FrameNet Brasil (UFJF)
Natural Language Understanding with structured computational semantics
FrameNet Brasil (FN-Br) is a Computational Linguistics lab of the Graduate Programs in Linguistics and Computer Sciences at the Federal University of Juiz de Fora, Brazil. It is also the Brazilian branch of the FrameNet initiative, a global effort to build language resources based on Frame Semantics and Construction Grammar started in 1997 by Charles Fillmore, in the International Computer Science Institute, in Berkeley California. FN-Br builds lexical and grammatical language resources and apply them to address issues in Natural Language Understanding. Frame Semantics takes a cognitive approach to study the meaning of words and expressions. Instead of using formal logics (a common view in computational semantics field), the meaning is structured considering how the language users understand and use the words in a given context. FrameNet is the computational implementation of this idea, building a lexical resource cognitively motivated.
FN-Br implements a relational database storing a set of frames - or scenes -, the elements structuring these frames, the language specific material (words, MWEs and grammatical constructions), and several typed relations between all those items. Such a database provides a model and a structure for semantic representation of the knowledge about a given language. Currently, FN-Br includes data about Brazilian Portuguese, English and Spanish.
FN-Br database is the base for the development of different software: a frame-based multimodal web annotation tool, a domain-specific dictionary - the FrameNet Brasil World Cup Dictionary; a system for recommending tourist attractions - m.knob; a semantically enhanced sentence translator and a constructional parser - CARMA.
FrameNet Brasil (UFJF) 2020 Projects
Multilingual Projections of Semantic Frames and Frame Elements using Modified Existing Parsers and Neural NetworksMy proposal aims to accelerate future semantic annotation projection research by creating two main deliverables. First, before the GSoC 2020 starts,...
New Frame-Based Image and Video Annotation Pipeline for the FrameNet Brasil Web Annotation ToolThe main motivation for this project is to create a more automated and simplified video annotation pipeline using both image and textual data for...