Surprise - Daniel does not behave like the Hebrew chapters. I have introduced an Aramaic switch to help govern the verses in Aramaic. When adding a Hebrew chapter to my test data I have been approaching an error rate less than 0.8% since most of the word forms are now in the data. But for Daniel, the rate was approaching 20%. Still the rules can be coded with the same strategy.
My compile now takes about 5 seconds for the 2000+ lines of code. I think it would be more efficient if I put all the tests into a database and searched them by condition, but it is a big change and I am nearly finished. In a database, assuming I got it to work, the rules would be selectable by stem and by prefix and suffix and special vowel forms. So it would be subject to some potentially easier analysis.
I will probably continue with my algorithm since changing to a data driven approach would likely occupy some time. The process is slightly addictive so I hope it's done soon. Short holiday next week and must go get some sunshine in the garden now.
No comments:
Post a Comment