The Impact of Text Type on the Quality of Machine Translation from Croatian to English and French

master's thesis

• Title (croatian): Utjecaj tipa teksta na kvalitetu strojnog prijevoda s hrvatskog na engleski i francuski
• Title (french): L'impact du type de texte sur la qualité de la traduction automatique du croate vers l’anglais et le français
• Author: Iva Belina
• Mentor: Nataša Pavlović (mentor)
• Mentor: Marta Petrak (mentor)
• Committee member: Nataša Pavlović (član povjerenstva)
• Committee member: Marta Petrak (član povjerenstva)
• Committee member: Kristijan Nikolić (predsjednik povjerenstva)
• Granter: University of Zagreb; Faculty of Humanities and Social Sciences; (Department of English language and literature); Zagreb
• Granter: University of Zagreb; Faculty of Humanities and Social Sciences; (Department of Romance languages and literature); Zagreb
• Defense date and country: 2023, Croatia
• Scientific / art field, discipline and subdiscipline: HUMANISTIC SCIENCES; Philology; Anglistics
• Scientific / art field, discipline and subdiscipline: HUMANISTIC SCIENCES; Philology; Romance Studies

Abstract

The thesis explores the translation quality of five different text types (journalistic, administrative, literary, scientific, controlled) when translated by Google Translate from Croatian into English and French in order to help facilitate and speed up the translation process. By comparing the translation outputs, the study aims to identify the text type with the highest and lowest translation quality, as well as to compare the results of Croatian-English and Croatian-French translations. The MT quality assessment encompassed several automated metrics, namely BLEU, TER and METEOR, and human evaluation methods, specifically a questionnaire and error counting. Once normalized, the results revealed that administrative texts had the lowest MT quality for both language pairs, while the highest scored text types were scientific texts for Croatian-English translation, and controlled texts for Croatian-French translation. Some similarity has been shown in translation quality of the same text types in both target languages. The two lowest scored text types, administrative and journalistic type, produced similar relative scores, but most text types varied in their scoring, showing that Croatian-English and Croatian-French translations of the same texts are not of similar quality.

Abstract (croatian)

U ovom se diplomskom radu istražuje kvaliteta prijevoda pet različitih vrsta tekstova (novinarsko-publicistički, administrativno-poslovni, književnoumjetnički, znanstveni, kontrolirani) prevedenih pomoću Google prevoditelja s hrvatskog na engleski i francuski radi ubrzanja i olakšanja procesa prevođenja. Cilj je istraživanja analizom rezultata procjene strojnog prijevoda odrediti koja vrsta tekstova ima najvišu odnosno najnižu kvalitetu prijevoda, kao i usporediti rezultate hrvatsko-engleskih i hrvatsko-francuskih prijevoda. Procjena kvalitete strojnog prijevoda obuhvatila je nekoliko metrika za automatsku procjenu kvalitete (BLEU, TER i METEOR), te dvije metode ljudske procjene (upitnik i brojanje pogrešaka). Nakon normalizacije rezultata pokazalo se da su administrativno-pravni tekstovi imali najnižu kvalitetu strojnog prijevoda za oba jezična para, dok su najbolje ocijenjene vrste tekstova bile znanstveni tekstovi za hrvatsko-engleski prijevod te kontrolirani tekstovi za hrvatsko-francuski prijevod. Pokazalo se da postoji određena sličnost u kvaliteti prijevoda istih vrsta tekstova s hrvatskog na francuski i engleski. Naime, dvije najslabije ocijenjene vrste tekstova, administrativno-poslovna i novinarsko-publicistička, imale su slične ocjene u engleskim i francuskim prijevodima, no većina promatranih vrsta tekstova bila je različito ocijenjena, što pokazuje da hrvatsko-engleski i hrvatsko-francuski prijevodi istih tekstova nisu jednake kvalitete.

Abstract (french)

La présente thèse se penche sur la qualité de traduction de cinq types de textes (journalistique, administratif, littéraire, scientifique, contrôlé) lorsqu'ils sont traduits par Google Traduction du croate vers l'anglais et le français, afin de faciliter et d'accélérer le processus de traduction. En comparant les résultats des traductions, cette étude vise à identifier le type de texte affichant la meilleure et la pire qualité de traduction, ainsi qu'à comparer les résultats des traductions croate-anglais et croate-français. L'évaluation de la qualité de la traduction automatique a englobé plusieurs algorithmes, à savoir BLEU, METEOR et TER, ainsi que des méthodes d'évaluation humaine, telles qu’un questionnaire et le décompte des erreurs. Une fois normalisés, les résultats ont révélé que les textes administratifs présentaient la qualité de traduction automatique la plus faible pour les deux paires de langues, tandis que les types de textes les mieux notés étaient les textes scientifiques pour la traduction croate-anglais, et les textes contrôlés pour la traduction croate-français. Une certaine similitude a été observée dans la qualité de traduction des mêmes types de texte dans les deux langues cibles. Les deux types de textes les moins bien notés, les textes administratifs et journalistiques, ont obtenu des scores relatifs similaires, mais la plupart des types de textes ont obtenu des scores différents, ce qui montre que les traductions croate-anglais et croate-français des mêmes textes ne sont pas de qualité similaire.

• Keywords: machine translation; Google Translate; machine translation quality assessment; text types
• Keywords (croatian): strojno prevođenje; Google Translate; procjena kvalitete strojnog prijevoda; vrste teksta
• Language: english
• URN:NBN: urn:nbn:hr:131:099985
• Study programme: Title: English Language and Literature (double major); specializations in: Linguistics, Literature and Culture, Translation; Course: Translation; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra engleskog jezika i književnosti (magistar/magistra engleskog jezika i književnosti)
• Study programme: Title: French Language and Literature (double major); specializations in: Teaching, Translation, Research; Course: Translation; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra francuskog jezika i književnosti (magistar/magistra francuskog jezika i književnosti)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2023-09-18 13:47:01