Facebook pixel tracking

The Science and Information (SAI) Organization publishes open-access peer-reviewed journals in computer science and artificial intelligence.

Contact Info
Website thesai.org
Follow Us
Contact Info
Follow Us
Research Article | Open Access |

Confidence-Gated Indonesian Grammatical Error Correction Under Synthetic-to-Authentic Domain Shift

Author 1: Eko Suroso Author 2: Onok Yayang Pamungkas Author 3: Laily Nurlina Author 4: Sudiyono Author 5: Jimat Susilo Author 6: Eko Priyanto Author 7: Hera Septriana
International Journal of Advanced Computer Science and Applications (IJACSA) · Vol. 17, No. 9 · Published 2026

DOI: 10.14569/IJACSA.2026.0170927

Abstract

Indonesian grammatical error correction (GEC) requires decisions about when to apply a proposed edit under synthetic-to-authentic domain shift. We evaluate a selective correction framework using a reconstructed, fully executable experiment. An audit of 1,955 authentic records removed 36 incomplete pairs, 14 duplicates, 45 string identities, and five additional to-ken identities, leaving 1,855 pairs. Exact-text groups were partitioned into 1,297 training, 280 validation, and 278 test pairs. Synthetic priors were learned from 200,000 public development pairs. A cross-fitted logistic ranker combined synthetic mappings, authentic templates, spelling retrieval, and local language-model evidence. Thresholds were chosen on validation data under joint degradation and overcorrection budgets. At the 2% budget, the gate accepted five source edits, four correct: precision 80.00%, F0.5 4.03, coverage 1.80%, and word-error reduction 0.88%. No test source was degraded, but one of 278 correct-target controls changed. The precision interval was 37.55–96.38%. Removing the synthetic channel reduced available gold candidates from 110 to 109 and accepted true positives from four to three. Replacing the language-model feature with frozen IndoBERT produced 66.67% precision at 2.16% cover-age, with one degraded source and one changed control. The results support restrictive authorization on this corpus, show limited unique synthetic contribution, and expose a substantial candidate-recall bottleneck. They do not establish broad correction ability or readiness for educational deployment.

Keywords

How to Cite this Article

Suroso, E., Pamungkas, O. Y., Nurlina, L., Sudiyono, Susilo, J., Priyanto, E., & Septriana, H. (2026). Confidence-Gated Indonesian Grammatical Error Correction Under Synthetic-to-Authentic Domain Shift. International Journal of Advanced Computer Science and Applications, 17(9). https://doi.org/10.14569/IJACSA.2026.0170927

Suroso, Eko, et al.. "Confidence-Gated Indonesian Grammatical Error Correction Under Synthetic-to-Authentic Domain Shift." International Journal of Advanced Computer Science and Applications, vol. 17, no. 9, 2026, https://doi.org/10.14569/IJACSA.2026.0170927.

@article{Suroso2026,
  title     = {Confidence-Gated Indonesian Grammatical Error Correction Under Synthetic-to-Authentic Domain Shift},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {9},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Eko Suroso and Onok Yayang Pamungkas and Laily Nurlina and Sudiyono and Jimat Susilo and Eko Priyanto and Hera Septriana},
  doi       = {10.14569/IJACSA.2026.0170927},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170927}
}

Open Access — licensed under a Creative Commons Attribution 4.0 International License. Unrestricted use, distribution, and reproduction in any medium, even commercially, as long as the original work is properly cited.

Call for Papers - Important Dates

Volume 17 No. 11, November 2026
Paper Submission Due October 25, 2026
Review Notification November 15, 2026
Publication Date November 30, 2026