Leveraging Synthetic Data for Question Answering with Multilingual LLMs in the Agricultural Domain

TL;DR

This paper improves multilingual agricultural question answering by generating synthetic datasets in multiple languages and fine-tuning LLMs, resulting in more accurate and relevant responses for farmers in India.

Contribution

It introduces a method for creating multilingual synthetic datasets from agricultural documents and fine-tuning LLMs, enhancing QA performance in local languages.

Findings

Significant improvements in factuality and relevance of answers

Enhanced agricultural consensus in responses

Effective use of synthetic data for multilingual fine-tuning

Abstract

Enabling farmers to access accurate agriculture-related information in their native languages in a timely manner is crucial for the success of the agriculture field. Publicly available general-purpose Large Language Models (LLMs) typically offer generic agriculture advisories, lacking precision in local and multilingual contexts. Our study addresses this limitation by generating multilingual (English, Hindi, Punjabi) synthetic datasets from agriculture-specific documents from India and fine-tuning LLMs for the task of question answering (QA). Evaluation on human-created datasets demonstrates significant improvements in factuality, relevance, and agricultural consensus for the fine-tuned LLMs compared to the baseline counterparts.