DilatoMX/G11n_GenAI_Prompt_Datasets

🌍 GenAI G11n Prompt Evaluation Dataset

This dataset was developed to support manual evaluation of multilingual generative AI models based on coherence, translation accuracy, cultural adaptation, and multimodal consistency.

📦 Dataset Contents

The dataset includes:

• Prompts Type: A list of types of prompts for testing linguistic behavior in different contexts.
• Placeholders: A list of the used placeholders to localize the prompts usually from the locale's word dataset.
• Locale Folder: All data related specifically to one locale.
  • Assessment Categories:
    • Language & Grammar
    • Cultural Adaptation
    • Instruction & Response
    • Multimodal Consistency
  • Word Dataset: Collection of culturally relevant words for that specific locale.
  • Formatting: Guidelines and other aspects to evaluate beyond translation, specifics for each locale.

🧪 Purpose

This dataset enables the manual assessment of AI models on tasks such as:

• Following multilingual instructions
• Handling ambiguity in translation
• Adapting responses to regional or cultural nuances
• Generating appropriate outputs in audio, image, and code formats

✅ Evaluation Criteria

The prompts are designed to evaluate the following success criteria:

• Instruction & Response Coherence
• Translation Accuracy
• Cultural Adaptation
• Multimodal Consistency

👥 Evaluation Team

• Andres Castillo – G11n QA
• Edgar Castillo – G11n QA
• Patricia Oceguera – Linguistic Advisor
• Marcela Salgado – Review Support

📊 Evaluated AI Model Versions

• ChatGPT – GPT-4o
• Gemini – 2.0 Flash
• Copilot – GPT-4o
• DeepSeek – V3

📄 License

This dataset is released under the CC BY 4.0 License, allowing sharing, modification, and redistribution with proper attribution.

📌 Additional Notes

For detailed instructions on how to run the evaluation using this dataset, go to the model page:
G11n_GenAI_Assessment_Model under the organization Dilato Infotech Limited here on Hugging Face.