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.

Task ideas for Google Code-in

From Apertium
(Difference between revisions)
Jump to: navigation, search
Line 1: Line 1:
This is the task ideas page for Google Code-in, here you can find ideas on interesting tasks that will improve your knowledge of Apertium and help you get into the world of open-source development.
+
This is the task ideas page for Google Code-in ([[http://code.google.com/gci]]), here you can find ideas on interesting tasks that will improve your knowledge of Apertium and help you get into the world of open-source development.
   
 
The people column lists people who you should get in contact with to request further information. The time column gives the minimum estimated amount of time that should be spent on the task. '''It does not include time taken to install / set up apertium'''.
 
The people column lists people who you should get in contact with to request further information. The time column gives the minimum estimated amount of time that should be spent on the task. '''It does not include time taken to install / set up apertium'''.

Revision as of 12:45, 27 October 2010

This is the task ideas page for Google Code-in ([[1]]), here you can find ideas on interesting tasks that will improve your knowledge of Apertium and help you get into the world of open-source development.

The people column lists people who you should get in contact with to request further information. The time column gives the minimum estimated amount of time that should be spent on the task. It does not include time taken to install / set up apertium.

Task list

Area Difficulty Title Description Time
(hours)
People
code 2. Medium Cross a language pair: Occitan-French Using apertium-crossdics, build a dictionary for Occitan-French from Occitan-Catalan and Catalan-French, and clean up the result. 4—10 Francis Tyers
code 2. Medium Cross a language pair: Aragonese-Catalan Using apertium-crossdics, build a dictionary for Aragonese-Catalan from Aragonese-Spanish and Spanish-Catalan, and clean up the result. 4—10 Jimregan
code 2. Medium Cross a language pair: Romanian-French Using apertium-crossdics, build a dictionary for Romanian-French from Romanian-Spanish and Spanish-French, and clean up the result. 4—10 Francis Tyers
code 2. Medium Cross a language pair: Romanian-Italian Using apertium-crossdics, build a dictionary for Romanian-Italian from Romanian-Spanish and Spanish-Italian, and clean up the result. 4—10 Francis Tyers
code 1. Hard Convert existing resource: Urdu morphological analyser Take Muhammad Humayoun's Urdu Morphology and convert to lttoolbox format. 8—10 Francis Tyers
code 1. Hard Convert existing resource: Punjabi morphological analyser Take Muhammad Humayoun's Punjabi Morphology and convert to lttoolbox format. 8—10 Francis Tyers
code 1. Hard Convert existing resource: Kurdish morphological analyser Take the Alexina Kurdish Morphology and convert to lttoolbox format. 8—10 Francis Tyers
outreach 3. Easy Apertium on Macedonian Wikipedia Bulgarian WP has 107,355 articles, Macedonian WP has 42,112, less than half as many. Translate some articles from Bulgarian Wikipedia to Macedonian Wikipedia using Apertium, and then postedit them. Explain to the local Wikipedia community what you are doing beforehand. 1—4 Francis Tyers
outreach 3. Easy Apertium on Occitan Wikipedia Catalan WP has 290,059 articles, Occitan WP has 22,579, less than a tenth as many. Translate some articles from Catalan Wikipedia to Occitan Wikipedia using Apertium, and then postedit them. Explain to the local Wikipedia community what you are doing beforehand. 1—4 Francis Tyers
outreach 3. Easy Apertium on Asturian Wikipedia Spanish WP has 663,567 articles, Asturian WP has 13,869, almost a fiftieth as few. Translate some articles from Spanish Wikipedia to Asturian Wikipedia using Apertium, and then postedit them. Explain to the local Wikipedia community what you are doing beforehand. 1—4 Francis Tyers
quality 3. Easy Thorough checkup of bn-en morphological analyser While the current bn-en morphological analyser has a pretty good coverage, it should have been higher. Part of the reason is that a lot of verbs have one/two slight different surface forms that differ from the regular ones and the analyser misses them. Using lt-expand it's possible to generate all forms of the verbs, then manually check these and using another script (already in the pair) rebuild the analyser file. This checking will require a native speaker/expert on Bengali language Abu Zaher
code 2. Medium NSIS script Write an NSIS script to install the Cygwin version of Apertium on Windows. Jimregan
research 2. Medium Contrastive analysis: Macedonian and Albanian Create a set of test sentences (see various pages of 'Pending tests' and 'Regression tests' on the Wiki) for translation between Macedonian and Albanian. The tests should cover as many features of the languages as possible. Some of the examples might be able to be found in a grammar, others might need to be invented. This will not involve programming, only grammatical analysis. 4—6 Francis Tyers
research 2. Medium Contrastive analysis: Kurdish and Persian Create a set of test sentences (see various pages of 'Pending tests' and 'Regression tests' on the Wiki) for translation between Kurdish and Persian. The tests should cover as many features of the languages as possible. Some of the examples might be able to be found in a grammar, others might need to be invented. This will not involve programming, only grammatical analysis. 4—6 Francis Tyers
research 2. Medium Contrastive analysis: Hindi and Urdu Create a set of test sentences (see various pages of 'Pending tests' and 'Regression tests' on the Wiki) for translation between Hindu and Urdu. The tests should cover as many features of the languages as possible. Some of the examples might be able to be found in a grammar, others might need to be invented. This will not involve programming, only grammatical analysis. 4—6 Francis Tyers
research 2. Medium Contrastive analysis: Finnish and Estonian Create a set of test sentences (see various pages of 'Pending tests' and 'Regression tests' on the Wiki) for translation between Finnish and Estonian. The tests should cover as many features of the languages as possible. Some of the examples might be able to be found in a grammar, others might need to be invented. This will not involve programming, only grammatical analysis. 4—6 Francis Tyers
research 3. Easy Catalogue resources: Aromanian Catalogue all the available resources (grammatical descriptions, wordlists, dictionaries, spellcheckers, papers, corpora, etc.) for Aromanian, along with the licences they are under. Francis Tyers
translation 2. Medium Translate the HOWTO: Polish Translate the new language pair HOWTO into Polish, and go through it for a new pair of languages. When finished, upload to the Incubator. 2—3 Jimregan
translation 2. Medium Translate the HOWTO: Italian Translate the new language pair HOWTO into Italian, and go through it for a new pair of languages. When finished, upload to the Incubator. 2—3 Deadbeef
translation 2. Medium Translate the HOWTO: Norwegian Translate the new language pair HOWTO into Nynorsk or Bokmål, and go through it for a new pair of languages. When finished, upload to the Incubator. 2—3 Unhammer
quality 3. Easy Quality evaluation: Spanish and French Perform a human post-edition evaluation of the Spanish and French language pair. This will involve taking some free text (e.g. from Wikipedia or Wikinews), running it through the translator and then altering the output to be correct. Then using apertium-eval-translator to calculate the Word Error Rate. The minimum amount of text should be 4,000 words. 4—8 Francis Tyers
quality 3. Easy Quality evaluation: Spanish and Occitan Perform a human post-edition evaluation of the Spanish and Occitan language pair. This will involve taking some free text (e.g. from Wikipedia or Wikinews), running it through the translator and then altering the output to be correct. Then using apertium-eval-translator to calculate the Word Error Rate. The minimum amount of text should be 4,000 words. 4—8 Francis Tyers
quality 3. Easy Quality evaluation: Spanish and Asturian Perform a human post-edition evaluation of the Spanish and Asturian language pair. This will involve taking some free text (e.g. from Wikipedia or Wikinews), running it through the translator and then altering the output to be correct. Then using apertium-eval-translator to calculate the Word Error Rate. The minimum amount of text should be 4,000 words. 4—8 Francis Tyers
user interface 1. Hard Design a user-friendly interface for Apertium Apertium does not currently have a friendly user interface for translators. Look at other translation software on the market, and sketch out some ideas for how to design a user interface. This will not require programming, but could, for example involve using Glade to demonstrate the ideas. Jimregan
training 3. Easy Step-by-step "become a developer" guide Write a simple step-by-step guide (on the wiki) for pre-university students (of varying levels of computer literacy) to install a development version of Apertium and make a single change in a language pair. This should include everything, from checking out with SVN to requesting committer access on SourceForge. Document everything you do! 2—3 Mikel L. Forcada
training 3. Easy Step-by-step "constraint grammar" guide Write a simple step-by-step guide (on the wiki) for pre-university students (of varying levels of computer literacy) to install Constraint Grammar and fix 5 disambiguation problems in a single sentence, then committing to the incubator. This should include everything, from checking out with SVN to requesting committer access on SourceForge. Document everything you do! 2—3 Unhammer
quality 3. Easy Release freshness Go through all the released pairs and note down their date of last release and how many dictionary entries and rules they have. Then go to SVN and look at the module for the released pair and find out how many dictionary entries and rules it has. Put this into a spreadsheet and email the mailing list. Why? Our release cycle is very slow, and often we get pairs in trunk which have substantial improvements but have not been released. 2—4 Francis Tyers

