The correction of grammatical errors in natural language processing is a crucial task as it aims to enhance the accuracy and intelligibility of written language. However, developing a grammatical error correction (GEC) framework for low-resource languages presents significant challenges due to the lack of available training data. This article proposes a novel GEC framework for low-resource languages, using Arabic as a case study. To generate more training data, we propose a semi-supervised confusion method called the equal distribution of synthetic errors (EDSE), which generates a wide range of parallel training data. Additionally, this article addresses two limitations of the classical seq2seq GEC model, which are unbalanced outputs due to the unidirectional decoder and exposure bias during inference. To overcome these limitations, we apply a knowledge distillation technique from neural machine translation. This method utilizes two decoders, a forward decoder right-to-left and a backward decoder left-to-right, and measures their agreement using Kullback-Leibler divergence as a regularization term. The experimental results on two benchmarks demonstrate that our proposed framework outperforms the Transformer baseline and two widely used bidirectional decoding techniques, namely asynchronous and synchronous bidirectional decoding. Furthermore, the proposed framework reported the highest F1 score, and generating synthetic data using the equal distribution technique for syntactic errors resulted in a significant improvement in performance. These findings demonstrate the effectiveness of the proposed framework for improving grammatical error correction for low-resource languages, particularly for the Arabic language.
Computational prediction of ABL1 tyrosine-kinase inhibitor (TKI) resistance is difficult because available mutation-drug datasets are small, imbalanced, and heterogeneous. This study evaluates structure-aware resistance prediction under mutation-disjoint internal validation and temporal cross-study outcome validation. The development benchmark comprised 144 ABL1 mutation-TKI records (19 resistant and 125 susceptible) labelled by a ten-fold affinity-loss criterion. A fixed panel of 11 documented structural and energetic predictors was used to minimize data-adaptive feature selection within the validation framework. A calibrated ensemble combining Elastic-Net logistic regression, radial-basis-function support-vector machines, random forests, and XGBoost was evaluated using mutation-disjoint five-fold grouped out-of-fold prediction repeated ten times. Temporal validation used a later prime-editing resistance map, yielding 84 exact mutation-TKI matches across imatinib, nilotinib, bosutinib, and ponatinib; 61 had binary Sensitive/Resistant outcomes and 23 were retained as Intermediate. The 11-feature stack achieved grouped internal AUROC 0.645 (95% CI 0.429-0.830) and PR-AUC 0.230, with mean ± SD AUROC 0.583 ± 0.101 across repeated grouped splits. On the 61 binary later-study outcomes, AUROC was 0.605 (95% CI 0.431-0.874) and PR-AUC 0.835, modestly above the resistance-prevalence baseline of 0.770. External calibration was poor (Brier score 0.583; calibration slope 0.441). Elastic-Net showed somewhat higher temporal discrimination (AUROC 0.643; PR-AUC 0.877). These results indicate limited transportability across mutation identities and experimental platforms and support use of the framework for hypothesis-generating resistance screening rather than clinically calibrated treatment selection.