Final Year Project
Language Style Transfer on Non-Parallel Corpora
Learn more

Our Team

Three outstanding final year students in computer science, supervised by Professor Benjamin Kao.

Our mission is to create value and make a difference in the area of natural language processing.

Project Introduction

Language Style Transfer on Non-Parallel Corpora

Language style transfer is a popular task in Natural Language Processing (NLP) which aims to modify the style of a sentence while keeping its content unchanged. Previous work mainly focuses on using adversarial methods, which has struggled to produce high-quality outputs. In this project, we will first evaluate the performance of three most state-of-the-art approaches. We will then explore a new approach, iterative semantic matching, and compare it with previous methods. Our model will be evaluated on three commonly used benchmark data sets: Yelp sentiment data set, formality style data set, and a democratic-versus-republican political slants data set. Our approach can be applied to a wide range of applications, including stylistic dialogue systems, text formality conversion, and partisan news text generation.

Progress

What's done and what's next.

  • Sep 2018

    Milestone: Deliverable 1

    • Literature review
    • Project plan
    • Project website
  • Oct 2018

    • Clean data sets (formality, political slants)
    • Implement three baseline methods from literature reviews
    • Configure Amazon MTurk account
  • Nov 2018

    • Refine algorithm
    • Conduct error type analysis
    • Review literature on other domains work (including computer vision, reinforcement learning)
    • Complete human intelligence tasks with Amazon MTurk
  • Dec 2018

    Milestone: Deliverable 2

    • Deliver an unsupervised learning algorithm with style transfer in formality, political leaning and sentiment
    • Write-up the first report
  • Feb 2019

    • Collect dialogue data set (e.g., Semantic Web Interest Group IRC ChatLogs)
    • Explore unsupervised methods on large language corpus e.g. different translations of books
  • Mar 2019

    Milestone

    • Make broader application-oriented adaptations in three contexts, including human writing, understanding politics, and sentiments in digital assistants
  • April 2019

    • Prepare presentations and exhibition

Documentation

Download

Project Plan

Unavailable

Interim Report

Unavailable

Final Report

Methodology

Coming soon...

Results

Coming soon...

Contact Us

"A wise man's question contains half the answer." -- Solomon Ibn Gabirol