SauDial: The Saudi Arabic Dialects Game Localization Dataset

Published: 16 September 2024| Version 1 | DOI: 10.17632/mzdwkb2t6d.1
Contributors:
,

Description

SauDial: The Saudi Arabic Dialects Game Localization Dataset is a curated collection of parallel text samples designed for video game localization. It features content in English, Modern Standard Arabic (MSA), and four major Saudi dialects: Najdi, Hijazi, Janoubi, and Eastern. The dataset was initially generated using the OpenAI GPT-4o model and refined with pre-compiled dialect-specific resources. Each entry in the dataset includes: - Original English text - MSA translation - Dialectal translation - Game context and age rating information - Linguistic notes on dialect features The content covers various game genres, scenario types, tones, and age ratings, making it versatile for different game development needs. The dataset underwent thorough cleaning and editing to ensure dialectal accuracy, tonal appropriateness, and cultural fidelity. This resource is valuable for: - Game developers and localization teams - Researchers in translation, cultural, localization, and game studies - Training and fine-tuning Large Language Models (LLMs) - Educational purposes in translation and localization studies - Professional translators and localizers as a specialized translation memory (TM) The dataset aims to streamline game localization processes and enhance the authenticity of Arabic language representation in video games, particularly for the Saudi market.

Files

Steps to reproduce

Steps to reproduce the SauDial dataset: Setup: Install required Python libraries: openai, pandas, openpyxl, tenacity Obtain an API key from OpenAI for GPT-4o access Data Generation: Initialize the SaudiDialectGameLocalizationGenerator class Define parameters: Scenario types (e.g., epic battle, marketplace haggling) Dialects (Najdi, Hijazi, Janoubi, Eastern) Game types (e.g., RPG, Adventure, Platformer) Tones (e.g., Serious, Humorous, Whimsical) Age ratings (3+, 7+, 12+, 16+, 18+) API Interaction: Generate dynamic prompts combining the above parameters Send prompts to GPT-4o API Retrieve responses containing English text, MSA translation, dialect translation, context explanations, and dialect notes Data Processing: Parse API responses Assign localization difficulty and target audience Apply content uniqueness checks Highlight dialect-specific words Data Compilation: Organize processed data into a pandas DataFrame Export data to an Excel spreadsheet using openpyxl Post-Processing: Manual review and refinement of generated content Ensure dialectal accuracy and cultural appropriateness Verify tonal consistency and age-rating adherence Quality Assurance: Conduct linguistic validation by native speakers Cross-check cultural references and idiomatic expressions Final Dataset Preparation: Compile refined entries into the final dataset format Add metadata and usage instructions Note: The exact reproduction of the dataset may vary due to the non-deterministic nature of language model outputs and the manual refinement process.

Categories

Natural Language Processing, Machine Translation, Corpus-Based Translation Studies

Licence