We gratefully acknowledge support from
the Simons Foundation
and member institutions
Full-text links:

Download:

Current browse context:

cs.CL
new | recent | 1704

Change to browse by:

References & Citations

Computer Science > Computation and Language

Title:Translating Neuralese

Abstract: Several approaches have recently been proposed for learning decentralized deep multiagent policies that coordinate via a differentiable communication channel. While these policies are effective for many tasks, interpretation of their induced communication strategies has remained a challenge. Here we propose to interpret agents' messages by translating them. Unlike in typical machine translation problems, we have no parallel data to learn from. Instead we develop a translation model based on the insight that agent messages and natural language strings mean the same thing if they induce the same belief about the world in a listener. We present theoretical guarantees and empirical evidence that our approach preserves both the semantics and pragmatics of messages by ensuring that players communicating through a translation layer do not suffer a substantial loss in reward relative to players with a common language.
Comments: Fixes description of the model human
Subjects: Computation and Language (cs.CL); Neural and Evolutionary Computing (cs.NE)
Cite as: arXiv:1704.06960 [cs.CL]
  (or arXiv:1704.06960v4 [cs.CL] for this version)

Submission history

From: Jacob Andreas [view email]
[v1] Sun, 23 Apr 2017 18:46:42 UTC (7,288 KB)
[v2] Mon, 25 Sep 2017 21:28:48 UTC (7,285 KB)
[v3] Thu, 28 Sep 2017 15:10:24 UTC (7,472 KB)
[v4] Sat, 6 Jan 2018 15:29:58 UTC (7,472 KB)