FrameNet Brasil is a center for the development of Natural Language Understanding solutions in which linguists and computer scientists from different departments of the Federal University of Juiz de Fora join forces for the creation of innovative products.
FN-Br is currently the home for the following research initiatives:
The Frames and Constructions project implemented an online Constructicon for Brazilian Portuguese, which can be used as an input for the development of IT solutions related to Natural Language Understanding. Founded on the theoretical constructs of Frame Semantics and Construction Grammar, the main purpose of the project was that of, based on corpus evidence, identifying, analyzing and annotating constructions. The annotation of the exemplar sentences, as well as the analysis of the constructions is made through the FrameNet WebTool.
This project complements a broader initiative in which it is included: FrameNet Brasil. While the FN-Br Lexicon identifies and analyzes frame-evoking lexical units, a Constructicon is a repertoire of syntactic constructions, structures whose recognition, treatment and interpretation are key for Natural Language Understanding.
The project is developed in collaboration with the (a) Swedish FrameNet++ and Swedish Constructicon projects at Göteborgs Universitet, (b) the German FrameNet and Construction project at Heinrich-Heine Universität Düsseldorf, and also with the Berkeley FrameNet project. Such a collaboration aims at allowing for the contrastive study of constructions in Brazilian Portuguese, Swedish, German and English, as well as for the definition of construction annotation and alignment standards for multilingual purposes.
The following projects have been concluded by the FN-Br Lab:
he FLAME (Frame-based Language Aid Multilingual Experiment) project developed a multilingual FrameNet for the Summer Olympics, to be held in Rio de Janeiro, Brazil, in 2016. Building on the previous experience of developing the FrameNet Brazil World Cup Dictionary (www.worldcupdictionary.com.br), we present a multiplatform solution for the communication problems common in major international events – m.knob. Such a solution combines a domain-specific dictionary, an automatic translator and a knowledge retrieval engine focused on enhancing the Olympic experience in Rio. To achieve this goal, we combine the infrastructure of FrameNet with domain ontologies, statistical machine translation applications, linked open data and gamification elements in a multilingual perspective, covering four different languages: Brazilian Portuguese, English, French and Spanish. The FLAME project aims also to serve as proof of concept for the deployment of frame-based knowledge databases in the development of solutions for multilingual translation and knowledge retrieval.
The Copa 2014 FrameNet Brasil project (Salomão et al. 2011, 2013) developed a frame-based trilingual (Portuguese – Spanish – English) electronic dictionary covering the domains of soccer, tourism and the World Cup. Focused on human users, the final product was used by tourists, tourism professionals, journalists and the staff involved in the organization of the FIFA 2014 World Cup in Brazil.
The project uses the basic infrastructure, analytical categories and methodology developed for FrameNet (Fillmore et al. 2003; Baker et al. 2003; Ruppenhofer et al. 2010), which can be defined as an application of Frame Semantics to practical lexicography aimed at:
Copa 2014 was implemented as a web app. The database comprises interconnected framenet-style data – that is, frames, lexical units, frame-to-frame relations, annotated sentences etc. (see Baker 2003 et al. for a comprehensive view of FrameNet’s database structure) – produced for each of the target languages of the dictionary. The query system is a web service comprising a set of possible queries, which are accessible through a web interface. Query results are shown through the same interface. Users may access the information in the dictionary in four different ways: by searching a word, by typing a sentence, by browsing the list of frames grouped by cognitive domains, and by exploring the frame grapher.