It can't be pickled when it uses classifiers, I believe, because they
contain lambda (anonymous) functions. Do you happen to use
WordNetClassifier in your taxonomy? I think the best approach would be
to save the actual data that goes into the classifier (e.g., as a CSV
with term;type;value lines), and have a script that reads in the data,
and then appends the classifiers. Reading from text file is not always
slower than reading from a pickle.
If you happen to come up with code that does this, let me know then
I'll add it to the module. For example, there could be a
Taxonomy.read_csv() classmethod and a Taxonomy.write_csv() method.
Best,
Tom