Please use this identifier to cite or link to this item: http://hdl.handle.net/1843/57498
Type: Artigo de Evento
Title: On auxiliary verb in Universal Dependencies: untangling the issue and proposing a systematized annotation strategy
Authors: Magali Duran
Amanda Pontes Rassi
Adriana Silvina Pagano
Thiago Alexandre Salgueiro Pardo
Abstract: Auxiliary verbs are universally recognized as components of verbal constructions. While there is no shortage of scholarship on these verbs in various linguistic traditions, uncertainty still remains on the best way to annotate them for Natural Language Processing (NLP) purposes. This paper reviews the evolution of the concept of auxiliary verbs to gather insights into forms of representing them in an annotation scheme and raises some issues with a view to leveraging the potential afforded by them in different NLP tasks. Using Brazilian Portuguese as an instance language and Universal Dependencies (UD) as annotation model, we argue for (i) annotating inflected verbs as heads, (ii) annotating auxiliary interdependence in an auxiliation chain; and (iii) adopting a more consistent treatment of auxiliaries to encompass tense, aspect, modality and voice in auxiliation chains. We further propose auxiliary type as a feature to be annotated which can be easily implemented in existing and new treebanks with substantial gains in enriching the information that can be extracted for different NLP applications.
Subject: Ciência da Computação
Processamento da linguagem natural (Computação)
Língua portuguesa - Sintaxe
language: eng
metadata.dc.publisher.country: Brasil
Publisher: Universidade Federal de Minas Gerais
Publisher Initials: UFMG
metadata.dc.publisher.department: FALE - FACULDADE DE LETRAS
Rights: Acesso Aberto
URI: http://hdl.handle.net/1843/57498
Issue Date: 2021
metadata.dc.url.externa: https://aclanthology.org/2021.depling-1.pdf
metadata.dc.relation.ispartof: International Conference on Dependency Linguistics
Appears in Collections:Artigo de Evento



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.