The European Research Council (ERC), the body supporting and promoting outstanding scientific research since 2007, devised a new ranking to reward "15 ways in which the ERC transformed science" - a sort of the top 1% of proposals funded.

Sapienza is part of this ranking with two projects developed by Principal Investigator Roberto Navigli of the Department of Computer Science: the MultiJEDI - Multilingual Joint word sensE DIsambiguation project and the MOUSSE - Multilingual Open-text Unified Syntax-independent SEmantics project.

These projects have brought computer language learning and translation closer to the level of human comprehension. Not only has this been used to deliver large-scale lexical resources that are employed by language learners, universities and organisations across the world, but it could also have important implications for artificial intelligence, robotics and the future of machine translation.

Furthermore, Roberto Navigli will be one of the three highlight speakers in the ceremony scheduled for Thursday, May 6 to celebrate the 10,000th ERC winner. The event will take place online with the participation of the ERC President Jean-Pierre Bourguignon, the President of the European Parliament David Sassoli and the President of the European Commission Ursula Von Der Leyen.

A longstanding dream of many experts working in artificial intelligence (AI) has been to develop machines capable of dialogue and language comprehension, indistinguishable from that of a human.That is a challenge that Roberto Navigli has been grappling with for over 20 years.

'A key issue is the ambiguity of words, says Navigli. 'For example, if I say in English, "please call me a cab", what exactly am I talking about? The context should be clear for a human being, but a computer might think I am literally asking him to be called a cab. While humans take for granted that words have specific meanings based on context, it does not work that way for machines. Computers need to choose the contextual meaning of a word from a list, a process called 'word-sense disambiguation.'

In this field of research, Navigli has received two prestigious awards from ERC: a Starting grant for the project MultiJEDI - Multilingual Joint word sensE DIsambiguation and a Consolidator grant for the project MOUSSE - Multilingual Open-text Unified Syntax-independent SEmantics.

MultiJEDI is a project that between 2011 and 2016 focused on multilingual automatic natural language understanding to create large-scale lexical resources for many languages and enable multilingual text comprehension.

'We tackled the problem of lexical ambiguity," says Roberto Navigli, "and solved it by analysing the context in which words are used and exploiting artificial intelligence systems. This has led to the creation of the largest multilingual encyclopedic dictionary, BabelNet and, in 2016 of the startup Sapienza Babelscape, which deals with semantic and multilingual natural language processing, collaborating with national and European public bodies and with important international companies.'

MultiJEDI has also won prestigious award recognition, such as the Artificial Intelligence Journal Prominent Paper Award (2017) and the META prize (2015).

On the other hand, the second selected project, MOUSSE, will end in 2022. "Started in 2017 - Navigli adds - MOUSSE will allow computers to understand not only single words in context, but entire sentences, thus achieving a language-independent representation of meaning for even more sophisticated artificial intelligence applications."

The technology is based on increasingly advanced neural networks (the basis of sophisticated forms of artificial intelligence), capable of learning by exploiting mechanisms similar to human intelligence.

With additional ERC funding, Navigli now hopes to take another step toward providing a semantic basis for language-independent texts, thus allowing a computer to understand the meaning of a text regardless of the language in which it is written.

 'If we achieve this, then we will be getting closer and closer to what humans do with language', explains Navigli. 'If I read a text in French and want to translate it into English, I do not aim to translate it word-for-word but to convey the meaning. That is what interpreters do, and that is also what we are aiming for.'

Further Information

Roberto Navigli
Department of Computer Science
navigli@di.uniroma1.it