The story of how one translator taught himself what he needed to know to be able to take advantage of an opportunity.
Let me tell you the story of Terence Lewis.
Terence is a Dutch>English translator. In the 1990s, one of his clients, Siemens Nederland, required a machine translation (MT) solution for Dutch>English to enhance communication between Dutch and English speakers within Siemens. So, Terence developed a rules-based MT system based completely on Word macros. He released it in 1994, and Siemens used the system until it needed a standalone application that could be used anywhere.
So, Terence taught himself the programming languages Java and C++ and developed that as well. And again it was used successfully by Siemens and a number of other large multinationals for internal purposes.
Then statistical MT became a more feasible option, especially with the availability of the open-source Moses engine in the 2010s. Terence taught himself how to develop and use that code, took what he had already developed for his rules-based engine, and created a “hybrid engine,” where the statistical part of the program suggests lots of phrases and the rules-based part picks some and attempts to combine them correctly.
As with any MT system, it creates “RobotSpeak” (original quote by Terence), but of a kind that Siemens likes well enough for its purposes. According to Terence, it’s about as good as Google Translate, and even better when it’s applied on Siemens stuff.
Just recently, Terence launched MyDutchPal.com to the general public with the TrasyCloudBox, a Java-based tool that allows you to copy Dutch text into one half of the two-paned program and get the English MT that was generated on Terence’s server on the other half. It may be a little onbeholpen (read: clunky) and not really well integrated into the way translators work, so Terence wasn’t particularly dismayed when this didn’t become a huge bestseller.
Terence will soon begin offering the pre-translation of TMX and XLIFF files with his MT engine that can then be imported into translation environment tools. My sense (and Terence agreed) is that this also won’t be quite what the Dutch>English translation community is looking for. So, the next integration is already in the pipeline: the integration into memoQ as a plugin and, further down the road, into Trados Studio. Terence will do this in partnership with Jon Olds, who is also the developer of memoQ’s newly released Moses plugin.
As I chatted with Terence recently, he envisioned all kinds of possibilities to make this into a much more interesting plugin than just one that suggests MT segments. How about getting to all those many different phrases and presenting those to the translator through predictive typing? How about continuous training of the data? And so on and so forth.
I love this story. Not because I’m personally really that interested in Dutch>English MT, but because of how Terence responded to the possibility that opened up and took hold of that opportunity. He was a translator—not first and foremost, perhaps, but that’s what he was. Seeing a need, however, he taught himself what he needed to know to be able to provide the services that presented themselves to him, and throughout all of that he stayed true to his translator ideals (see “RobotSpeak” above).
Jost Zetzsche is the co-author of Found in Translation: How Language Shapes Our Lives and Transforms the World, a robust source for replenishing your arsenal of information about how human translation and machine translation each play an important part in the broader world of translation. Contact: firstname.lastname@example.org.