Usage
  • 220 views
  • 238 downloads

Conversational agent with common-sense: Responding to nonsensical statements

  • Author / Creator
    Konar, Anandh Perumal
  • Conversational agents, also known as chatbots, are designed to have a real-time conversation with humans. Closed domain chatbots are limited to a specific task they’re designed to do. They can be rule-based or information retrieval based chatbots while open domain chatbots are meant to mimic humans in conversation. For a conversational agent to have a human-like conversation, the agent needs to generate grammatically correct and coherent output relative to the input. However, it is also crucial for the agent to have common sense and be equipped with basic facts about the world. For example, consider the following statement: “He gave birth to a baby.” For humans, it is common sense that males cannot give birth. Therefore, the given statement is nonsensical. However, for a system, it is challenging to distinguish between sensical and nonsensical statements.
    Moreover, it is even more challenging for a system to generate an explanation stating why the given statement is nonsensical. We propose UNION, a unified end-to-end framework, to generate a meaningful explanation to a given nonsensical statement that utilizes several existing commonsense datasets to allows a model to learn more dynamically under the scope of commonsense reasoning. To perform model selection efficiently, accurately, and promptly, we also propose a couple of auxiliary automatic evaluation metrics so that we can extensively compare the models from different perspectives. Our submitted system results in an excellent performance in the proposed metrics. It outperforms its competitors with the highest achieved score of 2.10 for human evaluation while maintaining a BLEU score of 15.7.

  • Subjects / Keywords
  • Graduation date
    Fall 2020
  • Type of Item
    Thesis
  • Degree
    Master of Science
  • DOI
    https://doi.org/10.7939/r3-bv31-kt07
  • License
    Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.