Treebank expansion

[sylvainkahane]

We would like to join to a treebank a file with several pieces of information which can be used to expand the treebank when it is uploaded to grew-match.

• metadata concerning the corpus itself : metadata associated to the corpus itself or to part of the corpus (genre, etc.), metadata associated with speakers (for spoken corpora)
• metadata concerning some features, especially features with continuous values. For each feature feat with continuous values, we need min(feat), max(feat) and a relevant gap(feat) (for instance for vowel lengths, min=0ms, max=200ms and gap=5ms, which is the supposed threshold for a difference to be perceptible). It allows us to transform feat into a set of binary features: For each value x = min(feat) + k.gap(feat), with x ≤ max(feat), we generate a feature feat[x] such that: if feat > x, feat[x] = 1, else 0.
• some other expansion do not need an additional file, such as the span and the flux (see other issues).

[sylvainkahane]

Bruno proposes to add the possibility to write a Grew grammar expanding or modifying the treebank. Possible applications:

• Generate a corrected version of the treebank by replacing any value of feature Feat by the value of CorrectFeat, when this feature exists.
• Generate a pure text-based version of the treebank, where amalgams such as French du is not analyzed as de+le. This is useful to train a parser.
• Add the domination relation, for a given set of dependency relations (see #26).

The new treebank will be temporarily available in Grew-match, could be requested and exported.