[aspell-devel] Understanding the suggestion algorithm

[Mindaugas Idzelis]

This email is addressed mostly to Kevin -- since I have not been able to get a hold of him through his email address I’m trying the mailing list. However, if there are any other friendly aspell hackers out there I’d like your help!

 

            I’m developing a Java-based spell checker. I’ve look at the aspell suggestion code many times, but I’m having trouble following the algorithm in the file suggest.cpp. One of the major questions I have is: when you are modifying the code to pick up other suggestions (the double-metaphone codeword for a word) are you interchanging and modifying the original checked word, or the soundslike code for that word? And why? What happens when you choose one over the other? Another question I have is – it seems that the variable “String phoneme” in the struct “OriginalWord” is never used. Is this true? This file is not very well commented – and the comments that are present are hard to follow. I think it would be a great help to all aspell developers if the algorithm was better commented or documented somewhere else.

 

            I would really appreciate if anyone can took a look at my spellchecking code. It’s available at http://jazzy.sourceforge.net/ I’m also attaching the java source for the suggestions algorithm I’ve adapted from yours. (For the complete source code of the application, check out the website) Right now, I would say it comes up with decent suggestions, but no where near the accuracy of what aspell produces. Basically, I’m strictly using the results of the DoubleMetaphone algorithm combined with some interchanging/deletions. Again, I’m neither a spelling expert nor a linguist – and I would greatly appreciate any help you can provide. Thanks

 

Mindaugas Idzelis

 

SpellDictionary.java
Description: Binary data