MahaParaphrase: A Marathi Paraphrase Detection Corpus and BERT-based Models

2025-08-24

AI Summary

Paraphrase detection in Marathi—a morphologically rich, script-diverse, low-resource Indian language—faces significant challenges due to scarce annotated data. Method: We introduce L3Cube-MahaParaphrase, the first large-scale, human-annotated paraphrase detection corpus for Marathi, comprising 8,000 high-quality sentence pairs. Leveraging this resource, we conduct the first systematic evaluation of Transformer-based models (e.g., BERT) on Marathi paraphrase detection via supervised fine-tuning. Contribution/Results: The corpus fills a critical data gap for low-resource Indian languages and enables downstream applications such as question answering and data augmentation. All data and benchmark models are publicly released, establishing a reproducible, multilingual NLP benchmark and foundational infrastructure for future research.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Knowledge Representation Languages

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Large pretrained models with web data
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Paraphrases are a vital tool to assist language understanding tasks such as question answering, style transfer, semantic parsing, and data augmentation tasks. Indic languages are complex in natural language processing (NLP) due to their rich morphological and syntactic variations, diverse scripts, and limited availability of annotated data. In this work, we present the L3Cube-MahaParaphrase Dataset, a high-quality paraphrase corpus for Marathi, a low resource Indic language, consisting of 8,000 sentence pairs, each annotated by human experts as either Paraphrase (P) or Non-paraphrase (NP). We also present the results of standard transformer-based BERT models on these datasets. The dataset and model are publicly shared at https://github.com/l3cube-pune/MarathiNLP

Problem

• Detecting paraphrases in low-resource Marathi language
• Addressing limited annotated data for Indic NLP tasks
• Evaluating BERT models on Marathi paraphrase identification

Innovation

• BERT-based models for paraphrase detection
• Marathi paraphrase corpus with 8000 pairs
• Human-annotated dataset for low-resource language