Difference between revisions of "User:Fpetkovski"

From Apertium
Jump to navigation Jump to search
Line 1: Line 1:
 
[[GSOC 2013 Application - Improving the lexical selection module]]
 
[[GSOC 2013 Application - Improving the lexical selection module]]
[[GSOC 2013 Report]]
 
==Documentation / HOWTO==
 
* [[Corpus based preposition selection - HOWTO]]
 
* [[Building a pseudo-parallel corpus]]
 
   
 
[[GSOC 2012 Report]]
==Reports==
 
* [[Lexical feature transfer - First report]]
 
 
* [[Lexical feature transfer - Second report]]
 
 
==TODO==
 
 
* '''Try generating corpus from monolingual SL corpus:'''
 
** Оваа лабавост на регулативите се одразува врз третманот на уапсените корисници на дрога.
 
<s>*** Run through lexical transfer <code>mk-en-biltrans</code>
 
*** Run through <code>apertium-lex-tools/scripts/biltrans-to-multitrans.py</code>
 
*** Run through the rest of the pipeline from <code>apertium-transfer -b</code> onwards
 
*** Run through <code>apertium-lex-learner/irstlm-ranker</code></s>
 
** This will give:
 
<s>*** SL:TL selection possibilities
 
*** probabilities from the TL language model for each selection</s>
 
** Select a subset for training where one translation has a substantially higher proportion of the probability mass than the rest.
 
** Look at finding out how to work out what "substantially" should be.
 
 
* '''Improve current method:'''
 
<s>** Split test corpus in two (dev, test)
 
*** Rerun the experiments and check with test corpus
 
*** Look at dev corpus to see what kind of patterns there are in lines that aren't getting matched
 
** Look at combining the 1-feature with the 2-feature model as backoff.
 
</s>
 
* '''Evaluation'''
 
** Try pair bootstrap resampling between best system and default translation for both WER and BLEU.
 
 
* ''' Check the bidix entries that were added automatically'''
 
 
[[Category:Users|Fpetkovski]]
 

Revision as of 12:19, 11 April 2013