MT/IE: Cross-lingual Open IE

Attention-based sequence-to-sequence model for cross-lingual open IE.

Summary

A tensorflow implementation of "MT/IE: Cross-lingual Open Information Extraction with Neural Sequence-to-Sequence Models" (EACL 2017) by Sheng Zhang, Kevin Duh, and Benjamin Van Durme.

Dependencies

• python 2.7
• tensorflow r0.12 or later

Train

We provide you a small toy dataset (10K) to play with. To start training on this dataset, simply run:

./run.sh

Evaluate

After training for a while, you can start evaluation by:

python -m mt_ie --do_decode=True

Note: multi-bleu.perl from mosesdecoder is included for your convenience.