Next: Annotation
Up: PEDANT
Previous: The storing of PEDANT
All text are first aligned on sentence level, using a program based
upon the algorithm putforth by Church and Gale. This algorithm is
built upon the assumption that a short sentence will be
translated with a short sentence and long with a long sentence. A more
detailed description of this algorithm is described in Gale and Church
GaleChurch93.
The different types of aligned pairs are:
1-0, 0-1, 1-1, 2-1, 1-2, 2-2, 3-1, 1-3. We have found the percentage
of correct alignments to be above 95%. Within the 5% of misstakes
that the program makes we mostly find examples of 1-0 or 0-1 alignment
which are easily to correct. All 3-1 or 1-3 alignments are made
manually as the program used today cannot automatically deside whether
it should be a case of 1-0 followed by a 2-1 or a 3-1.
Daniel Ridings
Sun Mar 31 09:05:43 METDST 1996