SUMMARY

SUMMARY OF FAILFINDER EFFORTS DURING THE 4TH MT MARATHON

Lo�c : 
My work consisted in making a unifying interface to show several sentence level metrics and alignments in a easy-readable and fashion way.
I provide a prepare_data.pl script which role is to converts raw data to javascript which will later be used in the html file.

prepare_data.pl needs as input a basename (let's call it base). 
2 kinds of data are considered for now : 
 - header : stored in base.header. Contains the column names.
 - data : stored in base.data. Guess what's inside ;)
 

Jonathan:
- Worked on Error Categorizor and Search Path Analyzer in the java/ directory.
- Error Categorizor requires forced and vanilla n-best lists from cdec or Moses.
- Search Path Analyzer requires search path dumps from cdec or Moses.
- We provide converters for n-best lists, but the search path formats are currently patches
  and are not in the trunk distribution of these decoders
- Scripts for running cdec with forced decoding are checked in
- Still waiting on work in Moses search graph output and forced decoding to be completed.

TODO list : 
 - include Martin/David other people from Charles University's alignment results ont top of this page.
 - include realistics errors from Jon's work.