MATRIC: Machine Translation and Respeaking in Interlingual Communication

Start date

2020

End date

2024

About the project

Summary

Interlingual communication in multilingual events (such as conferences, seminars, political events, etc.) is typically afforded via simultaneous interpretation, with highly skilled simultaneous interpreters working in the booth. However, with the advent of new technologies it has become possible to design alternative interlingual communication workflows. MATRIC is a pilot project that aims to explore to what extent communication can be ensured by a workflow involving a human respeaker who respeaks the same language into a speech recognition (SR) software to produce a clear transcript for machine translation (MT) engine.  

Our experiment will look at a ‘hybrid’ scenario, where respoken output, i.e. the text recognised by the SR software, is then fed into the European Union’s machine translation engine (E-Translation) to produce translated output in 6 foreign languages. The benchmark for measuring the accuracy of communication will be the transcripts of interpretation delivered highly skilled simultaneous interpreters employed in the European institutions (in the experiment we use authentic speeches from the European Parliament for which simultaneous interpretations already exist and are publicly available in the recordings accessible via the European Parliament’s Multimedia Centre). As part of this pilot initiative, we plan to develop an evaluation framework for comparing outputs from different modes but with the same communicative objective.

Project Outcomes

The aim of the project is to see if the suggested hybrid workflow can facilitate communication in interlingual settings. An additional outcome will be an evaluation framework for comparison of interlingual and multimodal data.

The following resources were developed:

  • A corpus of data from professional EN-EN respeakers aligned with data from professional interpreters.
  • An evaluation interface for comparing aligned pairs of text across two different delivery modes.
  • An evaluation framework for both quantitative and qualitative analysis of interlingual data delivered through different modes. 

Funder

Contact

For enquiries or potential collaboration on this topic please contact Dr Elena Davitti, the Principal Investigator of the project.

See other research projects carried out at the Centre for Translation Studies.