Thread by @translation_eu

Good morning from Brussels! We're attending #METAFORUM2022 today, bringing you updates from the world of language technology.

There are roughly 100 people in the room. And several hundred following along on YouTube: .

Join us!

Kicking things off now is @GeorgRehm who also attended our hybrid DG TRAD Conference last year. He underlines that #multilingualism is at the heart of the European idea. 24 EU languages plus countless regional and minority ones!

@GeorgRehm META-Net and others have been working on digital language equality for over a decade. One important contribution by @Europarl_EN was the resolution on this topic: europarl.europa.eu/doceo/document…

@GeorgRehm @Europarl_EN Language technology (including #AI or #NLP) is a huge market in the 50 billion USD range. Europe needs to become a key player, says @GeorgRehm. Overcome fragmentation and treat all languages equally.

If you'd like to ask questions, you can attend via Zoom. Link is on this page: european-language-grid.eu/meta-forum-202… #Interactive

Lastly, there is an online project expo that you can check out here: european-language-grid.eu/meta-forum-202… (during the lunch break)

A word of welcome now from @HajicJan (Charles University) - and a look back at what has happened since the last META-Forum conference.

Here's @JonasAndrulis (founder of AI startup @Aleph__Alpha) with a recorded keynote on AI sovereignty and multi-modal models.

AI has been somewhat disappointing for certain tasks, for example due to bias. Simply put, the world is too complex to be easily broken down into categories. So a model will see two motorcycles where there's actually a horse-drawn carriage.

Humans have extremely good "world models" (with rewards for correct prediction), use "fast thinking" (D. Kahneman) and pattern recognition. Engineers are trying to teach machines how to get better at predicting.

You can see them try (and sometimes fail) when using word prediction in a text message or for suggested search terms in your search engine of choice. The challenge here is to better understand context.

Modern models can do impressive things with a mere prompt: Produce a text that is a letter from Goethe to Schiller. The result may be accurate in style, but if it mentions television ads, well...

This also works across languages: You can ask the model a question in English about a text in French. 🤯 #multilingualism

The big challenge now is to add more cultural context. And context that is not just US-centric… As Andrulis says: The world is exciting and wild! For some wild examples, see @Sigmoid_Freud 😉

AI models are becoming bigger and bigger all the time to cope with the amount of data and the challenging tasks. This is the foundation of the next industrial revolution.
(For critique of this approach, see the work of @timnitGebru, @mmitchell_ai et al. - "stochastic parrots")

After the excellent keynote, a quick recap from @GeorgRehm on what the "European Language Grid" is. A platform for the language technology community and a hub for sharing a vast amount of resources: european-language-grid.eu/grid/

Stelios Piperidis (@athenaRICinfo) now gives us an overview of the architecture powering the #EuropeanLanguageGrid. The nerdier side of #METAFORUM2022!

Impressive numbers now from @kbontcheva: The #ELG provides hundreds of services for EU official (and other) languages. #MachineTranslation, speech and character recognition, text-to-speech and more! #metaforum2022

We're back from the coffee break - attendees were very happy to connect in real life again. Next up: a functional overview and demo of the new European Language Grid. #metaforum2022

For a live demo of #SpeechToText and #MachineTranslation, check out the ELITR page here: quest.ms.mff.cuni.cz/elitr/demo/

🔴 New live-streaming link! 🔴

This should be really interesting: A selection of #ELG projects, including for the Basque and Welsh languages.

An important part of working with big language data is anonymisation, especially for public administrations (like us 😉). @manuelhrrnz presents the MAPA project on #ELG.

european-language-grid.eu/wp-content/upl…

#metaforum2022

Now: @TiedemannJoerg on OPUS-MT

https://twitter.com/PatrickPlaten/status/1532397965909827584

Digital language equality counts! One example: @ileturia presents a smart speaker for the #Basque language, based on the @mycroft_ai project. Works in Basque and brings you Basque news and music, among many other things. (Can it play Europarl Radio, too? 😉)

What about Welsh? Gareth Watkins from @BangorUni about the @Cymraeg push for more language technology support. More information in this PDF: gov.wales/sites/default/…
#metaforum2022

Language = more than words! Svetla Koeva about @BgSciencesBAS's *multilingual* annotated image corpus. dcl.bas.bg/en/mic21/

What are microservices? @tlindhknuutila explains. #ELG #METAFORUM2022

Curious about all these projects (and more)? The online project expo is now open: european-language-grid.eu/meta-forum-202…. See you after lunch, we resume at 13:30.

We're back - with a hybrid panel discussion on AI initiatives around European languages. The panel includes @OpenGPTX, @ai_humane, @vision_claire, @CLARINERIC and @EFNILorg.

Europe needs its own language models and has some catching up to do, says Joachim Köhler (@eis_bonn). We see a pattern here…

New developments in AI and NLP are putting semantics and formal categorisation on the backburner - James Crowley, @GrenobleINP at #metaforum2022

Some discussion now about the role of EU institutions: one attendee says they "don't believe in minority languages". @phgelin offers some context, such as limited resources. Commission and Parliament DO support language equality in many ways!

@phgelin @dfiser3 also points to the multilingual corpora the EP provides (or is used as source for) - e.g. interpreting studies (@terptraining).

"If we don't take action, we'll all end up working for Google & Co." - James Crowley

Cooperation is key. Countries cannot build the infrastructure alone. Scientists have shown how to do this through projects like CERN. - Philipp Slusallek (@vision_claire) #metaforum2022

The community needs to evolve from text processing to proper natural language processing - @dfiser3 #HearHear #metaforum2022 #NLP

Recaffeinated, we're about to start the language technology industry panel, including Yolina Petrov (@Identrics), @schwarzpe (Phonexia), Fredrik Olsson (@gavagai_corp), Enric Plana (@M47_AI), Andres Mellik (@Cognuse) & Oskars Rumpēters. #metaforum2022

@Identrics @schwarzpe @gavagai_corp @M47_AI @Cognuse Here's what European lang-tech companies actually do: automated media monitoring / info extraction / text classification / customer feedback analysis / labelling and annotation of data / use data for disease management and rehabilitation … #MetaForum2022

… speech analytics / voice authentication / conversational #AI / services for law enforcement - and these are just a few examples!

As we approach the end of day one, @phgelin introduces us to the concept of the "language data space" and what the European Commission contributes. #MetaForum2022

@phgelin The European Data Space ecosystem at a glance #MetaForum2022

@GeorgRehm now wraps up the day with an outlook on the future of the European Language Grid. Since the project is about to reach its end, a not-for-profit entity will take over.

Lots to chew on today - and lots of interesting things to come, such as two (!) open-access books on European Language Equality and the European Language Grid later this years. We also learn about new and related projects. Looking forward to tomorrow!

#MetaForum2022