Publicação

A BERT-powered writing assistant for academic purposes in european portuguese

Ver documento

Detalhes bibliográficos
Resumo:In this paper, we will present the process of developing a resource that we consider to be useful for both native and non-native college students in the process of writing Portuguese academic texts: a BERT-powered Writing Assistant for academic purposes in European Portuguese. The Writing Assistant includes two main components: a phrase bank, that will be created using open scientific data in the form of scientific papers found in repositories, and a search engine, that uses BERT models for semantic searches. To create the phrase bank we will loosely follow the methodology developed by John Morley, creator of the Academic Phrasebank of the University of Manchester. The phrase bank will be based on 40 scientific papers taken from the repository of University of Minho. The corpus will be initially annotated, using some of the categories proposed by Morley, then the categories will be revised to better represent the reality of Portuguese academic discourse. The annotated phrases will then be simplified and stripped of any particular academic content. This phrase bank will “feed” the search engine. The search engine works with BERT machine learning models that allow us to make semantic searches. Students would just have to write a word, expression or sentence in the search bar to find equivalent or similar expressions on our phrasebank, even if the user has little to no knowledge of the vocabulary used in academic discourse, because Bert models are able to infer semantic context and find relevant results.
Autores principais:Araújo, Sílvia
Outros Autores:Aguiar, Micaela Maria Assis; Monteiro, José
Assunto:Academic Literacy Search Engine Phrase bank BERT
Ano:2023
País:Portugal
Tipo de documento:comunicação em conferência
Tipo de acesso:acesso aberto
Instituição associada:Universidade do Minho
Idioma:inglês
Origem:RepositóriUM - Universidade do Minho

Registos relacionados

Registos relacionados