Humans vs. Artificial Intelligence: the human translator wins the first battle !

Humans vs. Artificial Intelligence: the human translator wins the first battle !

Sejong University in South Korea and the International Interpreting and Translation Association recently organized a translation competition in which professional human translators and three well-known automated translation programs were pitted against each other: Systran, Google Translate and Papago.

The human translators had 50 minutes to translate four selected texts (a literary translation and then a more general translation) from English into Korean and vice versa. Internet access was permitted. As an aside, it should be noted that none of the texts provided to translators had ever been translated. The three programs were fed exactly the same texts and performed these translations five times faster.

The quality of the translations was then evaluated according to three criteria (scored out of a maximum of 5 points each): precision, expression and logic. Out of a perfect score of 60 for the four texts, the human translators scored almost 50 points while the best of the three artificial intelligence programs only scored 28 points! However the individual score of each automated translation program was not disclosed, so we do not know which of the three performed the best or the least poorly …

According to the organizers of this competition, it has been reported that a recurring problem of artificial intelligence programs is that they use the NMT approach for translating (Neural Machine Translation, which differs from the automatic statistical approach). The latter uses a network of artificial neurons and deep learning techniques, which are not capable of understanding context or the text on a deeper level …

Unfortunately, in texts translated using this approach, there are too many grammatical mistakes. A Systran representative, one of the drivers of automated translation, also acknowledges the difficulties in translating nuance and emotions, the achilles heel of artificial intelligence programs, still to geared towards producing quantity rather than quality.

It appears that human translators, the real ones, still have many years ahead of them even if artificial intelligence programs are slowing bridging  the gap, as a result of new inventive and innovative software updates. However this is where the greatest of paradoxes lies, since these artificial intelligence programs are programmed and configured by humans … The consumer will have the choice between opting for the work of a human which has been enriched through thinking, experience and a sensitivity or for an automated translation, which is cheaper or even free with some software, but where the quality is sorely lacking.

“The ideal reader is a translator. He is able to dissect a text, to peel away the skin, strip it to the bone, follow the path of each artery and vein, and then create a new living being. ” Alberto Manguel

The human translators had 50 minutes to translate four selected texts (a literary translation and then a more general translation) from English into Korean and vice versa. Internet access was permitted. As an aside, it should be noted that none of the texts provided to translators had ever been translated. The three programs were fed exactly the same texts and performed these translations five times faster.

The quality of the translations was then evaluated according to three criteria (scored out of a maximum of 5 points each): precision, expression and logic. Out of a perfect score of 60 for the four texts, the human translators scored almost 50 points while the best of the three artificial intelligence programs only scored 28 points! However the individual score of each automated translation program was not disclosed, so we do not know which of the three performed the best or the least poorly …

According to the organizers of this competition, it has been reported that a recurring problem of artificial intelligence programs is that they use the NMT approach for translating (Neural Machine Translation, which differs from the automatic statistical approach). The latter uses a network of artificial neurons and deep learning techniques, which are not capable of understanding context or the text on a deeper level …

Unfortunately, in texts translated using this approach, there are too many grammatical mistakes. A Systran representative, one of the drivers of automated translation, also acknowledges the difficulties in translating nuance and emotions, the achilles heel of artificial intelligence programs, still to geared towards producing quantity rather than quality.

It appears that human translators, the real ones, still have many years ahead of them even if artificial intelligence programs are slowing bridging  the gap, as a result of new inventive and innovative software updates. However this is where the greatest of paradoxes lies, since these artificial intelligence programs are programmed and configured by humans … The consumer will have the choice between opting for the work of a human which has been enriched through thinking, experience and a sensitivity or for an automated translation, which is cheaper or even free with some software, but where the quality is sorely lacking.

“The ideal reader is a translator. He is able to dissect a text, to peel away the skin, strip it to the bone, follow the path of each artery and vein, and then create a new living being. ” Alberto Manguel

Medical and Pharmaceutical

Medical translation: a challenge of growing importance

6 June 2018
Over the last few decades, medicine has seen more revolutionary changes and advances than any other scientific field. On the one hand, research has been progressing in leaps and bounds in a wide range of fields such as cardiology, genetics, our understanding of the aging process and medicinal treatments. This is a result of excellent training programs and the work of international research teams.
Can and should we translate poetry?

Can and should we translate poetry?

6 June 2018
Because it often plays with the form, wording and musicality of language, more than any other literary genre, poetry does not easily lend itself to translation. However, people have nonetheless been translating it for centuries. How then, is it possible to translate what some deem to be untranslatable?