Constituency parse probabilities bug

[etsipidi]

Thank for this library! I'm using the constituency parser and its probabilities, and I noticed that while the parse trees are correct, there seems to be a bug in the returned probabilities. The probabilities matrices are missing a necessary row and column each (the column being the rightmost column, which holds the 'S' start symbol) and because of that, the matrices are currently not matching the parse trees.

The issue is in this line:

parser/supar/models/const/crf/parser.py

Line 124 in bebdd35

batch.probs = [prob[:i-1, 1:i].cpu() for i, prob in zip(lens, s_span)]

and I'm proposing this fix:

            batch.probs = [prob[:i, 1:i+1].cpu() for i, prob in zip(lens, s_span)]