Apertium has moved from SourceForge to GitHub.
If you have any questions, please come and talk to us on #apertium on irc.freenode.net or contact the GitHub migration team.

Bosnian-Croatian-Montenegrin-Serbian and Slovenian

From Apertium
Jump to: navigation, search

Description about this project will be added soon.

Contents

[edit] Source

https://apertium.svn.sourceforge.net/svnroot/apertium/trunk/apertium-hbs-slv

[edit] Work plan

This is a draft workplan for development of the Bosnian-Croatian-Montenegrin-Serbian and Slovenian translator.

  • Trimmed coverage means the coverage of both of the morphological analysers after being trimmed according to the bilingual dictionary of the pair, that is, only containing stems which are also in the bilingual dictionary.
  • Testvoc for a category means that the category is testvoc clean, in both translation directions.
  • Evaluation is taking n words and performing an evaluation for post-edition word error rate (WER). The output for those n words should be clean.
Week Dates Trimmed coverage Reached Testvoc Evaluation Notes
0 23/04—21/05 45% <abbr> <ij> 500 words Preliminary evaluation. Translate the story in both directions with total coverage and without diagnostics. Get a baseline WER. Work on disambiguation, the morphological ambiguities in the story should be resolved.
1 21/05—27/05 50% <cnjcoo> <cnjadv> <cnjsub> <pr> - Conjunctions and prepositions should be clean. Verbs should be tagged for transitivity.
2 28/05—03/06 53% <adv> -
3 04/06—10/06 59% <num> 200 words Work on disambiguation. Numerals should be clean.
4 11/06—17/06 63% <prn> <det> -
5 18/06—24/06 68% <n> -
6 25/06—01/07 70% <adj> 500 words Midterm evaluation. Work on disambiguation.
7 02/07—08/07 73% - -
8 09/07—15/07 75% - -
9 16/07—22/07 77% - 200 words Work on disambiguation.
10 23/07—29/07 80% <vblex> -
11 30/07—05/08 - - - Closed categories clean.
12 06/08—12/08 - - - Open categories clean.
13 13/08—19/08 - - - Disambiguation and transfer.
14 20/08—24/08 - - - Final evaluation. Tidying up, releasing

[edit] Improving coverage - A two week report

The work on improving the coverage for the slovenian -> bosnian-croatian-montenegrin-serbian language pair should be done 4 hours per day, for two weeks straight.
Day 0 denotes the current state of the language pair, and each subsequent entry will denote the coverage and testvoc reached for that day.

The corpora used for this task can be found here: http://www.statmt.org/europarl/v7/sl-en.tgz

Coverage was calculated using the following script


echo total;
cat europarl-v7.sl-en.sl | apertium -d ~/Apertium/apertium-sh-sl/ sl-sh-morph | sed 's/\$[^\^]*\^/$\n^/g' | wc -l
# This expression evaluates to 14340199

echo unknown;
cat europarl-v7.sl-en.sl | apertium -d ~/Apertium/apertium-sh-sl/ sl-sh-morph | sed 's/\$[^\^]*\^/$\n^/g' | grep '*' | wc -l

coverage = 1 - unknown / total


Dat Dates Coverage reached Coverage reached (2) Target Notes
0 20-02-2013 87.12% - -
1 01-03-2013 88.39% - [87.56, 88.24]
2 02-03-2013 90.05% - [87.78, 88.80]
3 03-03-2013 91.53% - [88.00, 89.36]
4 04-03-2013 92.36% - [88.22, 89.93]
5 05-03-2013 92.95% - [88.44, 90.49]
6 06-03-2013 93.55% - [88.66, 91.05]
7 07-03-2013 94.11% - [88.88, 91.61]
8 08-03-2013 94.54% - [89.10, 92.17]
9 09-03-2013 94.95% 87.52% [89.32, 92.74]
10 10-03-2013 95.11% 91.46% [89.54, 93.30]
11 11-03.2013 95.26% [89.76, 93.86]
12 13-03-2013 95.41% 93.23% [89.98, 94.42]
13 15-03-2013 95.50% 93.57% [90.20, 94.99]
14 [90.00, 95.00]

The work on improving the coverage for the slovenian -> bosnian-croatian-montenegrin-serbian language pair should be done 2 hours per day, for two weeks straight.
The two tasks do not have to be run in parallel.


Day Dates Testvoc reached Notes
0 26-03-2013 99.76% Regional variants added, some translation mistakes ironed out, a cpl of transfer rules
1 28-03-2013 99.81% Nouns, proper nouns and number categories clean
2 02-04-2013 99.87% Verbs clean
3 03-04-2013 99.90% Adjectives clean sl->bhs
4 04-04-2013 99.92% Adverbs almost clean
5 05-04-2013 99.93% Adverbs bhs->sl left
6 06-04-2013 100% Testvoc clean
7 09-04-2013 100% Strange generation errors cleaned
8 14-04-2013 100% Generation duplicates cleaned
9 15-04-2013 100% 20 transfer rules added
10 16-04-2013 100% 21 transfer rules added
11 17-04-2013 100% particles that testvoc didn't catch clean
12 18-04-2013 100% numerals uncaught by testvoc cleaned
13
14

[edit] See also

Personal tools