How Interlaza adjusts difficulty in matching-to-sample activities

Imagine this identity trial: a learner selects the circle that matches the sample while the distractors are barely visible. In the next trial, the distractors gain contrast and the learner answers correctly again. Those are two correct responses with help; they are not yet two independent responses.

More choices, less support, or a sample that disappears changes what we ask the learner to do. Here is how Interlaza uses responses to adjust practice.

The next trial needs context

A learner selects the correct image. Before increasing difficulty, it helps to know how that response happened: were the choices easy to tell apart? Was there a visual prompt? Was the sample still visible?

A prompted response and an independent response provide different information. Adapting an activity means looking at the conditions of each response alongside the final percentage.

Interlaza combines response history, an estimate for each target, and adjustment rules. The Instructor configures the activity and reviews whether those changes make sense for the learner.

What happens in a matching-to-sample trial

In matching-to-sample, or MTS, a sample is presented and the learner selects an option from a set of comparisons. The correct response depends on the sample and the relation being taught.

In an identity activity, for example, a picture of a dog appears and the learner selects that same picture from several choices. Other activities may involve matching a photograph to a pictogram or a spoken word to a picture.

Designing that relation and its alternatives matters as much as counting correct responses. Grow and LeBlanc’s recommendations (2013) for auditory–visual discriminations explain how stimulus arrangements and prompts can help or hinder control by the sample.

Four conditions that change the task

These conditions help explain difficulty. They do not form a required ladder that every learner must follow in the same order.

Reducing support does not mean making every error impossible. Fading aims to support responding and transfer control to the relevant stimuli; it does not guarantee an error-free session.

See what changes in a trial

The demonstration lets you respond, manually change choice count, and choose how the sample appears. After each response, watch the help in the next trial and the independent streak. The circle changes position between trials.

The example uses one concept and the basic rules for three phases of help. It saves no data and does not adjust choice count automatically. Its BKT figure is illustrative: it uses the default identity parameters and changes only after independent responses; responses with help do not update it. The cover is also an illustration, not an app screenshot.

Answering correctly with help and responding independently

With basic opacity fading, the correct comparison stays visible while distractors move from an opacity of 0.10 to approximately 0.617 and finally to 1. With the initial three-phase settings, one correct response advances a phase and one error moves back a phase. Each target has its own progression; the Instructor can configure other criteria.

The streak used to fade help includes correct responses with help. The streak used to reach the mastery criterion requires independent responses: a correct response with help adds nothing to that streak and interrupts it. The engine’s base criterion is three consecutive independent correct responses; the editor offers other criteria. Meeting an activity criterion does not by itself check transfer or maintenance.

What informs automatic adjustments

Automatic comparison-count adjustment is off by default. It applies only when the effective configuration enables it for a compatible activity; its presence in the engine does not mean every session uses it.

When enabled, to adjust the number of comparisons, Interlaza combines an estimate for each target, calculated using Bayesian Knowledge Tracing (BKT), with recent correct and incorrect response streaks. It allows an initial observation period, spaces changes apart, and respects activity limits. While the learner still needs substantial support, it may keep the number of choices unchanged.

For how that estimate is calculated, which responses update it, and how to interpret its threshold, see How BKT helps Interlaza pace practice — without declaring learning. Here we focus on the visible result: what changes in the next trial and which conditions to review.

Increase, hold, or reduce: three examples

• Increase: in a compatible activity with automatic adjustment enabled, independent responses are stable and the estimate is sufficiently high. The engine may add a comparison if valid alternatives are available and the configured limit allows it.
• Hold: there is not enough information yet, or prompts are still needed. Keeping conditions unchanged allows further observation before introducing a change.
• Reduce: the estimate is low or errors recur. The engine may remove a comparison if the lower limit allows it. Fading has its own rules for restoring support.

These are examples of how the system works, not instructions for interpreting an individual learner. An error may also involve ambiguous materials, difficulty touching the screen, or an interruption. The Instructor reviews that context.

Three changes you can recognize on screen

From three comparisons to four. The sample and the relation being taught stay the same, but upcoming trials include one more choice. This requires a compatible activity, valid alternatives, and a limit that allows it. Removing a choice reduces the number of alternatives; it does not, by itself, change the relation being taught.

From faint distractors to fully visible distractors. The choices stay the same, but the correct response becomes less visually prominent. This is visual prompt fading. Progression is tracked for each target: one session can include targets at different support levels.

From a short delay to a longer one. In delayed matching, the sample appears first, then disappears, a blank interval follows with neither sample nor comparisons visible, and finally the choices appear. When delay progression is enabled, that interval changes according to its own rules. Keeping choices visible while hiding the sample would be a different presentation.

These changes increase different demands: examining more alternatives, responding with less support, or selecting after a delay. Identify which change occurred before interpreting the next response.

Sample presentation changes what you can observe

In the last two modes, responses occur after the sample disappears. The demonstration shows the sample for two seconds and offers a few example delays. The app lets the Instructor configure duration and delay in half-second steps; the demonstration’s options are not its complete list.

When reviewing speed, distinguish total latency, which can include presentation and delay, from response latency, measured from the moment comparisons appear. A longer delay can increase the first measure without the learner taking longer to choose.

Adaptive difficulty helps pace practice through changes that can be reviewed. Its usefulness depends on a well-designed activity and the Instructor’s observation.