README.md

metadata

configs:
  - config_name: default
    data_files:
      - split: Dialect2SQL
        path: Dialect2SQL.csv
      - split: train
        path: train.csv
      - split: validation
        path: validation.csv
      - split: test
        path: test.csv
task_categories:
  - text-generation
  - translation
language:
  - en
  - ar
tags:
  - text-to-sql
  - question-to-sql
  - nlq-to-sql
  - SQL
  - english-to-sql
  - low-resource-languages
  - darija
  - Arabic-dialect
size_categories:
  - 1K<n<10K

Dialect2SQL

Dataset Description

Dialect2SQL is a novel dataset designed for the Text-to-SQL task in Arabic dialects, with a particular focus on Moroccan Darija.
It provides natural language questions written in Darija, paired with corresponding SQL queries and database schemas.
The dataset enables research on low-resource natural language interfaces to databases (NLIDB) in non-standard Arabic varieties.

---

Dataset Summary

Dialect2SQL aims to bridge the gap between Arabic dialects and structured query understanding.
The dataset consists of manually and semi-automatically curated Darija–SQL pairs mapped to realistic database schemas spanning multiple domains (e.g., education, e-commerce, banking, and transportation).
Each entry contains:

• a Darija question (darija_question)
• an optional English translation (en_question)
• the SQL query (sql)
• the database schema (db_schema)
• the database name (db_id)

---

Supported Tasks and Leaderboards

Task: Text-to-SQL / NLQ-to-SQL

Given a natural language question written in Darija (Moroccan Arabic), the goal is to generate the correct SQL query that retrieves the requested information.

Example Task:

Field	Example
darija_question	شحال من تلميذ عندو أكثر من 15 ف الرياضيات؟
en_question	How many students scored more than 15 in mathematics?
sql	SELECT COUNT(*) FROM Grades WHERE subject = 'Mathematics' AND grade > 15;
db_schema	CREATE TABLE Grades(student_id number, subject varchar, grade real);
db_id	School_DB

---

Languages

• Darija (Moroccan Arabic) — primary source language
• English — provided for reference and cross-lingual studies

---

Data Splits

Split	Size (approx.)	Description
train	~7,000	Main training set
validation	~1,000	Development split
test	~1,000	Evaluation split
Dialect2SQL	full dataset	Combined dataset file

---

Dataset Structure

Each row in the dataset includes:

Column	Description
db_id	Database identifier
db_schema	SQL table definitions
darija_question	Question in Moroccan Darija
en_question	English translation (optional)
sql	Target SQL query

---

Use Cases

• Fine-tuning text-to-SQL models for Arabic dialects
• Research on multilingual and dialectal NLIDB systems
• Cross-lingual transfer learning for SQL understanding
• Evaluating low-resource adaptation of code LLMs (e.g., Qwen, StarCoder, Codex)

---

Limitations

• The dataset currently focuses on Moroccan Darija, and performance may not generalize to other Arabic dialects.
• Some questions are written using Arabic script, while others mix Latin characters (Arabizi), reflecting real user input diversity.
• SQL coverage is limited to single-domain, schema-bounded tasks.

---

Citation

If you use this dataset, please cite the following paper:

@inproceedings{chafik2025dialect2sql,
  title={Dialect2SQL: A Novel Text-to-SQL Dataset for Arabic Dialects with a Focus on Moroccan Darija},
  author={Chafik, Salmane and Ezzini, Saad and Berrada, Ismail},
  booktitle={Proceedings of the 4th Workshop on Arabic Corpus Linguistics (WACL-4)},
  pages={86--92},
  year={2025}
}