Re: Spellcheck Problem
Hi,
> using aspell
> seems to suggest much better alternatives than openoffice's built in
> spellchecker.
BTW, I looked at the en_GB.aff file and they do *not* use the replacement
table feature of MySpell. So their suggestions mechanism will ONLY catch
common typing problems (additional letter, ommitted letter, missing space,
swapped adjacent letters, etc) and won't do any substring substitutions at
all which will allow common phonetic spellings mistakes to be replaced
with correct versions.
See the en_US.aff file and look at the REP table entries. These will
hanbdle a number of common phonetic mappings to proper substrings for
better suggestions.
They can improve their suggestions by creating a table similar to the one
used in en_US.aff
With those in place MySpell will do quite well in making typical
suggestions and correcting typical typing errors including phonetic ones.
Kevin Atkinson's spellchecker will often do a better job at *very* poorly
spelled words (needing multiple phonetic substitutions and multiple
character misplacements) at the cost of using *two* hash tables (one
dedicated to phonetic use only - a memory hog), no affix compression at
all (I gave the myspell affix code to him but he has not integrated it in
yet) which results in simply huge dictionaries for some European languages
and consumes precious memory, the need to use endian specific binary
dictionaries (instead of building the dictioanry from simple text files on
the fly) thereby creating the need to have arch specific dictionary files
that can't be edited by users....
So please consider your common phonetic spelling mistakes and contribute a
replacement table (REP entries) to the authors of the en_GB dictionary to
help improve the suggestion capability of MySpell or contribute
development help to Kevin Atkinson and help him make Aspell build on
non-gcc based systems, incorporate affix compression, allow on the fly
dictionary building from text files (probably can't be done for phonetic
tables...), etc.
Kevin
Reply to: