Evaluating automatic sentence alignment approaches on English-Slovak sentences.

Area of Science:

• Natural Language Processing
• Computational Linguistics

Background:

• Parallel corpora are crucial for Natural Language Processing (NLP) applications.
• Sentence alignment, identifying sentence correspondences between texts, is a fundamental step in creating parallel corpora.
• Existing automatic sentence alignment methods include length-based, lexicon-based, translation-based, and neural network approaches.

Purpose of the Study:

• To evaluate the performance of Bertalign against established sentence alignment tools.
• To compare the effectiveness of five different aligners: Bilingual Sentence Aligner (BSA), Hunalign, Bleualign, Vecalign, and Bertalign.
• To assess performance on a custom English-Slovak parallel corpus.

Main Methods:

• Utilized five distinct sentence alignment tools: BSA, Hunalign, Bleualign, Vecalign, and Bertalign.
• Developed a custom parallel corpus using texts from 2021-2022 for the English-Slovak language pair.
• Evaluated aligner accuracy and performance metrics.

Main Results:

• Vecalign and Bertalign achieved statistically significant best performance.
• Bilingual Sentence Aligner (BSA) exhibited the poorest performance.
• Hunalign and Bleualign showed comparable F1 scores, though Bleualign yielded more varied results.

Conclusions:

• Vecalign and Bertalign represent the most effective tools for English-Slovak sentence alignment based on the evaluated metrics.
• The study highlights performance disparities among different automatic sentence alignment approaches.
• Further investigation into Bleualign's performance variability may be warranted.