Abstract
Neural machine translation (NMT/MT) replicates human brain networks and is often considered a “black box” due to its unexplained structure and function. To use NMT effectively, target-language (TL) learners should pre-edit source-language (SL) documents, similar to how humans paraphrase first-language (L1) texts before translating to second-language (L2) (Tsuji, 2024). Previous studies have shown that pre-editing SL texts improves the quality of MT output. However, the specific differences between NMT output with and without editing have been explored less. This study aims to compare NMT translations with and without pre-editing and examine the impact of pre-editing on the quality of TL texts. The study involved 54 Japanese university students with intermediate English proficiency, each composing a Japanese text which was translated into English using NMT. Three language researchers analyzed issues in unedited NMT outputs, and systematically categorised them into lexical, grammatical, semantic and formatting errors. Comparing the results with those of Tsuji (2024) exploring issues in pre-edited NMT outputs showed that NMT output without pre-editing displayed more frequent errors than NMT output with pre-editing. In particular, the semantic issues in unedited NMT output significantly limited readability. On the contrary, the pre-edited NMT translations described in Tsuji (2024) were generally comprehensible and contained minor errors that did not significantly affect comprehensibility. This study indicated that pre-editing SL texts is essential for improving the quality of NMT translations. Errors illustrated in this study could potentially be used as an indication for students learning L1 paraphrasing skills required for the human L2 translation process.
Author Information
Kayo Tsuji, Osaka Metropolitan University, Japan
Benjamin Neil Smith, Kwansei Gakuin University, Japan
Kiyo Okamoto, Osaka Metropolitan University, Japan
Paper Information
Conference: ACE2024
Stream: Foreign Languages Education & Applied Linguistics (including ESL/TESL/TEFL)
This paper is part of the ACE2024 Conference Proceedings (View)
Full Paper
View / Download the full paper in a new tab/window
To cite this article:
Tsuji K., Smith B., & Okamoto K. (2025) The Efficacy of Pre-editing as a Means of Improving NMT Output ISSN: 2186-5892 – The Asian Conference on Education 2024: Official Conference Proceedings (pp. 641-650) https://doi.org/10.22492/issn.2186-5892.2025.57
To link to this article: https://doi.org/10.22492/issn.2186-5892.2025.57
Comments
Powered by WP LinkPress