kimgerdes/basaa-treebank
Domain:
natural language processingRecord type:
dataset
Creator:
kim
Host:
First glossed, translated and audio-aligned SUD dependency treebank of Basaa (Bàsàá, A43, ISO bas). Annotations CC-BY-SA-4.0; copyrighted audio/PDFs referenced, not bundled.
# Basaa (A43) treebank
Materials toward **the first glossed, translated and audio-aligned dependency treebank of Basaa**
(Bàsàá, ISO 639‑3 `bas`, Glottolog `basa1284`, Guthrie **A43** / group A.40, Cameroon).
- 🌳 **Browse & annotate (with playable audio):** the *Bàsàá* project on Arborator-Grew —
- 🔊 **Demo / data hub:** (sentence list with in-browser audio players)
- The treebank is annotated in **SUD** (Surface-syntactic Universal Dependencies), with a UD copy of
the literature sample for comparison.
## What's here
| Path | What |
|---|---|
| `treebank/` | The treebank: **`basaa_audio.arborator.sud.conllu`** (21 audio-aligned sentences), SUD/UD samples, the forced-alignment scripts, and the syntactic analysis + upload notes |
| `glossary/` | `basaa_glosses.tsv` (~90 common words, glosses, noun classes); the *child*-paradigm morphology; the transcription & gloss review |
| `rules/` | Annotation rules extracted from the literature, by pipeline stage — transcription, glossing, parsing, and flagged contradictions |
| `recordings/` | **Metadata only** (manifests, forced-alignment offsets, Praat TextGrids, provenance READMEs). Audio is **not** in this repo — see §Copyright |
| `corpora/` | Provenance/README for the text & parallel data (OPUS); raw dumps not bundled |
| `docs/` | `SOURCES.md` + `resources.md` (annotated bibliography). The PDFs themselves are **not** redistributed |
| `TREEBANK_PLAN.md` | Gap analysis & roadmap to scale beyond this seed |
| `treebank/index.html` | The demo page deployed at elizia.net |
## Status
- **21 audio-aligned sentences**: 10 isolated "child" utterances (clean — transcription, Hamlaoui-style
glosses, SUD parse, **word-level** forced alignment) + 11 *North Wind & the Sun* clause units
(sentence-level timing reliable; word forms & parse **draft**, to re-transcribe by ear).
- Audio time-aligned with **torchaudio's MMS** multilingual forced aligner; each token is clickable in
Arborator-Grew.
- A separate 5-sentence **lite …