Improving Speech Translation with Linguistically-Informed Representations

Add to your list(s) Download to your calendar using vCal

• Elizabeth Salesky (Johns Hopkins University)
• Friday 27 November 2020, 12:00-13:00
• Zoom.

If you have a question about this talk, please contact Guy Aglionby.

Join Zoom Meeting https://cl-cam-ac-uk.zoom.us/j/99290674837?pwd=cEYvd0pSSXgvN2VERUpmblZ3QzJiZz09

Meeting ID: 992 9067 4837 Passcode: 999939

End-to-end models for speech translation (ST) more tightly couple speech recognition (ASR) and machine translation (MT) than a traditional cascade of separate ASR and MT models, with simpler model architectures and the potential for reduced error propagation. However, several challenges still remain to make end-to-end models perform as well as cascaded models, particularly in low-resource scenarios. Further, in the move towards more task-agnostic neural architectures, inductive biases for each task have largely been removed. In this talk, I will discuss some important considerations for building speech translation models (and why we should still draw inspiration from cascades), as well as three methods to re-introduce model biases through phonologically-informed representations and the situations where they are most beneficial.

This talk is part of the NLIP Seminar Series series.

This talk is included in these lists:

• All Talks (aka the CURE list)
• Cambridge Centre for Data-Driven Discovery (C2D3)
• Cambridge Forum of Science and Humanities
• Cambridge Language Sciences
• Cambridge talks
• Chris Davis' list
• Department of Computer Science and Technology talks and seminars
• Graduate-Seminars
• Guy Emerson's list
• Interested Talks
• Language Sciences for Graduate Students
• NLIP Seminar Series
• PMRFPS's
• Simon Baker's List
• Trust & Technology Initiative - interesting events
• Zoom
• bld31
• ndk22's list
• ob366-ai4er
• rp587

Note that ex-directory lists are not shown.