% Hey, emacs(1), this file is -*- coding: utf-8; -*- got it?

the files in this directory provide the input to the final black-box evaluation
of the LOGON demonstrator.  from the LOGON target corpus JHPSTG, there are the
following sub-sets that are used for quality assessments:
 
  jhk, psk, tgk --- known-vocabulary held-out data (word lists were available)
  jhu, psu, tgu --- unknown-vocabulary held-out data (completely unseen)
  jhe, pse, tge --- proportional sub-sets from the development corpus

for each segment, there are the following files:

  .fan    --- fan-out log from the final Fjell integration (15-jan-07);
  .all    --- post-processed fan-out output (from `summarize --output all');
  .oracle --- hand-annotated `best' outputs (thanks to victoria rosén)
  .smt    --- alternate translations from (baseline) SMT system

finally, there is a script to combine all LOGON outputs into a single file:

  ./align jhk.all jhk.oracle

produces output like the following:

  |< |Pakk sekken, reis til fjells og begynn å gå.|
  |@ |Fill your backpack, head for the mountains, and start walking.|
  |@ |Pack your rucksack, travel to the mountains, and start walking.|
  |@ |Pack your rucksack, travel to the mountains and start hiking.|
  |= |Wrap the back pack. travel in the mountains, and begin to go.|
  |> |Wrap the back pack. leave in the mountains, and begin to go.|
  |? |Wrap the back pack. leave to the mountains, and start to go.|

the |< (source sentence) and |@ (reference translations) lines have the exact
same interpretation as in the original fan-out log file.  the |= output is the 
one from the oracle annotations; the |> output is the one ranked highest by the
LOGON demonstrator, i.e. the one that was output at the top; finally, the |?
line shows the output with the highest BLEU score (in case there are ties, the
first such output is used, according to the LOGON ranking).

finally, files ending in `.logon' contain the outputs from:

  for i in jhe jhk jhu pse psk psu tge tgk tgu; do
    ./align ${i}.all ${i}.oracle > ${i}.logon;
  done

all files are encoded in UTF-8, with Un*x-style newlines.