ReGra Project: new challenges for automatic grammar revision

• Grant number: 00/10688-0
• Support Opportunities: Research Grants - Research Partnership for Technological Innovation - PITE
• Start date: March 01, 2001
• End date: February 28, 2003
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Maria das Graças Volpe Nunes
• Grantee: Maria das Graças Volpe Nunes
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Company: Itautec Philco S/A
• City: São Carlos

Abstract

ReGra is a grammatical reviser for Brazilian Portuguese that is regarded as the most wide-ranging and efficient of its category on the market. Microsoft acquired a license from Itautec-Philco, the co-sponsor of the project, and today makes ReGra available through its Office 2000 package, ensuring worldwide distribution. The ReGra Project was honored with two prizes, the Pawn of Technology, awarded by the São Carlos High Technology Park Foundation, in 1998, and the Alcatel National Prize for Technological Innovation, in 1999. This very positive scenario may give the impression that the quality of grammatical revision by ReGra had now reached satisfactory levels, and that the development, properly speaking, of the reviser had been concluded. However, tests showed that ReGra’s current performance still fell considerably short of what was desirable, with a significant number of omissions (39.48 per cent of all the incorrect sentences were not corrected) and of unwanted actions by the tool (there were 1.03 incorrect actions for each correct action). Therefore there still existed ample margin for work on improvements, which resulted in a far greater evolution than had actually been observed. This evolution demanded efforts to improve the current reviser, but - more importantly - it called for a change of paradigm, both in the use of linguistic strategies and in computer implementation, in particular because the performance of the current version had practically reached its peak in the approach adopted up to this moment. It proved to be indispensable for the strategies used in ReGra to be revised, and for new modules to be added to the existing ones for the revision of texts, in order to expand the number of correct actions of the tool, and, consequently, to reduce the number of its omissions and over-corrections. (AU)