franco334578/doric-conversations

Dataset Card for Doric Conversations

Dataset Details

Dataset Description

This dataset contains synthetic and human-reviewed conversational data designed to fine-tune Large Language Models to respond exclusively in Doric, the North-East Scots dialect. Conversations follow a chat-like structure ({"role": "user", "content": ...} / {"role": "assistant", "content": ...}) and cover everyday topics, adversarial English-only requests, multilingual prompts, and natural dialogue flow.

The assistant always replies in Doric regardless of the user's language (English/Doric/other). The dataset was designed specifically for Supervised Fine-Tuning (SFT) and uses the ChatML format.

• Language(s): Doric Scots (primary), English, and various user-prompt languages
• License: Apache-2.0

---

Uses

✅ Direct Use

This dataset is intended for:

• Fine-tuning base LLMs to respond exclusively in Doric Scots
• Evaluating multilingual → Doric translation behaviour
• Training dialectal conversational models
• Studying low-resource language adaptation

❌ Out-of-Scope Use

This dataset should not be used for:

• General English language modelling
• Formal Scots linguistics without dialect validation
• Safety-critical systems (medical, legal, financial advice)
• Deriving demographic or personal information

---

Dataset Structure

The dataset uses the ChatML format. Each row is a JSON object:

{
"messages": [
    {"role": "user", "content": ""},
    {"role": "assistant", "content": """}],
"meta":
  {
    "topic": "",
    "kind": "multi",
    "id": "",
    "lang": "",
    "group": ""
  }
}