Recipe for Miracles to Happen

Do you work in a place with frequent project meetings? Are you bored with taking minutes during the meetings and then revising them? Would you like a miracle to happen?

At the Translate Europe Forum (TEF) conference in November 2019, Dr. Ondřej Bojar presented a recipe for such a miracle of natural language processing:

  1. Define a task
    • You want automatic minutes from meetings.
  2. Provide training and test data
    • You have recordings from not-overly-confidential meetings with agenda and final minutes.
  3. Run a “competition” ‒ a shared task
    • You give it to the best universities to compete between themselves in this task and to develop the system.
  4. Repeat until satisfied, i. e. presumably for a number of years.
    • And now, you just sit back and observe the progress until it’s exactly what I’ve wanted.

Such approaches have already been successfully employed in other areas – WMT has been doing this for machine translations for more than 10 years already; and IWSLT (International Workshop for Spoken Language Translation) has been doing this for the discourse of spoken language. Now Dr. Bojar is starting up his project ‒ ELITR ‒ to do the same for meeting summarizations.

However such a miracle needs some fuel to work. At the end of his speech, Dr. Bojar made a request for your cooperation with scientific institutes. Machines can do incredible things for us, but they need to learn ‒ and that is why the data is needed. Recordings of any meetings, regardless whether business or project, and their minutes. With them, we can prepare this as a shared task and let the best universities compete in it. As a result, the current state of the art will improve, and after a couple of years the research community will come up with a workable solution. 

Donate your minutes! 

Ideal Meeting + Minutes

Here is a snippet from the recording:

And this is what the transcript and ideal minutes could look like:

GDPR and Consents

Collecting meetings for research purposes cannot happen without proper handling of people’s consent and personal data (GDPR). ELITR sets an example of how to make this easy with ready-made (but customizable) online forms: example.

Data Partners

We are grateful to the following bodies who are already providing us with experimental meeting recordings and minutes:

  • ELITR, our EU project itself
  • Bergamot (Browser-based Multilingual Translation), another EU project with Charles University on board
  • NEUREM3 (Linguistic Structure Representation in Neural Networks), Charles University project
  • IPR Praha, Prague Institute of Planning and Development

Leave a Reply

Your email address will not be published.Required fields are marked *