Exciting up-and-coming technology for interpreters: InterpretBank

Exciting up-and-coming technology for interpreters: InterpretBank

InterpretBank is an exciting piece of technology for interpreters which is gaining traction and is catching the eye of more and more #Terps. However, given that the team behind it seems constantly at work (do they ever sleep?), always tweaking functionalities, adding features (and occasionally breaking previous ones), a complete review of functionalities is not what I will do here because next month new ones will have stolen the limelight.

In very general terms (and by no means exhaustively – you should download the trial version and have a play if you are really interested), InterpretBank is a pretty competent preparation tool, able to take in a range of file formats, display their content, try to extract terminology automatically (a nice idea, but without the solid natural language processing techniques which tools such as the SketchEngine use, the results are not particularly mind-blowing yet), allow you to select terms manually (very sensible, given the above-mentioned), and search a very useful range of online sources for translations of your newly-identified terms.

Another very neat feature is that you are able to view your source files in a web browser with all the terms that exist in your termbases highlighted and annotated with their respective translations (as in the screenshot below, where InterpretBank is displaying one of the preparation files complete with a highlight and Slovene annotation). Brilliant!

So far it’s all good (even excellent) for preparation with a few resources. However, the situation changes quite significantly if you get too excited and start importing large termbases. It is possible (eventually, after seeing InterpretBank crash under IATE’s TBX export, then converting from TBX to Excel through a CAT tools such as memoQ, then further pruning the Excel export until it only has two columns: source and target) to import 660,316 terms (which IATE en-fr has); it is also very gratifying to click through the 66,032 pages in InterpretBank to remind yourselves of what these terms are (no? nobody actually does that? oh!).

It’s a pretty bad idea to try and do much else with it, though, and especially to expect to do it fast (forget how quick you think your latest MacBookPro runs with its latest i9 processor, super-fast SSD and 16GB of DDR4, 8GB of which are dedicated to your Windows 10 virtual machine running on the latest Parallels). For the last 30 min I’ve been trying to display in the web browser the hefty (but plausible as a meeting preparation document) Bologna Process Implementation Report to be annotated with the pretty large IATE EN-FR database. So far, apart from a couple of straight crashes and a Task Manager report of InterpretBank using around 27% of the CPU, 67% of the memory and a LOT of power, I don’t have anything to show for it…

This downside is particularly relevant, unfortunately, when combined with using Dragon NaturallySpeaking Professional v15, my preferred automatic speech recognition tool for English so far. Now, InterpretBank are very sensibly trying to manage expectations by stating in a very long menu entry that their support of speech recognition technology is still very much experimental. However, for me it’s the loveliest-looking carrot which this tool has on offer: imagine being in the booth, opening your preferred glossary (however large), feeding the sound from the floor through Dragon NaturallySpeaking and seeing a reasonable transcription of the speaker’s intervention appear in InterpretBank, while terms present in the glossary are also displayed in real time together with their target language equivalents. I know what you will say: yet another thing to increase the interpreters’ cognitive load, but this time this may actually be quite helpful. If it works.

And there’s the big IF. To be able to capture the screenshot above, I had to use a much smaller glossary (around 100 terms) and dictate quite slowly into Dragon. Still, the MacBook’s processor and memory were going crazy, Dragon was taking longer than usual to work its magic, and the result isn’t of fantastic quality.

Yet it’s a start. Maybe I have too many optimistic bones in my body, but I have this feeling that we are slowly moving away from the time of individual applications being developed independently of how linguists intend to (and for very good reasons should) use them. CAT/TEnT tools developers are slowly acknowledging that some linguists work better when they can dictate, too, not just type, and also when they can listen back to the source and/or target content (yes, I know we’ve still got a long way ahead of us to persuade developers to embed speech synthesis at source and target segment level into their solutions, but Rome wasn’t built in a day, either, so let’s persevere!).

The fact that InterpretBank chooses to keep their experimental feature (instead of scrapping it until further notice, thus saving themselves the hassle of hearing that picky people like me are writing about yet another situation in which it has failed) shows to me that they do care about their community of users, trusting them to keep experimenting until they find for themselves those situations and those technical set-ups where everything works as expected. By the way, working on a Mac gives me the opportunity to keep testing InterpretBank with three speech recognition engines, in fact: Dragon, Windows and Apple’s, too, and the results are not always terrible, in fact! I look forward to the time when the software’s been optimised for dealing with large prep documents and big glossaries much faster and, in the meantime, although in my opinion it isn’t quite a tool that’s sufficiently grown-up to be relied on in the booth, it is still competent enough to enhance the interpreting training practice.