mercredi 20 avril 2011

Thoughts on the future of translation technology

A recent panel discussion took place at MemoQFest on the subject "Has Translation Technology a Future".
What struck me about the panel's remarks was that:
1. they were primarily about tools; even the discussions of standards such as TMX and XLIFF were tool-oriented;
2. no one dealt with the importance of differing statistical reporting from tool to tool for translation data provided in the same standard format.
Here are my thoughts on these two issues:
1. The tool-oriented discussion. Even though the title of the discussion leads us first of all to tools, it is, I think, an error to consider tools outside of the Methods-Tools-Procedures relationship. The future of the technology we use depends on how we use the tools according to methods and procedures, not just on the tools themselves and the functionality they provide. And as far as standards are concerned, what about the method and procedure-oriented ones that exist for our industry such as EN 15038? How will the tools evolve along with these in the future? In particular, how are we going to use the tools in our costing methods and procedures. This brings us to my next point.
2. Statistics. The commercial model of our industry has been entirely based on word counts - and specifically, segment matches percentages -  for 20 years now. It is a paradigm. The obvious fact is that a statistical algorithm is necessary to produce these stats from aligned source and target language files, and discounts for fuzzy and context matches and such are only possible if the persistent data from the iterations of whatever algorithm is being used are stored in a format that allows us to use them again. So, we not only need either open or standard exchange formats, we need to know what criteria the various algorithms use (because they don't produce comparable results).
However, the real problem with rates that are based purely on segment matching and word count statistics is that they don't necessarily represent the cost of a translation. Statistical data are only one aspect of actual translation cost. This cost is better represented by the time taken for each stage in the translation process. This is because other factors can double or triple the cost. These factors include: subject field, quality of writing (is it, for example, clear?), the availability of relevant terminological data, document sensitivity (is the doc. internal or external, marketing or reporting) etc. Translation cannot be reduced to Adam Smith's pin factory. Nor can it be reduced to human resource management as last year's key note speaker seemed to think. It is more like agribusiness, where you have everything from factory farming to organic farming, from fast food to slow food, from cafeterias to gourmet restaurants, from hypermarkets (...) to small specialty shops.
I think that the current paradigm is running out of steam. More and more translators and LSPs are realizing that it is an abstraction that is no longer entirely satisfactory.
Finally given that Daniel Brockmann, one of the driving forces behind this paradigm (which has allowed our industry to grow rapidly ...), attended MemoQFest, wouldn't his participation in this discussion have made it more representative? Perhaps he didn't want to. If so, too bad. I for one am looking forward to a discussion in the future where all the actual issues are dealt with.