A Comprehensive Literature Review on Deep Learning–Driven Multilingual Chatbots for Low-Resource Languages with a Focus on Marathi–Hindi–English Interaction

Abstract

Conversational Artificial Intelligence (AI) has undergone substantial progress, evolving from rule-based systems to advanced transformer-driven multilingual models.However, research for low-resource Indian languages-particularly Marathi and Hindi-remains limited despite rapid technological advances.This review synthesizes studies from 2000 to 2025, covering rule-based chatbots, retrieval methods, Seq2Seq architectures, multilingual transformers, and selfsupervised speech models such as wav2vec 2.0 and HuBERT.The analysis highlights key linguistic challenges, including agglutination, free word order, transliteration, regional accents, and pervasive code-mixing.Although models like mBERT, XLM-R, and MuRIL significantly improve multilingual understanding, they still struggle with hybrid inputs and domain-specific conversational tasks.Persistent gaps include limited datasets, weak ASR-NLU integration, and insufficient cultural grounding.The review outlines future directions for developing robust, culturally aligned Marathi-Hindi-English chatbots.

Ähnliche Arbeiten