Cost-Efficient Cross-Lingual Retrieval-Augmented Generation for Low-Resource Languages: A Case Study in Bengali Agricultural Advisory

TL;DR

This paper tackles the lack of Bengali access to English agricultural manuals by proposing a translation-centric cross-lingual RAG pipeline that grounds responses in English sources and translates them back to Bengali. It leverages domain-specific keyword mapping to bridge colloquial farmer language with scientific terminology, dense vector retrieval over FAO/IRRI manuals, and a fully local, 4-bit quantized LLaMa-3 for grounded English generation, all deployed on consumer hardware. The approach yields source-grounded answers and robust out-of-domain rejection with an end-to-end latency of around 15–20 seconds, demonstrating practical deployability in resource-constrained settings. The study highlights the method's potential to democratize agricultural knowledge access, while outlining limitations such as translation quality, dialectal variation, static knowledge bases, and accessibility, and proposes future work on ASR, dialect normalization, ontology expansion, and real-world user studies.

Abstract

Access to reliable agricultural advisory remains limited in many developing regions due to a persistent language barrier: authoritative agricultural manuals are predominantly written in English, while farmers primarily communicate in low-resource local languages such as Bengali. Although recent advances in Large Language Models (LLMs) enable natural language interaction, direct generation in low-resource languages often exhibits poor fluency and factual inconsistency, while cloud-based solutions remain cost-prohibitive. This paper presents a cost-efficient, cross-lingual Retrieval-Augmented Generation (RAG) framework for Bengali agricultural advisory that emphasizes factual grounding and practical deployability. The proposed system adopts a translation-centric architecture in which Bengali user queries are translated into English, enriched through domain-specific keyword injection to align colloquial farmer terminology with scientific nomenclature, and answered via dense vector retrieval over a curated corpus of English agricultural manuals (FAO, IRRI). The generated English response is subsequently translated back into Bengali to ensure accessibility. The system is implemented entirely using open-source models and operates on consumer-grade hardware without reliance on paid APIs. Experimental evaluation demonstrates reliable source-grounded responses, robust rejection of out-of-domain queries, and an average end-to-end latency below 20 seconds. The results indicate that cross-lingual retrieval combined with controlled translation offers a practical and scalable solution for agricultural knowledge access in low-resource language settings

Figures (3)

• Figure 1: System Architecture of the proposed Translation-Centric Cross-Lingual RAG Pipeline. The system processes Bengali queries by translating them to English, enriching them with domain-specific keywords, and retrieving relevant information from English manuals before generating a grounded response.
• Figure 2: Latency Breakdown: Translation and LLM inference time compared to Retrieval time.
• Figure 3: Distribution of Retrieved Documents showing reliance on authoritative FAO and IRRI manuals.