Deep Semantic Role Labeling (SRL)

SRL deep learning model is based on DB-LSTM which is described in this paper : End-to-end learning of semantic role labeling using recurrent neural networks

The system consists of code related to two models for following tasks :

Predicate identification

Model trained using RNN network with CRF as final output layer

SRL model

SRL Trained on db-lstm network with CRF as final output layer.

Evaluation

We use evaluation/conlleval.pl for model evaluation for both the models . This script is taken from Conll2000 shared task. (http://www.cnts.ua.ac.be/conll2000/ ) Rest of the code files for data manipulation and feeding the data for neural networks.

Code Layout :

Data folder :

embeddings :

• The word embeddings have been trained using fasttext by facebook . The word embedding are represented by two files : Vocab.txt : the vocabulary words in w2v

• wordVectors.txt : the vector for words mentioned in vocab.txt. Please note that it is row-row mapping from word to vector.

• The w2v model is trained with skipgram , 32 dimensions on QQ+SRL data

Predicate_identifier

Data for predicate identifier model in conll05 format. Please note that this format is generated by SemanticRoleLabeler Srl

Models folder :

• Final - The final models for predicate identification and srl model
• Parameters- This folder is for saving model parameters while training
• PI - Folder to save predicate identifier model while training
• Srl - Folder to save srl model parameters while training

Paddle Docker

THe deep learning code is built using Paddle framework by Baidu (https://github.com/PaddlePaddle/Paddle)

Paddle docker is required to run the code in Docker environment.

Working in the docker environment: Go to semantic-role-labeling-neural/ - Run following command + docker run -it -v $(pwd):/paddle paddlepaddle/paddle:0.10.0 /bin/bash

if the docker image is not already present on the system, it will be downloaded. Also please check for newer version of the docker image

Model related code :

Predicate Identifier Model

Training the model :

pi_train.py

Running inference code from trained model :

Pi_infer.py

SRL model

Training the model

Srl_train.py

Running inference code from trained model :

Srl_inference.py