Difference between revisions of "Indirect contribution guide"

From Apertium
Jump to navigation Jump to search
Line 194: Line 194:
 
However, the difficult part is not getting it into this XML format, but getting each pair of verbs, and the important information for each pair (here: aspect), and making sure it is <i>consistent</i> and machine-readable.
 
However, the difficult part is not getting it into this XML format, but getting each pair of verbs, and the important information for each pair (here: aspect), and making sure it is <i>consistent</i> and machine-readable.
   
===Other===
+
===Other ways to help===
   
;Other ways to help.
 
 
* If you find mistakes in translation or pending test.
 
* If you find mistakes in translation or pending test.
 
** [[Contact|Let us know!]].
 
** [[Contact|Let us know!]].
 
** Correct them.
 
** Correct them.
 
*Ask on IRC or the mailing list (or, if you have one, your mentor) if there is another way to help.
 
*Ask on IRC or the mailing list (or, if you have one, your mentor) if there is another way to help.
  +
  +
===Other projects===
  +
  +
If you want to contribute to a free/open-source language resources project, but Apertium doesn't quite fit your intentions, here are some other resources:
  +
  +
* [http://omegawiki.org OmegaWiki] (Creative Commons BY & GNU FDL)
  +
* [http://tatoeba.org Tatoeba] (Creative Commons BY-SA)
   
 
[[Category:Documentation]]
 
[[Category:Documentation]]

Revision as of 17:00, 7 March 2015

Many people come to us with a question like "I'm not a programmer/linguist/whatever. Is there any way I can contribute?". This document is intended to show how you can make an "indirect" contribution, by documenting language resources, helping us to build bilingual test sets, translating, promoting, etc.

About This Tutorial

This tutorial will teach you:

  • How to create contrastive analyses.
  • How to catalog resources.
  • How to convert dictionaries.
  • How to translate.
  • How to help "Apertium" in other ways.

Basics

When in doubt, ask!

If you are participating as part of a programme such as Google Summer of Code, or Google Code-In, ask your mentor. Otherwise, ask for help on the IRC channel, or on the mailing list. Use the talk pages on the wiki -- leave your questions there, so you don't need to remember them later!

Create contrastive analyses

A 'contrastive analysis' is a set of example sentences which show the differences and similarities between a pair of languages. In a sense, it's a 'feature corpus' which we can use to develop and test rule hypotheses: if we see that the pattern noun + adjective becomes adjective + noun, then we have a good basis for building a rule. Think of it as 'raw input to a linguist': when we have a good enough idea of what a pair of languages look like, we later use these analyses to build translation rules.

One thing to note is that, when we see that something happens 9/10 times, or 8/10 times, etc., then we need to expand that exceptional part of the analysis, to get a better idea of what's happening: is it a certain class of words, or just a pure exception?

What you must do.

Your task is to make a set of test sentences in the first language, and translate them to the other. A translation in a third language may be useful in enlisting help, but is not required.

A sample sentence in wiki markup looks like this:

* {{test|First language abbreviation|First language.|Second language.|Other translation.}}

Examples:

* {{test|ru|Чашка большая.|Чашата е голямата.|The cup is big.}}

* {{test|el|Τι γίνεσαι?|как си?|How are you?}}

* {{test|bg|Вера се оглежда в огледалото.|Вера смотрит на себя в зеркало.|Vera is looking at herself in the mirror.}}

The following is a suggested list of features to provide coverage for. All language pairs have slightly different needs, but this list should provide a good general guideline.

  1. Simple syntax
    • Copula
    • Reported speech
    • Clitic placement
  2. Pronouns
    • Personal
    • Demonstrative
    • Relative
    • Possesive
    • Reflexive
    • Interrogative
  3. Nouns
    • General
    • Indefinite and definite forms
    • 1 Noun phrases
    • Indefinite (a, some)
    • Definite (the)
    • Demonstrative (this, that)
    • Quantified (a few, no, all)
  4. Numerals
    • Cardinal
    • Ordinal
  5. Adjective
    • Comparative
    • Superlative
  6. Adverbs
  7. Verbs
    • To be
    • General
    • Indicative mood
      • Present tense
      • Imperfect tense
      • Aorist tense
      • Perfect tense
      • Pluperfect tense
      • Future tense
    • Conditional mood
    • Imperative mood
  8. Questions

If you want you can also add

  1. Interjections
  2. Punctuation marks

You can find some examples here: Bulgarian and Russian, Bulgarian and Greek.

How to do it?

