Abimvikky/Yoruba-Constituency-Treebank

A manually annotated Yoruba constituency treebank containing 1,000 sentences with POS tags, phrase structures, translations, and notes for Yoruba NLP research. # Yoruba Constituency Treebank (Version 1.0) ## Overview This repository contains the **Yoruba Constituency Treebank**, a manually annotated collection of **1,000 Yoruba sentences** developed as part of an undergraduate research project in linguistics. The treebank is intended to support both **theoretical syntactic analysis** and **computational experiments** for an under-resourced language. The dataset captures a wide range of Yoruba sentence types, including simple, compound, complex, interrogative, imperative constructions, and serial verb constructions. Sentences were drawn from grammar books, the Yoruba Bible, BBC Yoruba articles, literary texts, and naturally occurring spoken Yoruba. Each sentence was manually annotated with constituency-based phrase structure trees, making the dataset suitable for linguistic analysis and machine learning applications. In addition to the treebank, the project experimented with **fine-tuning a transformer-based constituency parser** using the manually annotated data. This approach differs from traditional rule-based or English-trained parsers and reflects a data-driven method adapted to Yoruba syntax. --- ## Repository Contents | File | Description | | ---------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- | | `yoruba_treebank.csv` | The annotated dataset containing Yoruba sentences, English translations, sentence types, POS tags, constituency trees, and annotation notes. | | `yoruba_sentences.txt` | Raw Yoruba sentences without annotations; used as input for computational parsing experiments. | | `yoruba_eval.json` | Evaluation file derived from the treebank, used for testing parser outputs against gold-standard tre …