Dialog without Dialog Data: Learning Visual Dialog Agents from VQA Data

Abstract

Can we develop visually grounded dialog agents that can efficiently adapt to new tasks without forgetting how to talk to people? Such agents could leverage a larger variety of existing data to generalize to new tasks, minimizing expensive data collection and annotation. In this work, we study a setting we call "Dialog without Dialog", which requires agents to develop visually grounded dialog models that can adapt to new tasks without language level supervision. By factorizing intention and language, our model minimizes linguistic drift after fine-tuning for new tasks. We present qualitative results, automated metrics, and human studies that all show our model can adapt to new tasks and maintain language quality. Baselines either fail to perform well at new tasks or experience language drift, becoming unintelligible to humans. Code has been made available at https://github.com/mcogswell/dialog_without_dialog

19 pages, 8 figures

References in corpus (6)

• The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
• Best of Both Worlds: Transferring Knowledge from Discriminative Learning to a Generative Visual Dialog Model
• Evaluating Visual Conversational Agents via Cooperative Human-AI Games
• Compositional Languages Emerge in a Neural Iterated Learning Model
• Rethinking Action Spaces for Reinforcement Learning in End-to-end Dialog Agents with Latent Variable Models
• Examining Cooperation in Visual Dialog Models