huntila/reproducible-ethiopic-qa
Domaine:
natural language processingType de record:
dataset
Créateur:
hun
Hôte:
Repository for Reproducible Ethiopic QA.
# Ensuring Reproduciblity in Ethiopic Language Question Answering
## Question Answering (QA)
Question Answering (QA) systems answer natural language questions using specific underlying data sources. Ethiopic QA, in particular, is dedicated to responding to questions posed in Ethiopic languages, such as Amharic, Tigrinya, Afan Oromo, Somali, and others.
As shown in the figure below, it is a de facto to use datasets for training, evaluating, and comparing QA models.
In Ethiopic QA, despite the existence of QA models for languages such as Amharic, Tigrinya, and Afan Oromo, only a few publicly available datasets exist (See Dataset table). Furthermore, these datasets often lack detailed information in their accompanying dataset cards and do not always provide corresponding baseline models.
| Dataset | Language | Publicly Available Repo | Hugging Face (HF) Data Card | Method of Creation | Baseline Model | Baseline Code |
|--------------------------------------|----------|-------------------------|-----------------------------|-------------------------------------|----------------|---------------|
| Amh-QuAD Taffa et al., 2024 | Amharic | GitHub & HF | Yes | Crowd-sourcing | Yes | GitHub |
| AmaSQuAD Hailemariam et al., 2025 | Amharic | HF | Yes | Translation | No | - |
| Question Classification Destaw et al., 2022 | Amharic | HF | No | Collected from Community-based QA platform | Yes | GitHub |
| TigQA Teklehaymanot et al., 2024 | Tigrinya | GitHub & Zenodo | No | Expert curated | Yes | GitHub |
| TiQuAD Gaim et al., 2023 | Tigrinya | No | No | Crowd-S …