HRG Parser

Recommended Environment

• Linux
• Java 8
• Python 3.6
• GCC 7.0

Data Preparation

You need get DeepBank1.1 data first.

Split the punctuation from the lexicon

Compile the java project "pseud" and run: java jigsaw.treebank.RedwoodsTreeReader <"profile" dir of deepbank1.1>

The outputs are in out directory. You should manually split them into train, dev and test set into directory java_out_train, java_out_dev, java_out_test respectively, and put them into a common directory main_dir_base.

install python dependency

Run pip3 install -r requirements.txt

Extract grammar

Modify deepbank_export_path and main_dir_base in extract_sync_grammar.py and run it. The output trees, derivation and grammar will be in deepbank-preprocessed directory.

Train a POSTagger

Modify paths in scripts/train_leaftag.py and run it.

Train a Phrase Structure Parser

Modify paths in scripts/train_span_lite.py and run it.

Predict with Trained Phrase Structure Parser

Run main.py span predict and follow the instruction.

Train a HRG Parser

Modify paths in scripts/train_udef_lite.py and run it.

Predict with Trained HRG Parser

Run main.py hrg predict and follow the instruction.