The Linguist is a languages magazine for professional linguists, translators, interpreters, language professionals, language teachers, trainers, students and academics with articles on translation, interpreting, business, government, technology
Issue link: https://thelinguist.uberflip.com/i/1077738
26 The Linguist Vol/58 No/1 2019 ciol.org.uk/tl FEATURES management of terminology dictionaries and jobs. CAI software can provide real-time assistance during the interpreting session, using automatic speech recognition to 'listen' to the original speaker and display translations of salient terms when they are detected. This allows machines to help with some of the more difficult aspects of interpretation, such as preparation and quick recall of technical terms, while allowing interpreters to do what they do best: understand the core of what the speaker wants to say, and convey this in a reliable and culturally appropriate way. I think that these developments are particularly exciting as they view computers as tools that professionals can use to improve their performance, instead of envisaging them as drop-in replacements for highly skilled professionals performing a difficult and nuanced job (a view I see as both naive and unrealistic). Creating a distraction? One difficulty with CAI interfaces is the potential for distraction. Given that simultaneous interpretation is already a cognitively challenging task, at its worst, a CAI interface could be another distraction that the already overloaded interpreter needs to pay attention to. Thus, it is imperative that a good interface understands when assistance is necessary, and only steps in when its help will be beneficial. To this end, a group of researchers at Carnegie Mellon University, the University of Maryland and the University of Washington are examining the potential of CAI interfaces that can automatically determine when and what type of assistance is required. Sponsored by the National Science Foundation in collaboration with interpreter partners in Maryland, Bologna and Murcia, we aim to apply techniques from artificial intelligence to enable a CAI system to detect when it should step in and when it should step back in order to avoid confusion or mental overload. When the CAI system uses automatic speech recognition to listen to the source sentence, it needs to detect the terminology for which interpreters may require assistance, and what translations to display. A first requirement is that the system must be highly confident that what it 'heard' is correct – if there is a chance that the speech recognition made a mistake, it is probably better for the system to back off. Is computer-assisted interpreting the future for the industry or simply a distraction, asks Graham Neubig I am currently a professor of computer science, but some 10 years ago I worked for local government in Japan, doing a variety of jobs where native English proficiency was useful: translating, proofreading, greeting foreign dignitaries and interpreting. Among these experiences, I found that interpretation was both an essential tool for crosscultural communication and a really hard job. The difficulty of the task was compounded in simultaneous interpretation. Listening to the original speaker and accurately generating speech in the target language at the same time requires careful background research into the topics that a speaker may cover, intense concentration during the session, and often collaboration as a two- person team to make it possible to switch off when fatigue starts to set in. Recently, a small group of technologists and researchers (myself included) have started to ask the question: what can computers do to help interpreters tasked with this extraordinarily difficult job? This has led to a new genre of technology called computer-assisted interpretation (CAI), aiming to provide automatic assistance for interpreters preparing for assignments, including in the interpreting booth. Commercial software such as InterpretBank (www.interpretbank.com) and Interpreter's Help (www.interpretershelp.com) attempt to assist with a number of aspects of interpretation, including automatic Does not compute? To guess which terminology is likely to be difficult for interpreters, the system learns from databases