Date of Award
12-2025
Culminating Project Type
Starred Paper
Styleguide
apa
Degree Name
Computer Science: M.S.
Department
Computer Science and Information Technology
College
School of Science and Engineering
First Advisor
Maninder Singh
Second Advisor
Jalal Khalil
Third Advisor
Stephanie Culver
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Keywords and Subject Headings
Machine Translation, Sentiment Preservation, OPUS-MT, Cross-Lingual Evaluation, Multilingual NLP
Abstract
This research investigates whether machine translation systems, particularly OPUS-MT by Helsinki-NLP, preserve sentiment across languages, with a focus on translations from Spanish to English. Although traditional metrics such as BLEU, F1, and COMET evaluate linguistic similarity, they often fail to detect sentiment shifts—an important factor in understanding the true intent of a text. This study applies sentence-level sentiment analysis using a Spanish-English parallel corpus and evaluates translation performance using both conventional model metrics and sentiment shift analysis by comparing sentiment labels before and after translation. The findings show that OPUS-MT preserves sentiment with over 95% agreement between source and translated texts, indicating that modern neural translation models reliably maintain emotional tone with only minor neutralization effects. Statistical and star-level analyses further justify the use of traditional evaluation metrics, as they align closely with sentiment preservation outcomes.
Recommended Citation
Xiong, Matthew, "Sentiment Preservation in Multilingual Machine Translation" (2025). Culminating Projects in Computer Science and Information Technology. 63.
https://repository.stcloudstate.edu/csit_etds/63
The code translates Spanish sentences into English using OPUS-MT, applies sentiment analysis to both versions, and measures how well the translation preserves the original sentiment.
