Explore models of dialog in this lecture from CMU's Neural Networks for NLP course. Dive into chat-based and task-based dialog systems, covering generation-based models, neural approaches for response generation, and techniques to promote diversity in conversations. Learn about hierarchical encoder-decoder models, persona-based neural dialog, and retrieval-based chat systems. Examine challenges in dialog evaluation and methods to address them. Investigate traditional task-completion dialog frameworks, including neural approaches for natural language understanding, dialog state tracking, and language generation from dialog states.
Overview
Syllabus
Intro
Types of Dialog
Two Paradigms
Generation-based Models (Ritter et al. 2011)
Neural Models for Dialog Response Generation
Hierarchical Encoder- decoder Model (Serban et al. 2016)
Dialog allows Much More Varied Responses
Diversity Promoting Objective for Conversation (Li et al. 2016)
Diversity is a Problem for Evaluation!
Using Multiple References with Human Evaluation Scores (Gallay et al. 2015)
Learning to Evaluate • Use context, true response, and actual response to learn a regressor that predicts goodness (Lowe et al. 2017) • Important similar to model, but has access to reference
Problem 3: Dialog Agents should have Personality
Personality Infused Dialog (Mairesse et al. 2007)
Persona-based Neural Dialog Model (Li et al. 2017)
Dialog Response Retrieval
Retrieval-based Chat (Lee et al. 2009)
Neural Response Retrieval (Nio et al. 2014)
Smart Reply for Email Retrieval (Kannan et al. 2016)
Traditional Task-completion Dialog Framework
NLU (for Slot Filling) w/ Neural Nets (Mesnil et al. 2015)
Dialog State Tracking
Language Generation from Dialog State w/ Neural Nets (Wen et al. 2015)
Taught by
Graham Neubig
