ganeshjawahar / ELMoLex

Final parser submitted by ParisNLP team for CoNLL 2018 Shared Task on Multilingual Parsing

ELMoLex - Connecting ELMo and Lexicon features for Dependency Parsing

Our CoNLL-18 submission system for Parsing which uses Deep Biaffine Parser as its backbone.

Features

• ELMo features
• UDLexicons features
• LSTM pre-training through NLM (NLM Init. feature)
• Read/Write directly in CoNLL-U
• Integration with CoNLL-18 evaluation script

Quick Start

• Install the dependencies (ensure conda is working before running the setup):

bash setup.sh

• Check if everything works through a dry run (should take less than 30 minutes):

bash sample_run/run.sh check

• Perform a complete run of the parser on en_lines treebank:

bash sample_run/run.sh complete

Finale Run

Instructions to start the training/testing for the finale run can be found here.

Resources

• Raw Treebanks
• UDLexicons
• FastText Word Embeddings

Acknowledgements

• NeuroNLP2 from XuezheMax
• Chu-Liu-Edmonds from bastings

License

ELMoLex is GPL-licensed.