Multilingual corpus improves translation of Indian language dialogues
BaatCheet: A Multilingual Corpus for Dialogue Translation in Indian Languages
Computation and Language
Summary
Translating everyday conversations is harder than translating formal sentences because of informal language and how people talk to each other. The authors created a large collection of dialogues in Indian languages called BaatCheet to help improve translation in these situations. They tested multiple language models that got better when trained on this data, showing it helps translation tools understand casual speech better. They checked the translations with computers and human reviewers to make sure they worked well.
What this means in practice
- •For translation software developers: Build better translation tools that understand informal and conversational Indian languages by using the BaatCheet dialogue dataset for model training.
- •For chatbot builders: Improve multilingual chatbots to handle natural dialogues in Indian languages by training on conversation-style translations from BaatCheet.
Authors
Priyanka Dasari, Yuvrajsinh D. Bodana, Vandan Mujadia, Arafat Ahsan, Dipti Misra Sharma, Parameswari Krishnamurthy
Abstract
Existing translation models are typically trained on sentence-level and formal text, limiting their ability to capture everyday conversational dialogue phenomena such as informality, speaker interaction, and discourse coherence. Most existing Indic translation resources and evaluation benchmarks focus on sentence-level or formal text, making it difficult to assess translation quality of the dialogue phenomena. In this work, we introduce BaatCheet, a multilingual dialogue corpus named after the Hindi term for conversation or chitchat, containing approximately 49,000 dialogues for dialogue translation across five translation directions. We fine-tune five open-source LLMs across seven training data configurations and find that fine-tuning yields substantial gains over zero- and few-shot baselines. To comprehensively evaluate dialogue translation quality, we employ multiple evaluation strategies, including automatic metrics, LLM-as-judge, and human assessments using an SQM-guided Direct Assessment (DA) Protocol.