Home Linguistics & Semiotics Chapter 10. Domain-adapting and evaluating machine translation for institutional German in South Tyrol
Chapter
Licensed
Unlicensed Requires Authentication

Chapter 10. Domain-adapting and evaluating machine translation for institutional German in South Tyrol

  • Antonio Giovanni Contarino and Flavia De Camillis
View more publications by John Benjamins Publishing Company
Corpus Use in Cross-linguistic Research
This chapter is in the book Corpus Use in Cross-linguistic Research

Abstract

Building on a prior small-scale study on machine translation adaptation in the same language pair and domain (De Camillis 2021), this chapter reports on a (i) domain-adaptation, (ii) quality assessment, and (iii) automatic legal terminology evaluation experiment for legal South Tyrolean German. After collecting, sentence-aligning, and cleaning the LEXB parallel corpus, we used this bilingual resource to domain-adapt a ModernMT engine. Performance improvements were measured in terms of automatic quality metrics. The machine-translation of South Tyrolean legal terms was evaluated using an ad hoc automatic terminology evaluation tool. We observed a significant boost in performance and term accuracy in the output of the ModernMT adapted engine, but the improvement in legal terminology translation was deemed unsatisfactory.

Abstract

Building on a prior small-scale study on machine translation adaptation in the same language pair and domain (De Camillis 2021), this chapter reports on a (i) domain-adaptation, (ii) quality assessment, and (iii) automatic legal terminology evaluation experiment for legal South Tyrolean German. After collecting, sentence-aligning, and cleaning the LEXB parallel corpus, we used this bilingual resource to domain-adapt a ModernMT engine. Performance improvements were measured in terms of automatic quality metrics. The machine-translation of South Tyrolean legal terms was evaluated using an ad hoc automatic terminology evaluation tool. We observed a significant boost in performance and term accuracy in the output of the ModernMT adapted engine, but the improvement in legal terminology translation was deemed unsatisfactory.

Downloaded on 12.9.2025 from https://www.degruyterbrill.com/document/doi/10.1075/scl.113.10con/html
Scroll to top button