@inproceedings{sizov-etal-2024-analysing,
title = "Analysing Translation Artifacts: A Comparative Study of {LLM}s, {NMT}s, and Human Translations",
author = "Sizov, Fedor and
Espa{\~n}a-Bonet, Cristina and
Van Genabith, Josef and
Xie, Roy and
Dutta Chowdhury, Koel",
editor = "Haddow, Barry and
Kocmi, Tom and
Koehn, Philipp and
Monz, Christof",
booktitle = "Proceedings of the Ninth Conference on Machine Translation",
month = nov,
year = "2024",
address = "Miami, Florida, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.wmt-1.116",
pages = "1183--1199",
abstract = "Translated texts exhibit a range of characteristics that make them appear distinct from texts originally written in the same target language. With the rise of Large Language Models (LLMs), which are designed for a wide range of language generation and understanding tasks, there has been significant interest in their application to Machine Translation. While several studies have focused on improving translation quality through fine-tuning or few-shot prompting techniques, there has been limited exploration of how LLM-generated translations qualitatively differ from those produced by Neural Machine Translation (NMT) models, and human translations. Our study employs explainability methods such as Leave-One-Out (LOO) and Integrated Gradients (IG) to analyze the lexical features distinguishing human translations from those produced by LLMs and NMT systems. Specifically, we apply a two-stage approach: first, classifying texts based on their origin {--} whether they are original or translations {--} and second, extracting significant lexical features (highly attributed input words) using post-hoc interpretability methods. Our analysis shows that different methods of feature extraction vary in their effectiveness, with LOO being generally better at pinpointing critical input words and IG capturing a broader range of important words. Finally, our results show that while LLMs and NMT systems can produce translations of a good quality, they still differ from texts originally written by native speakers. Specifically, we find that while some LLMs often align closely with human translations, traditional NMT systems exhibit distinct characteristics, particularly in their use of certain linguistic features.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sizov-etal-2024-analysing">
<titleInfo>
<title>Analysing Translation Artifacts: A Comparative Study of LLMs, NMTs, and Human Translations</title>
</titleInfo>
<name type="personal">
<namePart type="given">Fedor</namePart>
<namePart type="family">Sizov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Cristina</namePart>
<namePart type="family">España-Bonet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Josef</namePart>
<namePart type="family">Van Genabith</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Roy</namePart>
<namePart type="family">Xie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Koel</namePart>
<namePart type="family">Dutta Chowdhury</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Ninth Conference on Machine Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Barry</namePart>
<namePart type="family">Haddow</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tom</namePart>
<namePart type="family">Kocmi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Philipp</namePart>
<namePart type="family">Koehn</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christof</namePart>
<namePart type="family">Monz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Miami, Florida, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Translated texts exhibit a range of characteristics that make them appear distinct from texts originally written in the same target language. With the rise of Large Language Models (LLMs), which are designed for a wide range of language generation and understanding tasks, there has been significant interest in their application to Machine Translation. While several studies have focused on improving translation quality through fine-tuning or few-shot prompting techniques, there has been limited exploration of how LLM-generated translations qualitatively differ from those produced by Neural Machine Translation (NMT) models, and human translations. Our study employs explainability methods such as Leave-One-Out (LOO) and Integrated Gradients (IG) to analyze the lexical features distinguishing human translations from those produced by LLMs and NMT systems. Specifically, we apply a two-stage approach: first, classifying texts based on their origin – whether they are original or translations – and second, extracting significant lexical features (highly attributed input words) using post-hoc interpretability methods. Our analysis shows that different methods of feature extraction vary in their effectiveness, with LOO being generally better at pinpointing critical input words and IG capturing a broader range of important words. Finally, our results show that while LLMs and NMT systems can produce translations of a good quality, they still differ from texts originally written by native speakers. Specifically, we find that while some LLMs often align closely with human translations, traditional NMT systems exhibit distinct characteristics, particularly in their use of certain linguistic features.</abstract>
<identifier type="citekey">sizov-etal-2024-analysing</identifier>
<location>
<url>https://aclanthology.org/2024.wmt-1.116</url>
</location>
<part>
<date>2024-11</date>
<extent unit="page">
<start>1183</start>
<end>1199</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Analysing Translation Artifacts: A Comparative Study of LLMs, NMTs, and Human Translations
%A Sizov, Fedor
%A España-Bonet, Cristina
%A Van Genabith, Josef
%A Xie, Roy
%A Dutta Chowdhury, Koel
%Y Haddow, Barry
%Y Kocmi, Tom
%Y Koehn, Philipp
%Y Monz, Christof
%S Proceedings of the Ninth Conference on Machine Translation
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, Florida, USA
%F sizov-etal-2024-analysing
%X Translated texts exhibit a range of characteristics that make them appear distinct from texts originally written in the same target language. With the rise of Large Language Models (LLMs), which are designed for a wide range of language generation and understanding tasks, there has been significant interest in their application to Machine Translation. While several studies have focused on improving translation quality through fine-tuning or few-shot prompting techniques, there has been limited exploration of how LLM-generated translations qualitatively differ from those produced by Neural Machine Translation (NMT) models, and human translations. Our study employs explainability methods such as Leave-One-Out (LOO) and Integrated Gradients (IG) to analyze the lexical features distinguishing human translations from those produced by LLMs and NMT systems. Specifically, we apply a two-stage approach: first, classifying texts based on their origin – whether they are original or translations – and second, extracting significant lexical features (highly attributed input words) using post-hoc interpretability methods. Our analysis shows that different methods of feature extraction vary in their effectiveness, with LOO being generally better at pinpointing critical input words and IG capturing a broader range of important words. Finally, our results show that while LLMs and NMT systems can produce translations of a good quality, they still differ from texts originally written by native speakers. Specifically, we find that while some LLMs often align closely with human translations, traditional NMT systems exhibit distinct characteristics, particularly in their use of certain linguistic features.
%U https://aclanthology.org/2024.wmt-1.116
%P 1183-1199
Markdown (Informal)
[Analysing Translation Artifacts: A Comparative Study of LLMs, NMTs, and Human Translations](https://aclanthology.org/2024.wmt-1.116) (Sizov et al., WMT 2024)
ACL
- Fedor Sizov, Cristina España-Bonet, Josef Van Genabith, Roy Xie, and Koel Dutta Chowdhury. 2024. Analysing Translation Artifacts: A Comparative Study of LLMs, NMTs, and Human Translations. In Proceedings of the Ninth Conference on Machine Translation, pages 1183–1199, Miami, Florida, USA. Association for Computational Linguistics.