Re: Thoughts on using aspell for Indian language checking

On Fri, 17 Nov 2006, Gora Mohanty wrote:

> On Fri, 2006-11-17 at 00:18 -0700, Kevin Atkinson wrote:
>> On Thu, 16 Nov 2006, Ethan Bradford wrote:
>>
>>> The patch was against 0.60.4.
>>>
>>> I don't really have an opinion on your design questions, except a question
>>> of my own: why do you want the extra information?  I wanted to get the
>>> scores out so that I could (1) filter on them and (2) combine them with
>>> other information (frequency information from another source) to better sort
>>> the result list.
>>
>> I would like to know that too.
>
> I think that at least the word score and soundslike score should be
> available. I am not so sure about the scores after each processing
> step. The reason that I wanted them in the first place was that
> suggestions for Hindi seemed to get better if I turned off
> fine_tune_score(). However, at the moment this needs to be tested
> in more quantitative terms.

Then I suggest you do what I do.  Hack suggest.cpp and try both versions. 
Rather than go through the trouble of exporting Aspell internals.

>
>> I want to be careful about exposing too much of Aspell internals since
>> they are subject to change.
>
> I am in two minds here. I see your point about not arbitrarily opening
> up everything. However, I would guess that anyone mucking around with
> testing probably wants to get their hands dirty. Would putting access
> to aspell internal details in another header file, say aspell_testing.h
> be an acceptable alternative? People could then be warned that the
> interface to aspell_testing.h was subject to change.

If you just want to do this for testing I think going through the trouble 
of designing an API to export Aspell internals is an overkill.  You 
should do what I do and modify Aspell directly.   If you need more 
information on what is going on just print it to stdout.  You will 
notice that there is a "DEBUG_SUGGESTION" macro which if defined outputs 
some useful info.