Certain problematic time expression produce a large number of "bad" parses, and this affects the training. Using a limit on max_stack_depth mitigates the problem. Replacing DummyScorer with a NaiveBayesScorer could further improve things. This option will be explored in the near future.
Certain problematic time expression produce a large number of "bad" parses, and this affects the training. Using a limit on max_stack_depth mitigates the problem. Replacing DummyScorer with a NaiveBayesScorer could further improve things. This option will be explored in the near future.