Beyond Chunking: Node Integrity and Field Structure in Bengali Retrieval-Augmented Agent Memory
Abstract
This work studies how chunking and record structure affect Bengali retrieval-augmented systems and agent memory. It compares chunked text, intact full-node text, and field-structured retrieval across agricultural, legal, and secondary-education domains, examining both retrieval quality and downstream evidence grounding.
Key Empirical Receipts & Quantitative Highlights
- Compares chunked text, intact knowledge nodes, and fielded records across three Bengali domains.
- Studies whether preserving record integrity improves retrieval and citation grounding in agent memory.
Problem Context & Real-World Motivation
Low-resource retrieval systems often split records into chunks, which can separate a query-matching phrase from the field that identifies its source record.
Principal Research Contributions
Compares three knowledge representations: chunked text, intact full-node text, and fielded records.
Evaluates retrieval across agricultural advisory, statutory law, and secondary-science education.
Extends the representation comparison to retrieval-grounded agent memory and citation grounding.
BibTeX Citation
@inproceedings{reza2026beyondchunking,
title={Beyond Chunking: Node Integrity and Field Structure in Bengali Retrieval-Augmented Agent Memory},
author={Reza, Khan Raiyan Ibne and Maria, Sanjana Aktar and Nimi, Sumaiya Tabassum},
booktitle={Proceedings of the Third Workshop on KNOwledge GRaphs & Agentic Systems Interplay (NORA), AACL-IJCNLP},
year={2026}
}