Unit:
Κατεύθυνση Γλωσσολογία: θεωρία και εφαρμογέςLibrary of the School of Philosophy
Supervisors info:
Μικρός Γιώργος, Καθηγητής, Τμήμα Ιταλικής Γλώσσας και Φιλολογίας, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Υφαντίδου Έλλη, Καθηγήτρια, Τμήμα Αγγλικής Γλώσσας και Φιλολογίας, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Νικηφορίδου Βασιλική, Καθηγήτρια, Τμήμα Αγγλικής Γλώσσας και Φιλολογίας, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Original Title:
Differences between Human and Machine-generated Institutional Translations: A comparative analysis using quantitative methods
Translated title:
Differences between Human and Machine-generated Institutional Translations: A comparative analysis using quantitative methods
Summary:
Machine translation, commonly referred to as MT, has gained popularity over the recent years; however, it has not yet reached the quality and naturalness of human writing. The present thesis aims to explore how human and automatic English translations of Greek institutional texts differ by comparing quantitative characteristics of the two translation types. Statistical analysis using independent samples t-tests revealed that the two corpora differed in a range of linguistic features including descriptive characteristics (e.g. word length), word information (e.g. parts of speech, word frequency), lexical diversity, syntax and cohesion; however, the degree of variation was not striking. In a follow-up examination, using Multilayer Perceptron neural network, the machine was able to classify correctly almost 82% of the texts as automatic or human-produced. These results suggest that the differences between HT and MT regarding the subgenre in question are detectable using machine learning techniques, but the distinction is not as clear-cut as expected. Further research is needed to determine whether the text properties that differ most in the two corpora can be used effectively as predictors of translation quality.
Main subject category:
Language – Literature
Keywords:
machine translation (MT), human translation (ΗΤ), translation quality assessment, institutional texts, Greek-English
Number of references:
110
