NOTES ON THE CONSTRUCTION OF THE WORD LIST A preliminary version of this spell checking dictionary was assembled with the help of my web crawler "An Crúbadán": http://borel.slu.edu/crubadan/ BUILDING TEXT CORPORA FOR MINORITY LANGUAGES Initially a small collection of "seed" texts are fed to the crawler (a few hundred words of running text have been sufficient in practice). Queries combining words from these texts are generated and passed to the Google API which returns a list of documents potentially written in the target language. These are downloaded, processed into plain text, and formatted. A combination of statistical techniques bootstrapped from the initial seed texts (and refined as more texts are added to the database) is used to determine which documents (or sections thereof) are written in the target language. The crawler then recursively follows links contained within documents that are in the target language. When these run out, the entire process is repeated, with a new set of Google queries generated from the new, larger corpus. EXTRACTING A CLEAN WORD LIST The raw texts downloaded using the scheme just described contain a lot of pollution and are unsuitable for use without further processing. I have been able to extract reasonably accurate spell checking dictionaries by applying a series of simple filters. First, the texts are tokenized and used to generate a word list sorted by frequency and the lowest frequency words are filtered out. Then, depending on the target language, correctly-spelled words from one or more "polluting" languages are filtered out to be checked by hand later. Usually this means English, but I also filter Dutch from the Frisian corpus, Spanish from Chamorro, etc. The remaining words are used to generate 3-gram statistics for the target language. These are used to flag as "suspect" any remaining words containing one or more improbable 3-grams. Finally, pairs of words differing only in the presence or absence of diacritical marks are flagged. Please contact me at the address below if you are interested in applying these techniques to a new language. Kevin Scannell <scannell@slu.edu> March 2004