Translation Workflow for Qualitative Interview Transcripts

A local browser workspace for translating qualitative interview transcripts in research. Transcript sections are translated in batches by forward translation, then by a blind back-translation whose request holds only the forward translation (never the original, the study context, the glossary or translator notes), and then by a reconciliation step that compares all three texts, flags discrepancies by category and severity and proposes a draft. Every draft needs a named human review decision before it counts as reviewed. Each stage can be run by copying prompt packets into any chat tool, including institutional chat services, or through the 'Mistral' or 'Anthropic' ('Claude') application programming interfaces. Exports include reviewed translations, a parallel review table, the exact prompts and replies, a hash-chained audit record and a draft methods paragraph. Projects are interchangeable with the 'Python' version of the tool.

qualitranslate (R)

QualiTranslate is a local workspace for translating qualitative interview transcripts for research. Each batch of transcript sections goes through:

1. Forward translation into the target language.
2. Blind back-translation: the request holds only the forward translation, never the original, the study context, the glossary or translator notes.
3. Reconciliation: the original, forward and back-translations are compared, discrepancies are flagged by category and severity, and a draft is proposed.
4. Named human review of every draft, before anything counts as reviewed.

The results export as a six-file research bundle (reviewed translations as CSV, a parallel review table, the project file, the exact prompts and replies, a hash-chained audit record, and a methods record with a draft methods paragraph). Projects are interchangeable with the Python version of QualiTranslate.

Install and start

install.packages("qualitranslate")
qualitranslate::run_app()

Before the CRAN release, install from a downloaded source file instead: install.packages("qualitranslate_1.5.1.tar.gz", repos = NULL, type = "source").

run_app() starts the app on http://127.0.0.1:8765 (or the next free port), opens it in your web browser and keeps running until you press Escape (RStudio) or Control+C. Your project stays in memory only: export it from the app before you stop. show_prompts() prints the translation instructions the app uses.

Ways to run each stage

• Manual: copy a prompt packet into any chat tool (for example ChatGPT, Claude, Microsoft Copilot or Mistral's Le Chat, including your institution's chat service) and paste the reply back.
• API: the Mistral or Anthropic (Claude) API with your own key. Keys stay in memory and are never exported.

Data protection

Transcript text leaves your computer only when you run a stage with a provider you chose, and then only to that provider: nothing in manual mode (you paste it into the chat tool yourself), and the selected API in API mode. Back-translation requests never contain the original. The app warns about common identifiers (e-mail addresses, phone-like and CPR-like numbers, dates, web addresses) but does not remove them and cannot detect names or places, so pseudonymise transcripts before translating them.

The app serves only the local address 127.0.0.1, checks the Host and Origin of every request and requires a per-session token.

Citation

Yalcinkaya A. QualiTranslate [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.23784148