Paper

Abstract Meaning Representation for Human-Robot Dialogue

---

Abstract

In this research, we begin to tackle the

\n

challenge of natural language understanding

\n

(NLU) in the context of the development of

\n

a robot dialogue system. We explore the adequacy

\n

of Abstract Meaning Representation

\n

(AMR) as a conduit for NLU. First, we consider

\n

the feasibility of using existing AMR

\n

parsers for automatically creating meaning

\n

representations for robot-directed transcribed

\n

speech data. We evaluate the quality of output

\n

of two parsers on this data against a manually

\n

annotated gold-standard data set. Second,

\n

we evaluate the semantic coverage and distinctions

\n

made in AMR overall: how well does it

\n

capture the meaning and distinctions needed

\n

in our collaborative human-robot dialogue domain?

\n

We find that AMR has gaps that align

\n

with linguistic information critical for effective

\n

human-robot collaboration in search and

\n

navigation tasks, and we present task-specific

\n

modifications to AMR to address the deficiencies.

Keywords: Dialogue systems, semantics, Natural Language Understanding

How to Cite:

Bonial, C. N., Donatelli, L., Ervin, J. & Voss, C. R., (2019) “Abstract Meaning Representation for Human-Robot Dialogue”, Society for Computation in Linguistics 2(1), 236-246. doi: https://doi.org/10.7275/v3c5-yd35