppose eight learners miss the same listening item. Three identify the reduced form after a replay, two need the transcript with the phrase underlined, and three still cannot connect the sound with the known written word. The follow-up groups nearly form themselves: replay practice, sound-to-text matching, and focused decoding work.

Scores can also hide inefficient success. A learner who answers correctly after 73 seconds, erases four attempts, and asks whether the instruction permits more than one answer has produced a different performance from someone who responds independently in nine seconds. The teacher need not time every item. One latency note or visible false start can explain why a nominally mastered form still collapses in conversation.

Keep it modest. Dynamic records become unusable when every hesitation receives a code. Select one language feature and one kind of prompt, then compare the same feature across two dates. A narrow record can change a lesson. A page of symbols usually becomes paperwork.

Stop there. Once the record shows the smallest successful prompt, the immediate transfer response, and one later unprompted response, adding codes for posture, confidence, speed, wording, and every abandoned attempt will make the evidence harder to use during planning rather than more precise.

Limits matter

Dynamic assessment cannot replace a standardized score. Teacher prompts vary, and a learner who receives more help has worked under different conditions. Reports for GEPT, TOEIC, or admission decisions need the published testing rules, not an improvised mediation scale.

The evidence base also contains many small studies and varied procedures. A recent three-level meta-analysis of computerized dynamic assessment reports positive effects across its included studies while stressing differences in prompt timing, design, assessment format, and methodology. That variation limits any single estimate. Computerized prompts can standardize a ladder, though they cannot respond to an unexpected misconception as precisely as a skilled teacher.

Nor should every exercise stop for five hints. During fluent discussion, repeated interruption can destroy the task. Use graduated prompts at selected diagnostic points, then let the conversation continue.

A four-column record

For one target this week, write four columns: task, independent response, least help needed, transfer response. Prepare no more than four prompt levels. Run the check with two or three items and keep the language neutral. The learner should experience assistance, not a sequence of rebukes.

At the next lesson, begin with an unprompted transfer item. A learner who needed a model on Friday and answers independently on Tuesday has shown a change that the earlier 6 out of 10 could never display.