This is quite an easy task if you know both languages. However, if you only know one language well, concentrate on translating to that language. We can always find help with the other direction later, and it helps to know what we need to find help with.

That said, if you know people who can help, ask them to help you!

Some tips:

  1. Ask your friends if they know the language you don't know.
    • If they do, ask them to help you.
    • If you know more than one person who speaks the language, you will finish the job faster and better
      • If your friend is busy, queue your questions and write them later.
  2. Get a textbook for the language that you don't know.
    • For some languages this is not a good idea.
      • They are hard to find.
      • They are expensive.
    • Look on the Internet for textbooks.
      • There are many textbooks and it is a good idea to look at more than one.

Catalogue resources

The task is to catalogue as many of the available linguistic resources as possible. Dictionaries, grammars, accademic papers, etc., are all resources that we can use to get a general idea of how the language works, which we can use to get an idea of how to translate from that language.

Lingistic resouces.
  • wordlists.
  • grammatical descriptions.
  • wordlists.
  • dictionaries.
  • spellcheckers.
  • papers.
  • corpora.
  • and more...
How to do it?
  • Use Google of course.
  • And other search engines, such as:
    • ScienceDirect.
    • JSTOR.
    • Duck Duck Go.

See also Wikipedia's article about search engines.

Translate

The task is to translate text (articles on this wiki, for example) to another language. You should speak the target language (the language you are translating to) natively or near natively to translate -- if you can't, it's best to leave the translation to someone who does.

Some tips.
  • If you aren't sure about something, ask a native speaker or, if possible, a specialist.
  • You must not change the meaning of what you translating.
    • Follow the meaning, not the words.
  • You must pay attention to language nuances.
  • You must pay attention to the tenses.
  • Follow the original style:
    • If the text style is wordy, colloquial, funny... follow it.
    • Pay attention to the punctuation.
Pay attention!
  • Don't translate if you don't know the language!
  • Don't take a translation task if you know that you can't do it!
  • Don't take a translation task if you don't have the time to do it!

Post-edit

Related to the above: post-edit an Apertium translation, and provide us with the result. Having a set of corrections to refer to can help us to refine the translator. We are particularly interested in Open Content text (text under Free licences, such as the GNU FDL or CC BY/BY-SA), which we can freely redistribute, and translations that are under a similar licence.

Even better, if you have used a Translation Memory (such as OmegaT), providing us with the TMX can help us in a variety of ways.

Convert dictionaries

There are many dictionaries available under free licences, that we would like to have converted to Apertium's format. However, it's not always as simple as taking words: Apertium (usually) allows only a single translation option per word; there are also some tagging differences that need to be present in Apertium's lexicon, to assist in grammatical operations.

Tagging

If the word list doesn't have part-of-speech information, you will need to add it.

Gender

If one or both of the languages have grammatical gender (male, female, neuter), the lexicon needs to have information about the gender when it is different for a set of words. It's not strictly necessary in other cases, but it is useful to have in the dictionary for other reasons, so we encourage you to add grammatical gender, always!

Aspect

Similarly, if you are translating verbs in a language with aspect pairs (i.e., Slavic languages), tag the aspect (even though it's not usually strictly necessary between languages with similar concepts of aspect):

  • perf = perfective.
  • imperf = imperfective.
How to do it?
  • If you have friends who speak the language, ask for their help.
  • Look in a dictionary.
    • You can buy a dictionary, they are not very expensive.
    • You can search for a free dictionary on the web.
    • It's a good idea to do both.
  • Double check: don't rely on a single source.


Examples:

Bulgarian→Russian

Say you've found a Bulgarian-Russion dictionary, and it says that the imperfective verb обяснява translates into the imperfective обяснять, while the perfective напише translates into the perfective написать, the converted version of this should look like:

<e><p><l>обяснява<s n="vblex"/><s n="imperf"/><s n="tv"/></l><r>обяснять<s n="vblex"/><s n="imperf"/></r></p></e>  
<e><p><l>напише<s n="vblex"/><s n="perf"/><s n="tv"/></l><r>написать<s n="vblex"/><s n="perf"/></r></p></e> 

However, the difficult part is not getting it into this XML format, but getting each pair of verbs, and the important information for each pair (here: aspect), and making sure it is consistent and machine-readable.

Other ways to help

  • If you find mistakes in translation or pending test.
  • Ask on IRC or the mailing list (or, if you have one, your mentor) if there is another way to help.

Other projects

If you want to contribute to a free/open-source language resources project, but Apertium doesn't quite fit your intentions, here are some other resources: