Punctuation and Capitalization Restoration using Bi-LSTM Network

Ashish Bansal

Abstract: In this paper we present Bi-Directional Long short-term memory (BLSTM) network with linear Conditional Random Field (CRF) to restore punctuation and capitalization in an unsegmented speech transcript. Our approach will restore both punctuation and capitalization using a single model. This is purely lexical based approach with pre trained glove embeddings as an input. The task was treated as a sequence tagging problem where the input is sequence of un- punctuated and un-capitalized words, and the output is a corresponding sequence of punctuated and capitalized words. We demonstrated the accuracy of proposed model are competitive with the state-of-the-art models and can do both the tasks in a single model.

Keywords: Punctuation prediction, Capitalization restoration, Neural network, true-casing, sentence segmentation

How to Cite?: Ashish Bansal, "Punctuation and Capitalization Restoration using Bi-LSTM Network", Volume 8 Issue 4, April 2019, International Journal of Science and Research (IJSR), Pages: 2020-2025, https://www.ijsr.net/getabstract.php?paperid=SR24809041216, DOI: https://dx.doi.org/10.21275/SR24809041216

Download Citation: APA | MLA | BibTeX | EndNote | RefMan