Using an lttoolbox dictionary

En français

This page is intended as an answer to the question "I've found one of these .dix files; how can I use it to analyse text?" First of all, it is worth explaining what a .dix file is: a finite-state transducer for a language encoded in XML. More information on this can be found at the page lttoolbox and monodix basics, but this page only concerns how it is used.

(If you haven't found a .dix file for your language yet, see List of dictionaries.)

Requirements

The most basic requirements are:

• lttoolbox — A finite-state toolkit
• apertium — A machine translation software platform

The second is necessary for the deformatters. The tools in lttoolbox have a set of escaped characters which must be escaped in running text (see Apertium stream format).

The page Installation shows how to install lttoolbox and apertium. On most systems, you don't have to install more than the Prerequisites.

Using the dictionary

Then, you take the .dix file (e.g. apertium-bn-en.bn.dix) that you have downloaded, and compile it:

Compile

See also: Compiling dictionaries

This compiles an analyser:

$ lt-comp lr apertium-bn-en.bn.dix bn.analyser.bin
final@inconditional 8 75
main@standard 6403 13351

Analyse

Note that the apertium-destxt command is important.

$ echo "উইকিপিডিয়ার বাংলা সংস্করণে স্বাগতম। এই বিশ্বকোষে যে কেউ অবদান রাখতে পারেন। ২১,২৫৫টি ভুক্তির ওপর কাজ চলছে।" | apertium-destxt | lt-proc bn.analyser.bin 
^উইকিপিডিয়ার/*উইকিপিডিয়ার$ ^বাংলা/বাংলা<adj><mf>/বাংলা<n><mf><nn><sg><nom>/বাংলা<n><mf><nn><sg><obj>$ ^সংস্করণে/*সংস্করণে$ ^স্বাগতম/*স্বাগতম$^।/।<sent>$ 
^এই/এই<det><dem>$ ^বিশ্বকোষে/*বিশ্বকোষে$ ^যে/যা<prn><p3><infml><rel><aa><mf><sg><nom>$ ^কেউ/কেউ<prn><p3><aa><mf><sp><nom>$ 
^অবদান/অবদান<n><nt><nn><sg><nom>/অবদান<n><nt><nn><sg><obj>$ ^রাখতে/রাখ<vblex><inf>/রাখ<vblex><past><hbtl><p2><fam>$ 
^পারেন/পার<vblex><pres><smpl><p3><pol>/পার<vblex><pres><smpl><p2><pol>$^।/।<sent>$ ^২১/২১<num>$, ^২৫৫টি/২৫৫<num>$ ^ভুক্তির/*ভুক্তির$ 
^ওপর/ওপর<adv>/ওপর<n><mf><nn><sg><nom>/ওপর<n><mf><nn><sg><obj>$ ^কাজ/কাজ<n><nt><nn><sg><nom>/কাজ<n><nt><nn><sg><obj>$ 
^চলছে/চল<vblex><pres><cnt><impers>/চল<vblex><pres><cnt><p3><infml>$^।/।<sent>$^./.<sent>$[][
]

because if unescaped special characters appear in the stream, you will get a std::exception:

$ echo "This is a test ^500" | lt-proc bn.analyser.bin 
This is a test std::exception

(on a Mac, you'll typically see a 9Exception)

Generate

When generating, you basically input the analyses given by the analyser, but only one analysis per lexical unit. The general input format is

^lemma<tag><tag2><tag3>$ ^otherlemma<othertag><tag2>$

E.g. to generate a couple of the analyses given in the analysis example above:

$ echo '^বাংলা<adj><mf>$ ^।<sent>$ ^এই<det><dem>$' | lt-proc -g bn.generator.bin
বাংলা । এই

See also

• List of dictionaries
• Daemon – using an lttoolbox dictionary as a "server", without re-loading the dictionary for each request