Make more specific

Area Difficulty Title Description Time
(hours)
People
research 3. Easy Create manually tagged corpora Fix tagging errors in a piece of analysed text, for use in tagger training. This will involve taking some free text (see above), running it through the analyser and tagger, and replacing incorrect analyses with the correct one. Jimregan
quality 1. Hard Improve a language pair Find some faults in an existing language pair and fix them. In particular minor→major pairs, e.g. Welsh-English, Basque-Spanish, Breton-French. Francis Tyers, Jimregan, Mikel L. Forcada
documentation 2. Medium Document undocumented features Find a feature that can't be found in the existing documentation (e.g. cascaded interchunk transfer), and write about it. Mikel L. Forcada
documentation 2. Medium Create a dictionary crossing guide Extract a tutorial guide to using crossdics from jimregan's brain. 2—3 Jimregan
outreach 2. Medium Writing a quick guide on 'What Apertium can and cannot do to help you with your homework'. Students around the world use Apertium (and other MT systems) to do their second-language homework. The documents would summarize the do's and don'ts, and could even elaborate on how students using Apertium for their homework could discover ways in which Apertium could be improved. Mikel L. Forcada
quality 1. Hard Fix some tagger errors in Swedish->Danish apertium-sv-da could be improved with a Constraint Grammar. Find 10 sentences that get wrong translations due to tagging, and write CG rules to fix them. The student should have good knowledge of Swedish, or at least some Scandinavian language. Unhammer
Personal tools