Toward Dialogue and Reasoning Mechanisms to Enable More Natural and Socially-Appropriate Task-Based Human-Robot Interactions
Abstract: When humans
communicate with one another they utilize a variety of subtle linguistic cues that
robots need to understand and reciprocate in order to achieve more natural and effective
collaborative interactions with humans. We begin by presenting a simple human-robot
interaction scenario and show that appropriate behavior by the robot requires the
ability to explicitly reason about ... read morea variety of social and introspective factors that
include (but are not necessarily limited to): the literal and non-literal intent of
utterances, the knowledge and capability of the robot, social roles and relationships,
goal priority and timing, and both norms of moral and social (e.g. politeness)
varieties. Many current natural language (NL) systems used in human-robot interaction
are unable modulate NL interactions appropriately given these considerations. Given this
need, we present a novel NL understanding and generation architecture for robotic agents
that provides a framework to achieve these desired capabilities. We begin by presenting
how this architecture implements both pragmatic reasoning and reasoning regarding the
belief states of interlocutors. We then show how this architecture can be utilized to
both understand and appropriately deploy certain sentential adverbial modifiers, which
enhance the informativeness and naturalness of natural language interactions. We also
show how the architecture can not only understand non-literal requests, but also
appropriately generate non-literal requests based on social context. Next, we
demonstrate how the NL architecture is also able to generate responses to both literal
and non-literal requests that conform to human sociolinguistic
Thesis (Ph.D.)--Tufts University, 2016.
Submitted to the Dept. of Computer Science.
Advisors: Matthias Scheutz, and Ariel Goldberg.
Committee: Anselm Blumer, Remco Chang, and Bertram Malle.
Keyword: Artificial intelligence.read less