Tense Identification in English Sentences Using an Ensemble Machine Learning Approach

Abstract

For efficient communication, comprehension, and writing in a variety of contexts including academic papers, legal documents, and media documents proper use of tense is essential. Notwithstanding its significance, tense usage is a common source of errors for both experts and students. By having access to a built-in tense detection tool, students can easily enhance their sentence construction skills through consistent self-study in early life education. As a solution for this, the study objective is to propose an automatic classification of English sentences using machine learning algorithms according to their tense: past, present, or future. We used a dataset with 1500 sentences that were split equally across the three tense groups. Tokenization and lowercasing were used in the preprocessing phase, and the Term Frequency-Inverse Document Frequency (TF- IDF) technique was then used to extract features. Then, six classification algorithms Naive Bayes, Random Forest, Decision Tree (J48), Support Vector Machine (SVM), Logistic Regression and ensemble learning by combining above five algorithms were tested in this research. Metrics including accuracy, precision, recall, F-measure, and error values were used in the evaluation. In terms of evaluation, the ensemble learning strategy outperformed individual models in all evaluations by achieving the best accuracy of 95.56%. In ensemble learning, the majority voting combination rule worked best in 70% training data. This work shows how machine learning may improve tense classification, providing a useful tool for both academic and professional contexts.