BiLSTM-Based Semantic Role Labeling

Developed a deep learning model for Argument Prediction (object and subject) of a sentence given a predicate on Nombank Dataset; based on dependency mappings and POS tags; used Tensorflow library for training on cloud GPUs. Reviewed literature and devised an approach using BiLSTM for classification.

The cleaned version data of % class from the nombank dataset is used here. However, this project can be used for any other class of nombank as well. The dataset %_nombank.clean.train is shared. Token Embeddings, POS embeddings and BIO embeddings of each sentence are concatenated. This concatenated input is given to BiLSTM model. The model outputs 2 tags Arg1 or Not Arg1. Made in collaboration with Srushti Pawar and Simran Makariye.