Since we had only just moved country before I attended BP19 in Bologna, I spent the few short days preceding my trip to Sheffield unpacking seemingly endless numbers of boxes. The timing was not great, but I had already booked conference tickets and hotels well in advance…
Five days after my return from Bologna, I travelled to the ITI Conference, held in the amazing Cutler’s Hall in Sheffield. This year’s topic: “Forging the Future of the Profession.”
It is hardly surprising that one of the recurring themes of the event turned out to be machine translation (MT), one of the industry’s prevalent trends. Experts from different areas of expertise approached the subject from different angles – from the technical and the linguistic angle as well as from a business point of view (meaning: will translators be replaced by machines soon?).
One of the early presentations on the first day of the conference tackled the question: does neural machine translation actually live up to the hype? Clever advertising and various press releases tend to suggest that MT engines are capable of producing translation quality that is more or less on par with that of a human. Reality, however, looks somewhat different: depending on subject of the text, length of the sentence and other factors, quality varies widely (and the texts don’t always work as texts since the machine only concentrates on single sentences at a time).
Another question brought up was whether interpreters would be able to survive in am MT world? The clear-cut answer? Yes! Voice translation tends to be especially difficult since speech recognition technologies add another layer of complexity to the task.
But what is the translator’s influence on machine translation engines? A more technical talk delved into the differences between statistical (“old” tech) and neural (all the rage) machine translation and enlightened us to a few interesting facts. Research shows that, although neural MT produces results that sound a lot better, the productivity gain compared to statistical MT is not significant.
One might also wonder: how will the use of MT affect our language and its development? When machine-translated texts are edited by translators, changes are kept to a minimum to reduce effort. These changes are then fed back into the system to train it for the next text – a process in which creativity and varied input are somewhat stifled.
MT was of course not the only topic of the conference. Other talks did, for example, address stress management for interpreters and useful tips regarding presentation skills.
After two days, packed with talks, social events and lots of networking, it was time to pack my bags again and travel back to Germany (where a few packed boxes were still waiting for me). Time to digest the varied input, develop new ideas and follow up with new connections!