jamesxy7 / bilstm-debias

This is a joint RNN model, which jointly models clean annotated data and noise projected data.

Welcome to BiLSTM+Debias

---

Introduction

This source code is the basis of the following paper:

Learning when to trust distant supervision: An application to low-resource POS tagging using cross-lingual projection, CoNLL 2016

Building

It's developed on clab/cnn toolkit.

• Install clab/cnn following clab/cnn.
• Add the source code to folder cnn/examples and add bilstm-dn to CMakeLists.txt.
• Make again.

Data format

The format of input data is as follows:

Tok_1 Tok_2 ||| Tag_1 Tag_2
Tok_1 Tok_2 Tok_3 ||| Tag_1 Tag_2 Tag_3
...

How to run

./bilstm-dn gold_data_file projected_data_file dev_data_file test_data_file max_epochs