Hemingway as a hardness sample for AI translation-Diepresse.com
Can you have literature translated by artificial intelligence? Waltraud Kolb conducts studies on machine translation of fiction and human revise.
How do you deal with a « Chimney Swift » when translating a Hemingway text-a bird that is at home on the American continent? « Schornstein sailor » is the name of the solution that delivers the machine translation. It can be considered correct, but is based on purely zoological criteria. The fact that the conscious bird in the short story « A very short story » is observed in the sky above the Italian Padua is not taken into account.
So would a similar European bird name be more suitable in the German translation? Is Ernest Hemingway tells more from an American or European perspective? Literature translators break their heads through such questions. The neural networks of Deepl, However, those who are trained with available translations have no sense for literary backgrounds. So -called post editors, who are commissioned to post -processing machine translations, also tend to follow their solutions. Priming is called this influence of the result presented by the machine on humans.
Wall sailor and Schwalbe
Waltraud Kolb – she is an assistant professor for literary translation on Center for Translation Science the University of Vienna – showed this effect in a study. Using several places in Hemingway short history, they compared five posted translations with five human new translations. The verbal statements, corrections and keyboard input of all translators during working on the computer were recorded with technical aids.
« /> « Hemingway is particularly suitable because his language perceived as relatively easy, » says Waltraud Kolb. Christian Sperr
The bird name mentioned was translated in all five post editions with the version of the machine translation. In the five pure human translations, on the other hand, there are four different bird names, including the European mouse sailor and the American-European smoke swallow. Apart from the greater variety of variants, it was also evident that the different images that cause such names in the minds hardly played a role for posteditors. The bird name is just one of many examples based on which Kolb is using the Problem of machine translation of literature deals with. « Hemingway is particularly suitable because his language is perceived as relatively simple and you could therefore assume that it is easier to translate from machines, » says the scientist.
Older texts overwhelm
She is currently working on a study in which the translation of Short stories by Kate Chopin, an American author from the late 19th century. She examines how much that of Deep and Chatt distinguish generated translations. It tends to cut a little better in comparison, with other texts it could be the other way around, says Kolb.
However, the researcher emphasizes unchanged: « None of the two systems produces a text that does not have to be revised, and some also indicates that an older text is more difficult for both systems than a more modern. »
Lexicon
Posted means subsequent editing. With regard to texts that have been translated machine, this means the process of post -processing by human translators.
Prime (Pre -embossing) is an effect that arises when posted with machine -created texts. The human translators are influenced by the machine translation when reworking, from the choice of words to the tone of a narrative.
