The Linguist

The Linguist 58-1 Feb-Mar2019

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

Contents of this Issue

Navigation

Page 26 of 35

@Linguist_CIOL FEBRUARY/MARCH The Linguist 27 FEATURES which terminology is likely to be difficult for interpreters, it learns from databases of interpreted speech paired with gold-standard translations, so it can make predictions based on the types of terminology that have been dropped or mistranslated in the past. Finally, to decide which translations to display for ambiguous words, the system processes large sets of documents in order to determine what kinds of translations tend to occur for particular topics. These technologies, combined together, could provide timely, appropriate assistance by incorporating various signals from the source speech. Another element which the system might usefully take into account when deciding whether to provide assistance is the performance of the interpreter. When an interpreter is generating fluent output, it is likely that there is less need for the system to step in; when the interpreter is struggling, more help is likely to be required. Towards this end, we are creating systems that listen to the interpreter's output and automatically assess their performance in real time. These methods work by combining a number of signals from the original and interpreted speech: they check whether the source words align with the interpreted words; measure the fluency of the interpreted speech based on its statistical similarity to large databases of written text; and count the number of hesitations, literally translated words and other features that may be indicative of the interpreter having trouble. This component can be learned from databases of interpreted speech paired with corresponding assessments of the quality of each interpretation. Towards a brighter future These methods of predicting interpreter performance could also be a useful tool in interpreter education. They could be used as first-pass quantitative methods for assessing how well interpreting students are doing, or as a means for interpreters to review their own performance and focus on problematic areas. Automatic systems have already had a significant effect on the everyday work of translators. I believe this is just the beginning of the incorporation of automated systems in the interpretation workflow. The system I have described is just one example; the potential for automated assistance that adapts to each interpreter's needs and wants is great. The future is bright, and we look forward to working with and learning from the interpreting community – be it professionals, schools or companies – in ushering in this new era together. Even if the system is relatively confident in its speech recognition results, it is not the case that every single word will need a translation; if the interpreter is likely to be able to recall the term easily then there is no need to display it. Finally, when deciding which translations to display for a particular term, it is important that these translations are topically and contextually appropriate. When translating the English word 'capital' into Japanese, for example, it would more likely be shuto in a political context and shihon in an economic one, and the system should be sensitive to this. Designing an intelligent system Our system tackles these challenges using a fundamental technology in the field of artificial intelligence called 'machine learning'. It attempts to learn how to predict which terminology should be displayed based on the information contained in existing databases. To decide when it should trust speech recognition, it learns from databases of automatically transcribed speech, along with gold-standard transcripts, predicting when it can or cannot trust the results. To guess MENTALLY CHALLENGING Clockwise from left: Computer-assisted interpreting (CAI) could aid professionals during phone interpreting; interpreters at the WTO Public Forum in Geneva; inside an interpreting booth; and CAI may be more difficult in between formal settings IMAGES © SHUTTERSTOCK

Articles in this issue

Archives of this issue

view archives of The Linguist - The Linguist 58-1 Feb-Mar2019