Search

V magistrski nalogi preverjam razumljivost nerevidiranih strojno prevedenih spletnih besedil pri končnih uporabnikih. Raziskavo sem izvedel z anketo, ki je vsebovala primere strojnih prevodov splošnih besedil, ki sem jih prevedel s prevajalnikoma Google Translate in eTranslation. Primeri vključevali napake štirih vrst, ki so bile predstavljene v kontekstu. Ta je lahko bil izključno besedilni, kombinacija besedilnega in vizualnega dveh vrst – s slikovnim gradivom, ki vpliva na razumevanje ali ne – ali vezan na pravilen izbor slike, na katero se besedilo nanaša. Vzorec 120 anketirancev je pokazal približno 59 % stopnjo razumevanja primerov, rezultati pa so bili boljši v kategorijah, kjer je bilo razumevanje vezano na slikovno gradivo oz. na izbor pravilne slike. This thesis tackles the issue of end-user comprehensibility of unedited machine translated web texts. The research was carried out by using a questionnaire, which contained examples of general texts, translated with Google Translate and eTranslation. The examples included four different types of errors, which were presented in context. The latter was either purely textual, a combination of textual and visual of two types – with pictures that affected comprehension or did not – or linked to the correct selection of a picture the text referred to. A sample of 120 respondents showed a comprehensibility rate of roughly 59 %, while the results were better in the categories where comprehensibility was tied to the visual material or the correct selection of an image.

This paper describes the TALP–UPC system in the Spanish–English WMT 2016 biomedical shared task. Our system is a standard phrase-based system enhanced with vocabulary expansion using bilingual word embeddings and a characterbased neural language model with rescoring. The former focuses on resolving outof- vocabulary words, while the latter enhances the fluency of the system. The two modules progressively improve the final translation as measured by a combination of several lexical metrics.