Menu
MapConnectionsTopicsBulgariaThe Night (K-TRIO)Marie Skłodowska-CurieFor studentsFor teachersMethodologySurprise me
H2020Individual fellowship2015–2017

LOVe · Linking Objects to Vectors in distributional semantics: A framework to anchor corpus-based meaning representations to the external world

Horizon 2020 — Marie Skłodowska-Curie Actions

Duration
2015-06-01 → 2017-05-31
EU contribution
€180,277
Participants
1
Scheme
MSCA-IF

Lines connect the coordinator with its partners.

Results in brief

Linking Objects to Vectors in distributional semantics: A framework to anchor corpus-based meaning representations to the external world

"We use language to talk about the world. If I say ""I like J. K. Rowling's books"", you will be able to understand what I say because you are able to connect the expression ""J. K. Rowling"" to a particular entity in the world, even if that person is not there in the moment. This, which comes naturally to us humans, is very difficult to do for machines. My project aims at understanding reference, the crucial property of language that allows us to link it to the external world, using computational tools. Specifically, it exploits distributional semantics and deep learning, a family of methods that can learn to represent language directly from language data generated by humans, such as texts drawn from the internet. The project has three main objectives: 1) To explore the representation of entity names (such as ""J. K. Rowling"") in distributional semantics. 2) To connect referential expressions (""the mug"", ""the book I bought"") to objects depicted in images. 3) To develop a semantic framework that combines information about entities and concepts in a meaningful way. We find that it is possible for current distributional models to learn to refer directly from data, and that distributional representations usefully represent the meaning of entity names. The project advances our scientific understanding of language, a defining trait of the human species, and makes significant progress towards building machines we can talk to, with the ensuing impact on our everyday lives. "

Data: CORDIS, © European Union

Project objective

Language mediates between concepts in our mind and the things they refer to in the world. Semantic theories are typically biased towards conceptual or referential aspects. My goal is to develop a theory of meaning that takes both aspects into account, and is supported by computational modelling experiments, so that it will also enable computers to match linguistic expressions with entities in the world. This is a highly interdisciplinary proposal that will bring computational linguistics, artificial intelligence, and theoretical linguistics forward.My model is based on distributional semantics, a scalable and flexible approach to computational semantics that, by inducing meaning representations from naturally occurring data with statistical methods, can model large portions of the lexicon and account for nuances in meaning that pose difficulties to traditional semantic theories. Distributional semantics has so far largely eschewed the reference issue, by testing its models on language-internal tasks. The project bridges this language-world gap, and integrates the distributional framework into a referential semantic theory. The project promises to advance our scientific understanding of language, a defining trait of the human species, and to make significant progress towards building computers we can talk to, with the ensuing strong impact on our everyday lives.Even though I am an established researcher in computational semantics and also contributed to semantic theory, I still need to fully develop my own line of research to become a leading, independent researcher in Europe. Carrying out the present proposal at the University of Trento CLIC laboratory will be a fundamental step towards achieving my goal, since CLIC is a world leader in distributional semantics. Conversely, my unique profile, addressing theoretical linguistic questions through computational means, will fill a gap in the lab, widening the scope and outreach of the research conducted at CLIC.

Original text from CORDIS.

Participants

  • UNIVERSITA DEGLI STUDI DI TRENTO · TrentoCoordinatorItaly

Links

Data: CORDIS, © European Union