Verbal Learning (Free Recall) Task

Version: v1 (current)

Present a word list, then capture spoken free recall after a delay. The task family includes the Rey Auditory Verbal Learning Test (RAVLT), the California Verbal Learning Test (CVLT), and the Hopkins Verbal Learning Test (HVLT).

Overview

A list of words is presented to the participant, one at a time, spoken aloud (played from researcher-recorded audio clips) and/or shown on screen. After a retention interval the participant is asked to say back as many words as they can remember, in any order. Their spoken recall is captured and the researcher scores it afterward against the list that was presented.

One task instance presents one word list and collects one recall. Repeated-trial designs (for example RAVLT's five readings of the same list, an interference list, and a delayed recall) are built by adding several Verbal Learning tasks to the study, each configured with its own list. See "Building a multi-trial study" below.

Scientific background

• The standard presentation rate is about one word per second. Presentation rate and the inter-word interval are experimental parameters; fix them per study so they are reproducible across participants.
• Recall can be immediate or delayed. Delays range from seconds up to roughly 20 to 30 minutes and are often "filled" with a distractor task to prevent rehearsal.
• Free recall is unconstrained speech. The participant may produce correct words, intrusions (words that were not on the list), and repetitions. The raw audio is the scientific ground truth, because these signals are exactly what a researcher scores; an automatic word matcher would discard them. That is why recall is captured as audio and an open transcript, never reduced to a single correct/incorrect flag.

How recall is captured

Recall is captured two independent ways, and recall is scored by hand afterward (the task does no automatic scoring and does not match the spoken words against the list):

1. In-task browser speech recognition (always, best-effort). During recall the browser's own speech recognition runs in open dictation and produces a transcript, stored in the task's summary file. This works on Chrome and Edge with the microphone permission granted. If the browser does not support it or the participant denies it, the task continues and records a marker noting that this source was unavailable.
2. Server-side transcription (when configured). The session is always recorded. If you add a Speech-to-Text analysis to the study, that recording is transcribed on the server, which is generally more accurate. The recall window is marked in the timeline so you can find the spoken recall in the transcription review screen.

When both transcripts exist they are exported as two distinct, unverified inference sources. Both are automated guesses: compare them to each other and fall back to the raw audio. Genuinely garbled audio is a loss.

Recording and transcription dependency

• The session recording is always on; there is no control to disable it.
• A recall transcript exists whenever in-task browser speech recognition is available (source 1), even with no analysis configured.
• The higher-accuracy server transcript exists only if you add a Speech-to-Text analysis to the study. Without it, recall is still recorded as audio and (on Chrome/Edge) captured by source 1, but there is no server transcript.

Presentation modes

• Visual (default) — words are shown on screen, timer-driven.
• Aural — each word is played as an audio clip you provide once in the task configuration: record it with your microphone directly in the study form (record, listen, then confirm or re-record) or upload an audio file. One clip per distinct word; every participant hears exactly the same recording, so the auditory stimulus is identical and reproducible across the whole study. You can play each clip back inline and preview the full list with "Play all". A Clip language setting labels the language of the recordings for the exported data. The study form will not let an aural configuration be saved while any word is missing its clip, or while a clip lasts longer than that word's presentation duration (a clip is never cut off; a shorter clip is followed by silence, and the configured duration always sets the pace). If a clip cannot be played on a participant's device, the task falls back to showing the words on screen and the session is flagged.
• Both — the clip plays and the word is shown together.

During aural or both presentation, a small headphone-level meter is shown to the participant for the whole presentation phase (not just while a clip is actively playing), as a visible sign that audio is part of the task. The level bars drop between words and rise again as each clip plays.

How recall ends

The recall window is bounded by two times:

• Minimum recall time — the participant cannot finish before this; the "Done" button is hidden until it elapses. This stops people quitting during an early pause (in free recall, words come in bursts with gaps, so an early "I'm done" usually isn't). Keep the recall prompt encouraging — "say any other words you remember, take your time."
• Maximum recall duration — a hard cap: recall ends here even if the participant never clicks Done, so a session cannot hang.

Between the two, a finished participant clicks Done and moves on. Two useful special cases:

• Minimum = 0 → fully self-paced (Done available immediately, capped by the maximum).
• Minimum = maximum → a fixed recall window with no early finish (no Done button) — the standard list-learning choice when you want everyone to get exactly the same retrieval time, so recall counts are comparable across participants.

In a strictly moderated session the moderator advances through the recall prompt; the moderator cannot shorten the word presentation or the retention delay (those run on their timers to protect the controlled timing).

Running presentation and recall separately

By default one task does the whole cycle: read the list, wait, recall. The "Mode" option lets you split that cycle when a design needs it:

• Presentation and recall (default) — the full cycle in one task. Use this for immediate-recall trials.
• Presentation only — read the list and stop. No recall is collected here.
• Recall only — go straight to the recall prompt, with no list read first.

