Shabda Chitra: Multi-Agentic Framework for Emotion - Conditioned Bengali Poetry Generation

ACL ARR 2025 July Submission294 Authors

Abstract: Shabda Chitra is a multi-agentic AI framework designed to produce Bengali poetry through emotion-conditioned topic modeling. We first identified latent themes from a corpus of Bengali texts using Structural Topic Modeling (STM), capturing nuanced socio-cultural and semantic dimensions. These topics are then used to condition a large language model within a collaborative multi-agent architecture, which composes original Bengali poems that reflect the core ideas of each identified topic while expressing target emotional states. By integrating emotion-aware topic modeling with specialized agent coordination, Shabda Chitra offers a structured yet creative approach to computational poetry in a low-resource language setting.

Paper Type: Long

Research Area: Efficient/Low-Resource Methods for NLP

Research Area Keywords: poetry generation, multi-agent systems, low-resource languages, topic modeling, emotion-aware generation, Bengali NLP, cultural AI, computational linguistics

Contribution Types: Model analysis & interpretability, Approaches to low-resource settings, Publicly available software and/or pre-trained models, Data resources

Languages Studied: Bengali

Submission Number: 294