Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards the Development of Balanced Synthetic Data for Correcting Grammatical Errors in Arabic: An Approach Based on Error Tagging Model and Synthetic Data Generating Model

Ahlam Alrehili, Areej Alhothali

TL;DR

This work tackles the scarcity of high-quality Arabic GEC data by introducing a two-model pipeline: an error tagging model based on DeBERTaV3 that identifies 26 ARETA error tags in correct sentences, and a synthetic data generation model using AraT5 that creates realistic erroneous sentences conditioned on those tags. The system leverages parallel data from QALB-14/15 and ZAEBUC and a large monolingual corpus to produce 30,219,310 synthetic sentence pairs, significantly expanding training resources for Arabic GEC. Empirical results show that DeBERTaV3-based tagging achieves state-of-the-art F1 on QALB-14/15 dev/test sets, while synthetic data—especially when combined with ARAT5 GEC—substantially improves correction quality (BLEU-4 and F1) compared to baselines. The approach delivers a balanced, diverse synthetic data framework for Arabic GEC with strong implications for low-resource languages and dialectal coverage, enabling more robust grammatical correction tools in real-world text processing.

Abstract

Synthetic data generation is widely recognized as a way to enhance the quality of neural grammatical error correction (GEC) systems. However, current approaches often lack diversity or are too simplistic to generate the wide range of grammatical errors made by humans, especially for low-resource languages such as Arabic. In this paper, we will develop the error tagging model and the synthetic data generation model to create a large synthetic dataset in Arabic for grammatical error correction. In the error tagging model, the correct sentence is categorized into multiple error types by using the DeBERTav3 model. Arabic Error Type Annotation tool (ARETA) is used to guide multi-label classification tasks in an error tagging model in which each sentence is classified into 26 error tags. The synthetic data generation model is a back-translation-based model that generates incorrect sentences by appending error tags before the correct sentence that was generated from the error tagging model using the ARAT5 model. In the QALB-14 and QALB-15 Test sets, the error tagging model achieved 94.42% F1, which is state-of-the-art in identifying error tags in clean sentences. As a result of our syntactic data training in grammatical error correction, we achieved a new state-of-the-art result of F1-Score: 79.36% in the QALB-14 Test set. We generate 30,219,310 synthetic sentence pairs by using a synthetic data generation model.

Towards the Development of Balanced Synthetic Data for Correcting Grammatical Errors in Arabic: An Approach Based on Error Tagging Model and Synthetic Data Generating Model

TL;DR

This work tackles the scarcity of high-quality Arabic GEC data by introducing a two-model pipeline: an error tagging model based on DeBERTaV3 that identifies 26 ARETA error tags in correct sentences, and a synthetic data generation model using AraT5 that creates realistic erroneous sentences conditioned on those tags. The system leverages parallel data from QALB-14/15 and ZAEBUC and a large monolingual corpus to produce 30,219,310 synthetic sentence pairs, significantly expanding training resources for Arabic GEC. Empirical results show that DeBERTaV3-based tagging achieves state-of-the-art F1 on QALB-14/15 dev/test sets, while synthetic data—especially when combined with ARAT5 GEC—substantially improves correction quality (BLEU-4 and F1) compared to baselines. The approach delivers a balanced, diverse synthetic data framework for Arabic GEC with strong implications for low-resource languages and dialectal coverage, enabling more robust grammatical correction tools in real-world text processing.

Abstract

Synthetic data generation is widely recognized as a way to enhance the quality of neural grammatical error correction (GEC) systems. However, current approaches often lack diversity or are too simplistic to generate the wide range of grammatical errors made by humans, especially for low-resource languages such as Arabic. In this paper, we will develop the error tagging model and the synthetic data generation model to create a large synthetic dataset in Arabic for grammatical error correction. In the error tagging model, the correct sentence is categorized into multiple error types by using the DeBERTav3 model. Arabic Error Type Annotation tool (ARETA) is used to guide multi-label classification tasks in an error tagging model in which each sentence is classified into 26 error tags. The synthetic data generation model is a back-translation-based model that generates incorrect sentences by appending error tags before the correct sentence that was generated from the error tagging model using the ARAT5 model. In the QALB-14 and QALB-15 Test sets, the error tagging model achieved 94.42% F1, which is state-of-the-art in identifying error tags in clean sentences. As a result of our syntactic data training in grammatical error correction, we achieved a new state-of-the-art result of F1-Score: 79.36% in the QALB-14 Test set. We generate 30,219,310 synthetic sentence pairs by using a synthetic data generation model.

Paper Structure

This paper contains 23 sections, 11 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Arabic Synthetic Data Techniques
  4. Arabic Grammatical Error Correction as Multi-label Classification Task
  5. BERT Pre-trained Models
  6. Text-to-Text Transformer
  7. Data
  8. Parallel Corpus
  9. Monolingual Corpus
  10. Methodology
  11. Error Tagging Model
  12. Synthetic Data Generation Model
  13. Data Generation
  14. Grammatical Error Correction
  15. Experiments
  16. ...and 8 more sections

Figures (3)

  • Figure 1: The error tagging model and the synthetic data generation model.
  • Figure 2: Label imbalances in (QALB-14, QALB-15, and ZAEBUC). Over half of the data contain some labels, while almost none occur, resulting in large label imbalances.
  • Figure 3: Examples of data generated by our AraT5 synthetic data generation model. We use our error tagging model to predict error tags. An error that is highlighted or bolded in a corrupt sentence is caused by the corruption model and is highlighted in that color.