Machine Translation: Top or Flop? The Results of an Itl Survey
By: Carina Mayr - itl Institut für technische Literatur AG
17 January 2020
What do companies’ employees think about machine translation (MT)? What do they think are its main benefits? And what are its risks? The people responsible for MT at itl wanted to get an idea about all that, and so they asked those coming to itl’s evening event “Machine translation - what lies behind it and how you can benefit from it”. The results were clear: although they recognised the benefits of machine translation, particularly in terms of the time it saves, at the same time they feared there is a loss of quality.
Machine translation is what everyone in the industry seems to be talking about these days. At this year’s itl evening event, the participants heard all the latest information on machine translation. Carina Mayr, Director of the Translation Department at itl AG, offered an open-minded view of MT and addressed the question of whether MT is now really as good as or better than human translation. What can it do, and what can’t it do? And, above all, what does it do for technical documentation?
In the run-up to the event, the participants answered six questions online, the results of which were discussed during the evening event.
The first question asked the respondents whether they already had any experience of machine translation. A large majority stated that they had already had experience of machine translation, above all at work. Only around a quarter said they had no experience at all of MT.
The next question asked what they saw as the main benefits of machine translation in their work. The most common answer was “faster translations”, followed by “cheaper translations” and “translation of less important languages or document types”. The respondents thus expected MT to enable them to get translations done more quickly, more cheaply and more often. Only a few of the respondents believed that machine translation will offer higher-quality translations.
In response to the question as to what risks the participants believed were associated with MT in their working environment, the most common answer was “Faulty or poor translations”. As with the previous question, about the benefits of MT, not only did few expect an improvement in quality, but they also expected poor-quality translations to a much greater extent. People have greater confidence in human translators (as things stand, at least) than in a machine translation algorithm. Further risks mentioned were data security, dependency on machines or their vendors and job losses as a result of streamlining.
The respondents were then asked what they expected from MT in terms of quality. Around 40% of them said they expected the same quality as from a human translator. Another 40% agreed with the statement “The main thing is to understand what is meant”, thus revealing that they tend to expect low-quality translations from MT. Only a small minority of the respondents expected better quality.
The respondents were clearly optimistic about the future of machine translation. More than three out of four of them agreed with the answer “Will establish itself in certain areas”. Only 10% of the respondents believed that MT would largely replace human translators, but none of them were of the opinion that MT is a passing phase driven purely by hype.
Finally, it emerged that the respondents’ opinions about artificial intelligence and machine translation were by no means uniform. When asked to give a general rating to artificial intelligence and machine translation, the average rating they gave was 3.19 out of 5 possible points/stars.
The results of our small survey reflect the differing opinions on the issue in the industry. The results of neural machine translation (NMT) are too promising to be disregarded, but there are justifiable doubts in the very strictly regulated world of technical documentation as to how the creative and often unpredictable algorithms of the MT engines will be able to comply with all the guidelines and directives. However, there may have to be a rethink here, as well.
Conclusion of the survey
It's still not clear what the future of machine translation will be. The people we asked see both risks and opportunities with MT. The extent to which machine translation prevails against human translators will depend on both price and the quality of the translation algorithms used. As things stand, machine translations are often post-edited by professional translators to prevent any loss of quality. That enables companies to combine the best of both worlds: fast, finely tuned machines, on the one hand, and thoughtful translators, on the other, who can identify inconsistencies, iron out problems and make company-specific adjustments.