toutoutoubun/Imbizo

a local research workbench for people studying code-switching in South African multilingual interview data # Imbizo-CS Workbench Imbizo-CS Workbench is a local research workbench for people studying code-switching in South African multilingual interview data. It is designed for humanities and social-science researchers who need to keep their material on their own computer. The app works with local project folders: your transcripts, annotations, logs, dictionaries, and exports stay inside the folder you choose. Imbizo-CS Workbench is not a cloud assistant. It does not require an account, subscription, API key, telemetry, or internet connection for core work. ## What You Can Do In The MVP - Create and open a local research project. - Import TXT, CSV, TSV, JSON, XLSX, ODS, ELAN EAF, Praat TextGrid, audio, and video files. - Continue annotating imported transcript data at token level. - Keep original transcript text separate from optional normalized text. - Add or override language labels manually. - Record Matrix Language, Embedded Language, switch type, linguistic status, trigger text, confidence, tags, memos, and morpheme splits. - Run local language-identification suggestions that remain editable. - Compute language proportions, switch count, switch density, dominant language, M-index, I-index, burstiness, trigger tables, and KWIC concordance. - Export CSV, XLSX, JSON, EAF, TextGrid, HTML, PDF, and quotation extracts. ### Optional Coarse Language Group Gate Local LID suggestions can now be run with an optional coarse language group gate. This is not a final language classifier. It groups candidate evidence into broad South African language families such as Germanic, Nguni, and Sotho-Tswana, then blocks only risky auto-annotation when the evidence is too weak or when closely related languages are hard to separate. Suggestions are still saved in `lid_suggestions.evidence_json` for audit and review. The gate is off by default, never overwrites manual or imported annotations, and does not change Matrix Language, Embedded Language, switch-type, tags, memos, morpheme s …