[PATCH-3-5] fdo#35270 - kill first-use grammar checker freeze ...

Németh László nemeth at numbertext.org
Wed May 23 07:58:43 PDT 2012


Hi,

First of all, many thanks for the fix to Daniel and Michael. (I
haven't checked yet, but the patch probably solved the similar problem
of the Lightproof editor.)

Indeed, there are some important philosophical differences in level of
the software and also level of its language modules.

I am one of those frustrated users, who are very skeptic about
"full-featured" grammar checking, see my article here:
http://libreoffice.hu/2011/12/08/grammar-checking-in-libreoffice/. In
the article I haven't mentioned the user-related problem of the
computer aided grammar checking, the limited understanding of grammar
errors and their descriptions. Difficult orthographic rules, moreover
linguicism (linguistic discrimination) combined with the computer may
have very frustrating results. If the (word-level) spell checker
accepts the most intrusive words (yes, it accepts them, but it doesn't
suggest them), we don't have to frustrate users with potentially
dubious or dummy default grammar checker warnings. Less (false
positives) are more. (See the fixes of the false alarms in the
simplest a/an article checking.) [A real life story about orthography
and spell checking: Hungarian dictionary of Hunspell has got the
capability to check the difficult "6-3 rule" of Hungarian word
compounding. It seems, this strict rule will be removed from Hungarian
orthography (and from LibreOffice), because it was too difficult and
sometimes controversial.]

About the software: Lightproof is a lightweight proofreader. Instead
of extra libraries or dictionaries, it uses Python for a glue language
between UNO components/language resources of LibreOffice, also fast C
regex library of Python, and for parsing logical expressions in
grammar checking rules (these expressions with arbitrary user
functions give the requested flexibility in the conditions of the
rules). For example, recent Lightproof modules uses Hunspell for spell
checking and morphological analysis (in fact, UNO solution is not
depend from Hunspell: the function spell() of Lightproof will use
Voikko in a possible Finnish module, or the system spell checker on
Mac OS X), Calc for measurement conversion and rounding, moreover, the
Hungarian Lightproof module uses Numbertext Calc function extension to
check number and number name integrity in money amounts.

About the possible integration: For the easiest integration of the
result of the different grammar checker developments, in the Summer we
(Daniel Korostil GSoC student and me) will check a simple "more
grammar checking (LanguageTool, AtD)" option in LibreOffice (now only
Lightproof) grammar checker settings. (Using LanguageTool via UNO by
Lightproof could be a common/simplier interface for the different
grammar checkers.)

Best regards,
László

2012/5/21 Michael Meeks <michael.meeks at suse.com>:
> Hi Daniel,
>
> On Sun, 2012-05-20 at 19:11 +0200, Daniel Naber wrote:
>> now that this is fixed I searched the archives for the reason LanguageTool
>> is not enabled by default. The slowness you talk about in [1] does not
>> sound like general Java overhead, it sounds like exactly the issue that has
>> just been fixed.
>
>        Indeed ! :-) good to have that nailed - thank you.
>
>>  Were there other reasons LanguageTool is disabled by default? Should this
>> decision maybe re-evaluated now?
>
>        So - there is/was of course, also the Java issue - that the JRE is not
> present on many machines, whereas we bundle our own python runtime. That
> can lead either to the horrible user-experience of complaining about
> missing JRE's on first-start, and/or simply silently disabling those
> components, or worse moaning about JREs as the user types which is not
> that pleasant either. The other issue that windows Java downloads were
> advertising the competing openoffice project in the past seems also to
> be fixed now.
>
>        I guess many Linux distribution (where Java is ubiquitous) ship
> languagetool, though no doubt having lightproof also enabled by default
> presents some issues there. There is a minor licensing issue around the
> MPL vs. LGPL.
>
>        Personally, I'm looking for a good, consistent out-of-box experience,
> and (given the size of our download) am not really over-eager about
> packaging / bundling an OpenJDK for windows.
>
>        Beyond that - I've no idea :-) presumably there are philosophical
> differences between lightproof and languagetool that I'm not clued-up
> on. It'd be interesting to hear what Laszlo & you think about it, and
> other people's views on the list ...
>
>        What hope is there for sharing work, rules, etc. ? could we (for
> example) have authoritative languagetool Java code, and use one of those
> java2py things to compile it to python ? [ crazy thoughts of course ].
> Would that offend the minimal modus operandi of lightproof ? could that
> light mode be made an option ?
>
>        Thoughts on clever technical solutions much appreciated :-)
>
>        Thanks !
>
>                Michael.
>
> --
> michael.meeks at suse.com  <><, Pseudo Engineer, itinerant idiot
>


More information about the LibreOffice mailing list