![]() |
|
GoSPELL
A Linguistic Knowledge-Based Spell-Checker for Brazilian Portuguese |
Starting Time: 1999
Goal
Develop an interactive isolated-word spelling-error corrector for Brazilian Portuguese that will not fail users when they are lost for words, or rather, good guesses of correction. We call this feature “utility-orientation”, which we intend to accomplish by means of massive (psycho)linguistic modeling.
This project focuses on the generation of linguistically plausible/helpful/useful suggestions of correction for a given string, presumed to be a word that has undergone some kind of recoverable, i.e. justifiable, distortion on the user’s part. Thus the detection of such a string as a spelling mistake is rather an assumption than an aim, but a very reasonable one, as we have ReGra’s vast instant-access lexicon at our disposal.
Current Status:
2001 –
· ongoing MSc project;
· finishing up models of (chiefly linguistic) phenomena connected to spelling-error production (knowledge representation);
· specifying the architecture and prototypical correction strategy.
GoSPELL's Features:
·
coverage of a range of cognitive
errors predictable on the basis of knowledge in the domains of phonetics,
phonology or morphology;
·
recovery of multiple errors;
·
attempt at validating neologisms
by assessing the morphological plausibility of unknown words and hypothesizing
corresponding blocking forms;
·
inference of inflections of new
words that have been taught by the user.
GoSPELL's Expected Products:
·
the spell-checker proper;
·
valuable insights into Portuguese
spelling, phonology and morphology;
·
a unified computational model of
word distortion in Brazilian Portuguese;
·
cross-language software components
for word analysis;
·
an instantiation of these
components for Portuguese employing…
·
… represented knowledge for word
and spelling analysis in Portuguese;
·
cross-language architecture for
both spell-checking and error correction in general.
Team
Jorge
Marques Pelizzoni (MSc student)
Maria das Graças Volpe Nunes (supervisor)
CAPES:
1999-2001;
Finep
(PADCT-CE, Proc. 88-98-059100-02-01): 2001.
FAPESP
(Proc. 00/10688-0): 2001-…;
Jorge Marques Pelizzoni (e-mail: jorgemp@icmc.sc.usp.br)