Supertagging in Alto
Alto has experimental support for supertagging in the alto_supertagging fork.
To parse sentences with a TAG grammar using the supertagger, you need the following files:
- A Tensorflow model bundle representing the trained supertagger. This can be created from a trained model using the convert.py script in the nTagM package.
- The output vocabulary file that was created when training the supertagger.
- A file with word embeddings for the input words - in particular, glove.6B.200d.txt. The file may be gzipped to save a few hundred MB of disk space.
- The Chen grammar (d6.clean.f.str).
- A supertagger configuration file (e.g. config_dropbox.yaml), which specifies (among other things) the filenames of the output vocabulary and the word embeddings.