Study and development of methods for automatic detection and correction of errors and inconsistencies in syntactically annotated corpora

• Grant number: 13/18090-6
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Effective date (Start): January 01, 2014
• Effective date (End): July 31, 2016
• Field of knowledge: Interdisciplinary Subjects
• Principal Investigator: Charlotte Marie Chambelland Galves
• Grantee: Pablo Picasso Feliciano de Faria
• Host Institution: Instituto de Estudos da Linguagem (IEL). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Associated research grant: 12/06078-9 - Portuguese in time and space: linguistic contact, grammars in competition and parametric change, AP.TEM
• Associated scholarship(s): 14/17172-1 - Study and application of explicit grammatical formalisms in the detection of inconsistency in treebanks, BE.EP.PD

Abstract

This research project has as its main goal the study, application, and further development, of computational methods for automatic detection and correction of errors and inconsistencies in syntactic annotated corpora (treebanks) such as the method proposed by Kato & Matsubara (2010), which is based on Synchronous Tree Substitution Grammar (Shieber & Schabes, 1990). Tied to the research project "Portuguese in time and space: linguistic contact, grammars in competition and parametric change" (Thematic Grants, FAPESP 12/06078-9), the present proposal will join their efforts, particularly those aimed at extending and consolidating Tycho Brahe Corpus (CTB), inasmuch as the method to be developed is to be evaluated and applied to it. As expected results, we have (i) the development of a new method for automatic detection and correction of inconsistencies with better results than the current state-of-the-art methods; (ii) the inclusion of the method as part of the syntactic annotation process and the application of it to CTB in order to make available a revised version of the current corpus; and, finally, (iii) based on the analysis of the most recurrent types of errors detected by the method, the update of the manual annotation guidelines in order to better train the team of annotation reviewers.