This repository contains an implementation of the Tensor-LSTM dependency parser used in arxiv link here. The parser as described in the paper can be run on any UNIX system with:
bash complete_test_and_train_pipeline.sh $TRAIN $VALID $TEST $LANGUAGE $FORMAT
$TRAIN, $TEST, and $VALID represent the location of the train, test, and validation files. $LANGUAGE represents the string name of the language within the ISO format, and $FORMAT should take either the value "mse" or "cross-entropy" and represents the loss function. Please note that we do not provide data or pre-trained embeddings.
We include a script and data for a single pilot test of the cross-entropy loss version of the parser using the English training and validation data in the Universal Dependencies treebank. This can be run as:
bash run_english_test.sh
To run on a GPU, replace the string "DEVICE=cpu" in the file complete_test_and_train_pipeline with "DEVICE=gpu"