Usability study subtitles that keep the task in context.

Make research clips easier to review without rewriting the evidence

A short usability-study clip can help a team revisit a difficult moment in a product. Polyglot can transcribe the authorized recording locally on an Apple Silicon Mac, let you review timed text, translate it and export separate SRT or WebVTT files. The research task is to preserve enough context for the team to understand what the participant was trying to do and what actually happened.

This workflow is for recordings you are authorized to process and share with a defined research audience. Keep a small context record beside each clip. Subtitles convey speech and necessary caption context; your observations and findings belong in separately labeled research notes. Polyglot does not analyze usability, identify speakers, select highlights or decide whether a task succeeded.

Approve the source cut and the people who can see it

Check the study’s permission record before importing or sharing a recording. It should cover the planned processing, subtitle or translation work, intended viewers and retention arrangements. Permission for an internal study does not automatically cover a public portfolio, sales presentation or social-media reel. Resolve uncertainty with the study owner before proceeding.

The GOV.UK user-research consent guide emphasizes explaining recording and sharing, and linking consent evidence to the corresponding research data. Follow your organization’s applicable requirements and agreed withdrawal process. Keep permission records private and retain a way to locate each distributed derivative if its use must change.

Prepare the approved clip in suitable editing tools first. Check faces, voices, screen content, account details, notifications and incidental speech. Removing a name from a subtitle does not remove it from the audio or picture, and a participant code does not make a recognizable recording anonymous. Polyglot does not redact or anonymize recordings. Local processing also does not replace access controls or authorization; the privacy page explains the app’s separate network boundaries.

Attach a task and intervention record to each clip

Record the study round, an approved participant code, the task goal, prototype revision, source recording time, final clip filename and intended audience. Note any moderator prompt or help given before the excerpt. GOV.UK’s moderated-testing guide describes participants attempting specific tasks and thinking aloud; those conditions matter when interpreting a short clip.

Keep the relevant question or prompt audible when practical. Otherwise, add clearly labeled context to the accompanying research notes or video introduction. Do not insert an unspoken task instruction into a participant’s subtitle. If an edit hides the intervention that enabled progress, extend or revise the clip before creating the final track.

This newly written fictional example shows a manual review record. It contains no real participant information and is not a transcription benchmark or an app-generated analysis.

Fictional clip review
Round: R2 / participant P04
Prototype: booking-v07
Task: change a booking date
Source: 08:14–08:31
Clip: date-task-v03.mp4
Audience: study team only

Moderator at 08:12:
“Try the calendar icon.”
Keep prompt in clip context.

Heard participant speech:
“I thought this saved it.”
Draft:
“I think this saved it.”
Action: restore “thought”.

Observed:
Returned to overview.
Outcome:
Save not confirmed in clip.
Finding:
Review with other evidence;
not a participant quotation.

Check who spoke, what they said and what the screen showed

Replay the approved cut while reading every cue. Distinguish participant speech from moderator speech and any audible device output, such as a screen reader. Add only verified labels where the speaker is otherwise unclear. Speech recognition does not establish speaker identity. Do not guess a label when overlapping voices make the source uncertain.

Preserve words that carry the participant’s mental model: “I thought”, “maybe”, “again” or “I expected”. Correct recognition errors without polishing uncertainty into confidence. In the example, “thought” refers to a prior expectation; changing it to “think” changes the evidence. If a phrase cannot be resolved from the recording, flag it for review rather than inventing a fluent sentence.

Compare the visible action with your observation note separately. Clicking a control is not proof that a change was saved. GOV.UK’s analysis guidance separates what people saw or heard from the findings they later draw. Keep the subtitle faithful even when the participant’s belief conflicts with the interface. Explain that conflict in the analysis, not by correcting the participant’s words.

Review meaningful non-speech audio and audible device feedback where needed to understand the clip. W3C describes captions as synchronized speech and relevant non-speech information, and calls for checking automatic results. Subtitles alone do not describe a silent cursor movement or provide a complete account of the visual task; plan any necessary description separately.

Translate uncertainty and interface references carefully

Review the source wording before translating. Have a reviewer competent in the target language check tense, hesitation, negation and the distinction between an expectation and a confirmed result. A translation should not make a confused participant sound as if they understood the task. Keep moderator help distinct from participant speech.

When a participant names an interface label, retain a usable connection to the label visible in that prototype. Record any explanatory translation in the companion context notes if needed; do not silently substitute a newer interface term. Polyglot preserves source timing and validates translated cue structure, but these checks cannot establish research meaning or task success. Review each delivered language and revisit affected translations after source corrections.

Accept the clip, track and context notes together

Export SRT or WebVTT according to the destination player’s requirements. Attach the file to the exact approved cut and play it with the captions enabled. A later highlight-reel edit needs another timing review; a track from the full session does not automatically fit a shortened clip. Polyglot exports separate subtitle files rather than burning text into the picture or uploading it to a research repository.

Check the opening prompt, moderator interventions, participant responses and the clip’s ending. On the intended screen size, verify that captions remain readable without covering the control or status message under discussion. Check the destination’s placement options where necessary; Polyglot is not an individual cue-positioning editor.

Confirm that the context note names the same task and prototype revision as the video. Keep observations, interpretations and next actions visibly separate. A compelling excerpt illustrates a moment; it does not by itself establish how common the issue was. Resolve any finding that depends on missing context before presenting it.

Finally, inspect the video, all subtitle languages, companion notes, filenames and sharing settings for excluded information. Deliver only the approved versions to the authorized audience, with a record of their locations and review status. Keep the original research material and permission evidence under the study’s retention and access arrangements. Release the bundle only when both caption playback and research-context review pass.

Keep exploring.