Support for GF grammars for the South African Bantu languages
# Enabling Localised Language Technology Applications Phase II: Developing Nguni computational grammars and resources
This repository disseminates various outputs of a project aimed at developing computational grammars and resources for the Nguni languages. The primary languages targeted are isiZulu (Zul), Siswati (Ssw), isiXhosa (Xho), and Sesotho sa Leboa/Sepedi (Nso). The work is framed around Grammatical Framework (GF), a grammar formalism and toolkit for multilingual natural language processing.
The broader motivation is to support low-resource South African languages through resource grammars, application grammars, treebanks, and lexical resources — enabling applications such as mission-critical translation, literacy support, parsing, and data augmentation to address data sparsity arising from the agglutinating morphology of Bantu languages.
## Repository structure
### `corpora/`
Contains parallel sentence corpora used as source material for lexicon and grammar development. Currently includes a CSV of parallel sentences drawn from Taljaard and Bosch (2024-03-22).
### `evaluation/`
Tools and data for evaluating grammar coverage and correctness.
- `evaltools.py` — Python utilities for reading GF parse output, running linearisation tests, and comparing parsed results against reference strings using the `pgf` Python bindings.
- `regression/` — Regression treebanks (in `.treebank` and `.trees` formats) designed to continuously monitor development of the isiZulu resource grammar, with a focus on accurate modelling of the verb phrase in main and relative clauses.
### `lexica/`
GF lexicon modules for the Nguni languages, sourced from Foundation Phase CAPS dictionaries.
#### `lexica/foundation_phase/`
A multilingual lexicon (`parallel.yaml`) covering vocabulary relevant to South African foundation-phase education, with entries for Siswati, isiZulu, and isiXhosa, annotated with noun class information.
### `presentations/`
Slide decks and presentation notes …