고유명사, 숫자의 기계번역: 원리와 오류분석
Machine translation for named entities: error analysis based on its mechanism
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study examines the characteristics and quality of machine translation (MT) for named entities (NEs). We first provided an overview of the underlying mechanisms and technical obstacles involved in NE translation, based on previous studies in this field to understand the challenges that MTs may encounter when translating NEs. Based on these theoretical considerations, we proposed three hypotheses regarding potential NE-related translation problems and error types: (1) MTs may struggle to translate NEs consistently; (2) MTs often fail to differentiate proper nouns from common nouns; and (3) without sufficient contextual support, low-frequency NEs may be translated incorrectly, resulting in non-words or semantically unrelated terms. We performed a statistical analysis of 120 translations produced by two standard MT systems (Google Translate and Papago) and two generative artificial intelligence models (ChatGPT 3.5 and ClovaX) to test these hypotheses. The results revealed that (1) consistency errors occurred in 4.21%–26% of all NE translations, (2) 19.55%–50.34% of proper nouns that could be confused with common nouns were incorrectly translated as such, and (3) 31.31%–35.72% of low-frequency NEs lacking sufficient context were mistranslated.
Abstract
This study examines the characteristics and quality of machine translation (MT) for named entities (NEs). We first provided an overview of the underlying mechanisms and technical obstacles involved in NE translation, based on previous studies in this field to understand the challenges that MTs may encounter when translating NEs. Based on these theoretical considerations, we proposed three hypotheses regarding potential NE-related translation problems and error types: (1) MTs may struggle to translate NEs consistently; (2) MTs often fail to differentiate proper nouns from common nouns; and (3) without sufficient contextual support, low-frequency NEs may be translated incorrectly, resulting in non-words or semantically unrelated terms. We performed a statistical analysis of 120 translations produced by two standard MT systems (Google Translate and Papago) and two generative artificial intelligence models (ChatGPT 3.5 and ClovaX) to test these hypotheses. The results revealed that (1) consistency errors occurred in 4.21%–26% of all NE translations, (2) 19.55%–50.34% of proper nouns that could be confused with common nouns were incorrectly translated as such, and (3) 31.31%–35.72% of low-frequency NEs lacking sufficient context were mistranslated.
참고문헌 (19)
- 성승은, "그림책의 고유명사 번역에 관한 고찰", 국제언어문학, 24, pp.143~169 (2011)
- "딥 러닝을 이용한 자연어 처리 입문(e-book)" (2024), 유원준[URL]
- "Language models are few-shot learners" (2020)[URL]
- Chang, T. A., "Language Model Behavior : A Comprehensive Survey", Computational Linguistics (2023)[DOI]
- Chen, J., "Learning in-context learning for named entity recognition", Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, 1 (2023)
- Chung, H. -Y., "Wie übersetzt NMT Eigennamen und Zahlen?", Lebende Sprachen, 63(1), pp.142~167 (2018)
- Koehn, P., "Six challenges for neural machine translation", Proceedings of the First Workshop on Neural Machine Translation, pp.28~39 (2017)
- Nadeau, D., "A Survey of named entity recognition and classification", Lingvisticæ Investigationes, 30(1), pp.3~26 (2007)
- "Neural Machine Translation and Sequence-to-sequence Models: A Tutorial" (2017)[URL]
- Riaz, H., "ELLEN : Extremely Lightly Supervised Learning For Efficient Named Entity Recognition", Lightly Supervised Learning For Efficient Named Entity Recognition.’, Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pp.5622~5636 (2024)
- Rogers, A., "A Primer in BERTology : What We Know About How BERT Works", Transactions of the Association for Computational Linguistics, 8, pp.842~866 (2020)
- Schuster, S., "When a sentence does not introduce a discourse entity, transformer-based models still sometimes refer to it", Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Association for Computational Linguistics, pp.969~982 (2022)
- Vermes, A., "Proper names in translation : An explanatory attempt", Across Languages and Cultures, 4(1), pp.89~108 (2003)[DOI]
- Vinay, J. -P., "Comparative Stylistics of French and English" (1995), Benjamins
- "TENER : Adapting Transformer Encoder for Named Entity Recognition" (2019)[URL]
- "Named Entity Recognition from Synthesis Procedural Text in Materials Science Domain with Attention-Based Approach" (2021)[URL]
- "Named Entity Recognition (NER): Unveiling the value in unstructured text"[URL]
- "BERT Visualization in Embedding"[URL]