NagaNLP Conversational Corpus
The NagaNLP Conversational Corpus, as part of the NagaNLP toolkit, is a human validated conversational dataset for Nagamese Creole, containing 10,018 instruction–conversation pairs and 311,684 tokens.
About Dataset
The NagaNLP Conversational Corpus, as part of the NagaNLP toolkit is a large-scale conversational resource developed for Nagamese Creole, a creole widely used as a lingua franca across Nagaland and neighboring Northeast Indian states. It contains 10,018 instruction–conversation pairs and 311,684 tokens covering everyday conversational scenarios (greetings, requests, small talk, general Q&A) in Nagamese. It gives a user prompt in Nagamese, generate an appropriate assistant response in Nagamese as well. The dataset is described in: Agniva Maiti, Manya Pandey, and Murari Mandal. (2025). NagaNLP: Bootstrapping NLP for Low-Resource Nagamese Creole with Human-in-the-Loop Synthetic Data. arXiv preprint arXiv:2512.12537. This dataset was identified and facilitated for onboarding as part of the Dataset Onboarding Support Team (DOST) initiative led by CivicDataLab (CDL), partnering with the Gates Foundation in collaboration with BHASHINI. CivicDataLab provided support for dataset discovery, validation, metadata preparation, and onboarding facilitation. All dataset ownership and intellectual property rights remain with the original author(s).
Purpose of Dataset
This Dataset is part of the NagaNLP Toolkit, the purpose of the NagaNLP Conversational Corpus is to provide foundational conversational data for the development and evaluation of generative NLP systems for Nagamese Creole. It can be used for instruction tuning, conversational language modelling, dialogue generation, question answering, chatbot development, summarization, machine translation, multilingual and cross-lingual learning, and research on low-resource and creole languages.
Dataset Metadata
Attribution-Non-Commercial 4.0 International (CC BY-NC 4.0)
Northeast India
Science, Technology and Research
Agniva Maiti
-
Structured
Static
Static
N.A.
26/08/26 07:22:43
Open
Hosted
Hybrid
The Conversational Corpus Was Created Using A Multi Stage Llm-to-human Bootstrapping Pipeline. Google Gemini 2.5 Pro Was First Guided Through Persona And Task Definition, Interactive Grammatical Elicitation, And Knowledge Consolidation Using Authentic Nagamese Materials Such As Articles And Social Media Posts. Once A Structured Understanding Of Nagamese Grammar And Vocabulary Had Been Established, Structured Few-shot Prompting With In-context Reinforcement Was Used To Generate A Large Set Of Conversational Instruction Pairs. The Model's Core Grammatical Rules And Authentic Nagamese Texts Were Periodically Reintroduced Between Generation Batches To Reduce Context Drift And Maintain Linguistic Consistency. After That It Underwent A Single, Crucial Human Validation Step. The Raw Generated Conversational Pairs Were Reviewed And Corrected By Native Speakers To Fix Any Grammatical Errors, Unnatural Phrasing, Or Logical Inconsistencies.
Activity Overview
0
3 
0 
23
Tags
- Large Language Model
- conversational
- Instruction-Tuning
- low-resource
- Nagamese
- Conversational AI
- northeast-india
- Nagaland
- Conversational Data
- Dialogue
- Low Resource NLP
- Low Resource Languages
- Natural language processing
- Corpus linguistics
- Conversational corpus
- Question answering
- Instruction tuning
- Text generation
- Language resources
- naga
License Control
Attribution-Non-Commercial 4.0 International (CC BY-NC 4.0)
Related Datasets
-- 0
- 4
- 0
21
-- 0
- 2
- 0
- 28
-- 0
- 3
- 0
- 26
© 2026 - Copyright AIKosh. All rights reserved.