FCE Veri Setindeki Açıklamalı İngilizce Dil Hatalarının Makine Öğrenmesi ve Derin Öğrenme Yaklaşımlarıyla Sınıflandırılması
Year: 2026
Type: conference-paper
Abstract: This study aims to classify annotated grammatical errors in written texts produced by learners of English as a Foreign Language (EFL) using machine learning, deep learning, and large language model-based approaches. To this end, errorcorrection instances were extracted from the *edits* field of the FCE (First Certificate in English) dataset, and model inputs were constructed using various information sources, including contextual text, POS-tagged context, erroneous words, corrected words, and error-correction pairs. Although the dataset originally contains 75 fine-grained error labels, each detailed label was merged with its corresponding higher-level category to mitigate the effects of class imbalance, resulting in 12 general error categories, such as verb, noun, and preposition/determiner errors. Among these 12 categories, the four linguistically meaningful classes with the largest number of samples were selected to formulate the classification task. The performance of SVM, BERT-base-uncased, TinyLlama-Embedding combined with LinearSVC, BiLSTM, and LightGBM models was comparatively evaluated using an 80/20% train-test split. Experimental results demonstrated that the BERT-base-uncased model achieved the highest performance, attaining an F1-score of 93%. Among the conventional machine learning methods, SVM yielded the best classification performance.
Institution: Marmara University
Language: Turkish
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7223686480