Translit plugin for pidgin

This is a pidgin plugin for

1. De-transliteraion of russian messages in ISO-9 translit.
2. Providing a keymap inside the pidging conversations.

Building and installing

In order to build a plugin, you have to have:

• Pidgin libraries libpurple.so
• GTK libraries and GLib-2.0
• Headers for pango, cairo, atk, gdk-pixbuf

After that run make and copy translit.so to ~/.purple/plugins. On the next start of pidgin, in plugins section you should see a plugin called Translit tools; enable it, and read help for /detrans, /nodetrans, /rus and /norus commands.

How does it work?

This plugin is using ISO-9 coding scheme for de-transliteration of the russian messages. For every conversation from the user included in the de-transliterarion list, the message is going to be de-coded using ISO-9 table and a set of custom exceptions. The plugin tries to leave as is html-tags, &xxxx; symbols and urls.

The main feature of the plugin is a set of custom exceptions, which were retrieved by analyzing a russian dictionary of hunspell. Consider the russian word ушла in a transliterated fashion -- ushla -- it can be decoded by simple per-letter decoding; so far, so good. Now let's look at сходить -- shodit'; any naiive decoder would decode it as шодить, which of course is wrong. In order to overcome this, one has to apply a set of rules to recognize such patterns. These rules can be found in the file called ru-special-words.def. INPUT (a, b) means that while de-transliteration a is going to be replaced by b. The longest-match principle is used while searching for the replacement candidate.

There are two more files involved in the decoding process:

1. ru-replacement.def which is just ISO-9 table
2. ru-capital-letters.def which is a table for replacing lowercase russian letters with capital.

As this table can be considerably large, we are using a trie data structure for fast matching. It works considerably fast -- 4 Mb can be detransliterated in 0.2 seconds on core i5.

Hacking

De-transliteration works outside the plugin context, and one can compile detrans-input binary by running make detrans-input which read a message from stdin and outputs decoded version on the stdout.

detrans-file binary, which can be built with make detrans-binary, accepts a file where each line is tab-separated correct russian word and transliterated version of it, and the binary will print out those pairs where de-transliteration wouldn't match the original. As an example of such a file see misc/ru-words-tr.txt.

Todo

• It would be nice to come-up with a minimal set of exceptions, to decrease a size of the binary.

• One can put some effort in making the tool aware of encodings, which may make it work a wee bit faster.

• The only word that currently fails is pasha. It is being decoded as пасха, not паша. Seems to be really non-trivial task to resolve it.

• I didn't get a chance to test it on windows.

• More testing :)

As always, pathces and suggestions would be highly appreciated.