Notice: Due to the high degree of overlap, this project has been abandoned in favor of Enchant

SpellCore is a an open-architecture spelling/orthography-changing engine for minority languages.

In a typical language development project, workers use four or more applications to produce vernacular materials (i.e. reading primers, dictionaries, books). Because these are non-commercially interesting languages, ready-made spell-checkers are not available. Some applications have their own way to spell check anyhow, but the resulting list of good and bad words is not generally reusable in the other applications. So the first thing SpellCore offers is a database of spelling information which is usable by multiple applications.

Recognizing the needs of languages where 'correct-spelling' is an elusive target, SpellCore goes way beyond 'correct/incorrect' lists to include plug-ins that measure suspiciousness. That is, if a word is not known to be right or wrong, does it seem likely to be wrong? Two measures that have proven useful are:

• Structure (e.g. syllable structure, composed of morphemes found in a dictionary and conforming to known morphological grammar rules).
• Similarity (e.g. one key away from a known-good word see edit distance)
• Equivalent Transliterations (Eaton & Penny)

Architectural Overview

A core library/api, named SpellCore (previously codenamed Panang), sits in-between the client pieces and particular plugins. Note: many possible plug-ins are shown here, but we don't need to build all these pieces to have a great system. The point is to have natural places to add in approaches that we know folks are currently using out there, in various kinds of languages.

Related Projects

See SpellCenter.

Who's doing this?

This project is being launched by a Palaso (PAyap LAnguage SOftware). Palaso is a small group of field-based developers working for SIL International, some working in the Linguistic Institute of Thailand's Payap University. We welcome others to join in.

Status

July 2007. Christopher Wilks and Eric Albright added support for HunSpell and ASpell engines, so our major-language support is strong. Next up is minority language support.

October 2007. We found out about Enchant so have moved our efforts to contributing toward Enchant.

Please post to the mailing list with your ideas/questions.