Felanalys av maskinöversättningar av svenska texter till kroatiska

Ljubas, Sandra. (2017). Felanalys av maskinöversättningar av svenska texter till kroatiska. Diploma Thesis. Filozofski fakultet u Zagrebu, Department of English Language and Literature > Chair of Scandinavian Languages and Literature. [mentor Antunović, Goranka and Simeon, Ivana].

Preview

PDF (Croatian) Download (2MB) | Preview

Abstract

The objective of the present study is to examine the errors found in the machine translations of various texts from Swedish into Croatian with the help of Google Translate, a free and publicly accessible web-based MT system. The material used in the analysis includes three corpora. The first corpus comprises ten source texts (ca. 80 standard pages): a technical, a political, four administrative and four scientific texts. The second corpus consists of human translations I have used as reference translations. Both source texts and their human translations were included in the master’s theses written by eight students at the Scandinavian Section of the Faculty of Humanities and Social Sciences in Zagreb. The third corpus contains machine translations of the same source texts. The study aims to provide preliminary insights into machine translation regarding a language pair that has not yet been investigated in the MT context as well as to test the hypotheses with respect to how GT operates, i.e. to examine whether GT retains the same number of sentences as in the source text, whether it produces spelling and writing errors by itself, whether there are any English words in translations and whether GT makes word order and comma errors when translating from Swedish into Croatian. Taking Simeon’s classification (2008) as a basis, I have divided the errors into eight categories: untranslated words, omitted and added words, lexical, stylistic, orthographic and morphosyntactic errors, word-order errors and two additional categories: consistency errors and unclear sentences. The results have shown that GT makes a substantial amount of morphosyntactic errors when translating into Croatian. However, lexical errors have the greatest impact on the level of (un)intelligibility. Furthermore, it can be said that GT fails to correctly render the Swedish formal subject “det”, reflexive pronouns, phrasal verbs, double negation, abbreviations, compounds etc. The lack of digital content in Croatian, the lack of people who are willing to use modern technology and have the necessary time and resources to develop it, the lack of trust concerning privacy and intellectual property rights, as well as the lack of financial means all lead to the conclusion that machine translation is not a threat to professional translators working with Swedish in Croatia. However, a potential threat can be seen in that a language may disappear without sufficiently developed language technologies that could be used in the digital world where so much takes place today.

Actions (login required)

View Item