# ViraPart: A Text Refinement Framework for Automatic Speech Recognition and Natural Language Processing Tasks in Persian

arXiv: https://arxiv.org/abs/2110.09086 · HTML: https://mlsanity.com/paper/2110.09086/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 7 of its 35 references that are in the map.
- First posted on arXiv 18 Oct 2021.

## Builds on (most cited first, 7 of 7)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [ParsBERT: Transformer-based Model for Persian Language Understanding](https://mlsanity.com/paper/2005.12515/) (2020)
- [Punctuation Prediction Model for Conversational Speech](https://mlsanity.com/paper/1807.00543/) (2018)
- [Automatic punctuation restoration with BERT models](https://mlsanity.com/paper/2101.07343/) (2021)
- [Persian Ezafe Recognition Using Transformers and Its Role in Part-Of-Speech Tagging](https://mlsanity.com/paper/2009.09474/) (2020)
- [Joint Persian Word Segmentation Correction and Zero-Width Non-Joiner Recognition Using BERT](https://mlsanity.com/paper/2010.00287/) (2020)

## Built on by (most cited first, 1 of 1)

- [PersianPunc: A Large-Scale Dataset and BERT-Based Approach for Persian Punctuation Restoration](https://mlsanity.com/paper/2603.05314/) (2026)

## Abstract (arXiv)

The Persian language is an inflectional subject-object-verb language. This fact makes Persian a more uncertain language. However, using techniques such as Zero-Width Non-Joiner (ZWNJ) recognition, punctuation restoration, and Persian Ezafe construction will lead us to a more understandable and precise language. In most of the works in Persian, these techniques are addressed individually. Despite that, we believe that for text refinement in Persian, all of these tasks are necessary. In this work, we proposed a ViraPart framework that uses embedded ParsBERT in its core for text clarifications. First, used the BERT variant for Persian following by a classifier layer for classification procedures. Next, we combined models outputs to output cleartext. In the end, the proposed model for ZWNJ recognition, punctuation restoration, and Persian Ezafe construction performs the averaged F1 macro scores of 96.90%, 92.13%, and 98.50%, respectively. Experimental results show that our proposed approach is very effective in text refinement for the Persian language.