Component Flags
NLP components in ClearNLP behave differently when different flags are set.
All component flags are defined in CFlag.
The component is used for collecting lexicons before training. For example when this flag is set, AbstractPOSTagger collects lexicons such as ambiguity classes using POSCollector.
This flag is not used for components that do not need to collect any lexicon before training (e.g., CommonTagger).
The component is used for generating training instances. When this flag is set, all features are extracted by using the oracle.
The component is used for generating training instances through bootstrapping.
When this flag is set, the component requires a statistical model, trained from previous training or bootstrapping, and extracts features predicated by this model instead of the oracle.
For instance, AbstractPOSTagger extracts features from part-of-peech tags automatically generated by the statistical model. See the following paper for more details about bootstrapping.
- Getting the Most out of Transition-based Dependency Parsing, Jinho D. Choi, Martha Palmer, ACL'11, 687-692, 2011.
The component is used for evaluating the statistical model. The gold-standard labels must be provided for this flag.
The component is used for decoding where no gold-standard labels are provided.