Splitting matters in two cases:

• Filled delay — to put other tasks (a distractor) between the list and the recall, use a Presentation only task, then your distractor task(s), then a Recall only task. A single Presentation and recall task can only wait passively (an unfilled delay) — it cannot have other tasks run in the middle.
• True delayed recall — a delayed recall should not re-read the list. Use a Recall only task for it (the words were already learned in the earlier presentation), so the participant simply recalls after the intervening tasks.

Building a multi-trial study (RAVLT and similar)

RAVLT-style protocols read the same list several times and then test delayed recall. Build this by chaining instances:

1. Add a Verbal Learning task for Learning Trial 1 with your list (Presentation and recall).
2. Copy it for Learning Trial 2 through Learning Trial 5 (same list).
3. Optionally add an Interference task with a different list (Presentation and recall).
4. Add an Immediate Recall of the first list (Recall only — the list was already learned above).
5. Run your distractor / delay tasks.
6. Add a Delayed Recall task (Recall only) after the delay.

Give each instance a distinct name (for example "Learning Trial 1" through "Learning Trial 5", "Interference", "Immediate Recall", "Delayed Recall") so the timeline markers and task index make each recall window easy to find. Prefer a filled delay — real tasks between the list and the recall, built with Presentation only then Recall only — over a long passive wait, because long in-task waits are fragile in the browser (mobile tabs throttle timers and sessions can idle out).

Reading recall from the transcription review

When a Speech-to-Text analysis is configured, open the transcription review for the participation and locate the recall window using the recall markers on the timeline (each Verbal Learning instance has its own recall start and end). Read the words recalled, including intrusions and repetitions, and score them against the presented list (which is exported in the task's summary file). When only the in-task transcript exists, read it from the summary file and cross-check against the audio.

Parameters

Parameter	Default	Notes
Mode	Presentation and recall	Or Presentation only / Recall only, to split the cycle
Words to present	10-word demo list	A table: one row per word (configured order = the target list), each with its own duration (ms) and gap to the next word (ISI, ms)
Randomize presentation order	Off	When enabled, each participant sees the words in a unique random order. The configured list stays the target list scored against; the actual shown order is recorded (see below).
Presentation mode	Visual	Visual, Aural, or Both (aural/both require an audio clip per word)
Font size	48 px	Visual / both modes
Audio clips	none	One clip per distinct word: record with the microphone or upload a file; inline playback and "Play all" preview; required before an aural configuration can be saved
Clip language	English (en-US)	Labels the language of the recordings for the exported data; does not change playback
Retention interval	0 ms	Delay before recall; 0 is immediate
Minimum recall time	10000 ms	Done hidden until this elapses; 0 = finish anytime; = maximum for a fixed window
Maximum recall duration	60000 ms	Hard cap; recall ends here even if the participant has not finished
Recall language	English (en-US)	Dropdown; language for the in-task recognizer
Retention / recall messages	provided defaults	Participant-facing text you can edit

Randomization and serial-position analysis

When Randomize presentation order is on, each participant sees the words in a different random order. This is useful for multi-trial designs (RAVLT-style) where presenting the same serial order every trial would inflate primacy and recency effects artificially.

The shown order is fully recorded so serial-position analysis remains valid:

• The task's summary file (verbal_learning_summary_{taskIndex}.json) includes a presentation_order array — a list of 1-based configured-list indices in the order they were shown. For example [3, 1, 2] means word 3 was shown first, then word 1, then word 2.
• Each stimulus_shown marker carries a list_index field — the word's 1-based position in the configured list — alongside word_position (its serial step in that participant's session).
• The words array in the summary file always preserves the configured (target) list order for recall scoring.

The configured list is always the target list scored against. Randomization is per-participant and reproducible when a random seed is set on the study.

Exported data (per-word and recall rows)

In the exported data and the Scoring tab, each word in the list appears as its own row (one row per presented word), plus one additional row for the recall itself when the task's mode includes recall. For example, a Presentation and recall task with a 15-word list produces 15 presentation rows plus 1 recall row (16 rows total); a Presentation only task produces 15 presentation rows and no recall row; a Recall only task produces just the 1 recall row. Word rows carry the word, its serial position, and its timing; the recall row carries the transcript and recall-window fields described above. Presentation is a passive activity (the participant is not asked to respond to each word as it is shown), so word rows never carry a right/wrong or "responded" flag — only the recall row does.

Validity caveats

• Aural presentation plays your recorded clips, so every participant hears an identical stimulus. Record the clips in a quiet environment at an even volume; use "Play all" to check the whole list sounds consistent before opening the study.
• On rare devices a clip cannot be played (for example an audio format the participant's browser cannot read). The task detects this automatically and switches the remainder of the list to on-screen presentation so no word is lost; the session is flagged with a marker so you know that participant's list was completed visually rather than aurally after that point.
• In-task speech recognition is effectively Chrome/Edge only and streams audio to the browser vendor; disclose this in your consent text (the platform already uses it for the Stroop task).
• The transcripts are automated inferences and can misrender unusual words. Score from the transcript(s), compare them when both exist, and fall back to the raw audio.
• Each participation adds recording (storage) and, during recall, browser speech recognition. A server transcript adds per-minute transcription cost only when that analysis is enabled. Chaining many instances (as in RAVLT) multiplies all of these.