SubmissionNumber#=%=#10
FinalPaperTitle#=%=#Are Large Language Models State-of-the-art Quality Estimators for Machine Translation of User-generated Content?
ShortPaperTitle#=%=#
NumberOfPages#=%=#11
CopyrightSigned#=%=#Shenbin Qian
JobTitle#==#
Organization#==#
Abstract#==#This paper investigates whether large language models (LLMs) are state-of-the-art quality estimators for machine translation of user-generated content (UGC) that contains emotional expressions, without the use of reference translations. To achieve this, we employ an existing emotion-related dataset with human-annotated errors and calculate quality evaluation scores based on the Multi-dimensional Quality Metrics. We compare the accuracy of several LLMs with that of our fine-tuned baseline models, under in-context learning and parameter-efficient fine-tuning (PEFT) scenarios. We find that PEFT of LLMs leads to better performance in score prediction with human interpretable explanations than fine-tuned models. However, a manual analysis of LLM outputs reveals that they still have problems such as refusal to reply to a prompt and unstable output while evaluating machine translation of UGC.
Author{1}{Firstname}#=%=#Shenbin
Author{1}{Lastname}#=%=#Qian
Author{1}{Username}#=%=#s.qian
Author{1}{Email}#=%=#s.qian@surrey.ac.uk
Author{1}{Affiliation}#=%=#University of Surrey
Author{2}{Firstname}#=%=#Constantin
Author{2}{Lastname}#=%=#Orasan
Author{2}{Username}#=%=#c.orasan
Author{2}{Email}#=%=#c.orasan@surrey.ac.uk
Author{2}{Affiliation}#=%=#University of Surrey
Author{3}{Firstname}#=%=#Diptesh
Author{3}{Lastname}#=%=#Kanojia
Author{3}{Username}#=%=#dipteshkanojia
Author{3}{Email}#=%=#dipteshkanojia@gmail.com
Author{3}{Affiliation}#=%=#University of Surrey
Author{4}{Firstname}#=%=#Félix
Author{4}{Lastname}#=%=#do Carmo
Author{4}{Username}#=%=#felixdocarmo
Author{4}{Email}#=%=#f.docarmo@surrey.ac.uk
Author{4}{Affiliation}#=%=#University of Surrey

==========
èéáğö