DG SCIC, the University of Ghent and the Johannes Gutenberg University Mainz started a research project to assess the potential benefits and usefulness of Automatic Speech Recognition technology in the booth. This research project is co-financed by DG SCIC within the scheme Grants for actions to support training in conference interpreting. It started in September 2020 and will continue until August 2021.
The researchers involved are Bart Defrancq, head of interpreter training at the University of Ghent, and Claudio Fantinuoli (Johannes Gutenberg University Mainz), founder of InterpretBank. Helena Snoeck (Ghent University) is the research assistant for the project.
The researchers conducted a survey amongst professional interpreters from DG SCIC to find out how they could make the work of interpreters easier and develop useful tools for them. As a next step, they asked DG SCIC’s interpreters to test the solution in an interpretation environment.
How can ASR help interpreters?
Automatic Speech Recognition, in short ASR, can successfully function as a support system to interpreters in the booth. The ASR system extracts information from the speaker’s audio in order to facilitate the work of interpreters. “Where humans tend to show weaknesses, machines tend to be remarkably strong, and it’s also the other way around”, says Bart Defrancq.
The system displays the numbers, names and terminology, which are mentioned by the speaker. The terminology that is used comes from an event specific database, which allows the interpreters to benefit from the years of experience of the organisation in other conferences and meetings.
It is important to emphasize that not everything that is being said is shown on the display, as this would create additional cognitive load. The purpose of the artificial booth mate is to provide help for interpreters help in specific areas. “Interpreters can use the system just like a booth partner,” says Claudio Fantinuoli, “and consult it whenever they need help.”
How were the tests organised?
An earlier experiment by the researchers with students of interpretation showed promising results. For example, automatic number recognition improves the overall performance for almost all number types. The test showed that numbers are presented by the system with an impressive accuracy of 95%. The researchers wanted to verify these results with the professional interpreters in DG SCIC, to make sure that those who will use and benefit from this project will have the best experience possible.
Twenty volunteer interpreters participated in the test. Ten from the Dutch booth and the other ten interpreters from the Spanish booth. The idea was to run the test in a Romance and a Germanic language, because of the differences between the ways numbers are formed in the two language families (compare for example, “treinta y dos” in Spanish with “tweeëndertig” in Dutch).
During the tests, the interpreters listened to four short speeches (five minutes per speech with two minutes between speeches, i.e. thirty minutes in total), and interpreted them into their language with the help of the speech recognition solution. For the purpose of this exercise, the volunteers recorded their renditions using DG SCIC’s Speech Repository and SCICrec, the recording software offered by the platform. SCICrec gives the possibility to embed two audio paths into one file, which allowed the research team to compare the work of the interpreters.
First impressions
Carmen Gomez von Styp Rekowski from the Spanish booth participated in the tests and found it very interesting to try out something new that can lead towards a solution helping interpreters in their daily work. “The Automatic Booth Partner could be helpful in the booth, provided that the results are reliable”, she said. “I think of it as a very useful tool. It doesn’t substitute a booth partner, but it could be a great plus.”
The tool is an aid in the booth, even though it adds to the amount of information that is available, which interpreters have to deal with in real time. “There are pros and cons. In the booth, we have to process a lot of information from different sources, and it is not easy to manage our attention with more tools than we used to. Sometimes we need to find and retrieve information on the internet, because we did not receive the information in a meeting document. If the speech recognition tool is reliable, easy to use and handy, it will be very helpful.”
Do you think the Automatic Boothmate will change your interpreting experience? How? Please share your thoughts in this dedicated forum.
Links
To keep up to date and to get more information about the experiment you can also take a look at this article, this one too and visit the University of Ghent.