Towards a Phonetic Brazilian Portuguese Spell Checker

Spell checking is no longer considered a big challenge for natural language processing, at least regarding the task of correcting documents during edition. Nevertheless, without human interaction, it is necessary to automatically choose the word that will more likely correct the misspelled word. Also, there is a further difficulty for spell checking: new types of errors on the web material have emerged due to the increasing participation of general public, especially when expressing opinions, feelings and requests, which take many characteristics from the spoken language. This paper presents the first efforts towards a new Brazilian Portuguese (BP) spell checker to deal with the challenges that emerged in the automatic processing of a web corpus, including a new phonetic algorithm to specifically address spelling correction in BP. The speller proposed here is able to correct 16% more words than Aspell, in a web corpus composed of reviews of products.