Wednesday, December 26, 2018

Rage against the machine translation: why so much fuss?


Large corporations such as Google or Amazon invest millions of dollars a year in the development of machine translation (Mt) technology. This could expand translation capacity and potential cost savings. However, although excitement about the productive aspect the machine translation can be measured, there is a much more important investment which is ignoring in the process. I speak of the emotional investment of users of machine translation in a broad sense. I mean all the people who have been affected by the machine translation, because there is a very important psychological aspect for all involved, ranging from customers who buy services of post-edit the linguists that are involved in the process. Do you think that it is all? Well, there are many more emotionally affected people: end-users and their sense of the usability of the product resulting from the machine translation, translation agencies that hire posteriors, academic institutions they work to develop post-editors and research of automatic translation, developers of motors in machine translation, etc. are many!
. . The main problem is that most of the linguistic human resources processing machine translation do not favorably, not only as a potential threat to their profession (something that very few actually supported), but also as a tool detestable that mainly it hinders and stands in his way. Rarely see it as a useful tool for achieving better results.
While these thoughts can be based on reasons legitimate (due to defective or traumatic first pilot experiments with the automatic translation), I can't help but feel disturbed by the intensity of some of his emotions on the subject, a certain bias generated by an acute closed vision, a strong barrier that hinders debate over the topic and its possible advantages.
Several studies on the subject have shown that individuals who interact in a way regular with this technology accept the results much more when the engines include some degree of network neural or learning deep (in comparison with the) system based on phrases). Is it happens? Simply because they are more likely to prefer a better flow (Syntax) than accuracy.
That is why I mention the emotional investment in the machine translation as a key element to reinvent the concept for users. Understand the latest changes that have been implemented in the process can help users of machine translation to overcome their fears.
It seems that the classical and more standardized machine translation (based solely on comparison statistics instead of the artificial intelligence) is much more appreciated by frequent users, taking into account that the latter is more efficient and easier to "fix" the post-edited; while pre-translated with classical technology text is evidence that post-edited is more problematic and erratic. This has probably fueled anger to automatic translation and its bad reputation. The majority of the users (if not all) accept more material pre-translated with statistical processing system with rules-based system. It seems that neuronal machine translation could be the best tool to move towards acceptance by frequent users.
If you have read some of my previous posts, you know that I tend to be very apocalyptic when it comes to flirting with the idea that machines gradually going to conquer the world, but it seems reasonable that the cooperation between the men and the machines could lead us a step in the evolutionary ladder that leads to the lighting either final self-destruction. Either way, looks like an exciting adventure that no one wants to lose. For better or for worse, all want secret that dizzying sense of loss of control, at least for a moment. Everything is part of this emotional rollercoaster ridiculous full of twists and turns that entails the acceptance of change and how to address it.