Reference (.ref) file specifications

  Each files of type "*.ref" contains a reference list of occurrences
for a key word, similar to the reference files used with the ROAD RALLY
corpus, which can be used by evaluation or training software.  They
have been produced automatically from the "*.mrk" form of transcription
files, in this format:

  - Any line beginning with ";" is an ignorable comment line.
    (Comment lines have been generated to make the files almost
     self-documenting.)

  - A non-comment line will contain these seven fields of data:
     1. the keyword base form;
     2. a bit-vector numeric code indicating features of this occurrence:

             CODE bits: (0=LSB, on the right)
               0 - variant of the basic keyword.
               1 - overlapped word (simultaneous speech).
               2 - mispronounced word.
               3 - transcriber was unsure of word.
     
     3. the numeric conversation i.d.;
     4. the speaker i.d.;
     5. T1, the time in seconds of the beginning of the occurrence;
     6. DUR, the duration in seconds of the occurrence;
     7. the keyword variant form (may be same as base form).


   These files are derived from the marked transcript ("*.mrk") files.
   They therefore contain the errors, for keywords, that the marked
   transcript files contain.  See "mrk_spec.doc".  NIST will release
   in June 1992 official corrected reference files for the keywords.