conll16st-v34-focused-rnns

System implementation of the paper Discourse Sense Classification from Scratch using Focused RNNs (presented at CoNLL 2016 conference). Implemented in Python 2 using Numpy, Keras with Theano.

Note: Same implementation was used on English and Chinese datasets. It achieved new state-of-the-art on Chinese blind dataset.

Check out the conference paper and presentation at:

http://gw.tnode.com/deep-learning/conll2016-discourse-sense-classification-from-scratch-using-focused-rnns/

Abstract

The subtask of CoNLL 2016 Shared Task focuses on sense classification of multilingual shallow discourse relations. Existing systems rely heavily on external resources, hand-engineered features, patterns, and complex pipelines fine-tuned for the English language. In this paper we describe a different approach and system inspired by end-to-end training of deep neural networks. Its input consists of only sequences of tokens, which are processed by our novel focused RNNs layer, and followed by a dense neural network for classification. Neural networks implicitly learn latent features useful for discourse relation sense classification, make the approach almost language-agnostic and independent of prior linguistic knowledge. In the closed-track sense classification task our system achieved overall 0.5246 F1-measure on English blind dataset and achieved the new state-of-the-art of 0.7292 F1-measure on Chinese blind dataset.

Usage

Script for applying both trained models for English and Chinese that were used on TIRA system (check its source code):

# tira_run_{en|zh}.sh <dataset_dir> <output_dir>
$ ./v34/tira_run_en.sh ./data/conll16st-en-03-29-16-trial ./output
$ ./v34/tira_run_zh.sh ./data/conll16st-zh-01-08-2016-trial ./output

For training each individual model use:

# train.py <experiment_dir> <train_dir> <valid_dir> [--clean] [--config CONFIG]
$ ./v34/train.py ./models-v34-a ./data/conll16st-en-03-29-16-train ./data/conll16st-en-03-29-16-dev --config='{"filter_fn_name":"conn_eq_0"}'

Afterwards apply the trained model to an unseen dataset with:

# classifier.py <lang> <model_dir> <dataset_dir> <output_dir> [--config CONFIG]
$ ./v34/classifier.py en ./models-v34-a ./data/conll16st-en-03-29-16-test ./output --config='{"filter_fn_name":"conn_eq_0"}'

For evaluation use the official CoNLL 2016 Shared Task scorer:

$ ./conll16st_evaluation/tira_sup_eval.py ./data/conll16st-en-03-29-16-test ./output ./output

License

This code is licensed under the GNU Affero General Public License 3.0+ (AGPL-3.0+). Note that it is mandatory to make all modifications and complete source code publicly available to any user.

gw0/conll16st-v34-focused-rnns

conll16st-v34-focused-rnns

Abstract

Usage

License