Sunwoo Ha

Date of Award

2019

Document Type

Thesis

Degree Name

Bachelors

Department

Natural Sciences

First Advisor

Lepinski, Matthew

Area of Concentration

Computer Science

Abstract

This thesis explores the topic of text generation using natural language generation and processing methods which allows machines to automatically create text. Over the past few years, there has been an increase in parodies of machine translation online. These posts have risen because it is difficult for machines to create sensible text which also has correct syntax. These machine generated text are unintentionally hilarious to humans and became a source of entertainment on social media. Generative Adversarial Networks (GANs) are gaining popularity due to their success in image generation. Since the network has such great success in creating fake images, it is tempting to see how well it will fare to the task of generating text. Other than the GAN, this thesis will also look at two other models which are more traditionally used for text generation: Markov chains and predictive text. The three models will create their version of a script for popular TV show, Friends. Although automatic evaluation techniques are improving for text generation, they are still widely debated in the academic community. I conclude with a proposal for human-based evaluations on the text.

Download

Share

COinS