NIPS Workshop "Let's Discuss: Learning Methods for Dialogue"

Humans conversing naturally with machines is a staple of science fiction. Building agents capable of mutually coordinating their states and actions via communication, in conjunction with human agents, would be one of the greatest engineering feats of human history. In addition to the tremendous economic potential of this technology, the ability to converse appears intimately related to the overall goal of AI.

Although dialogue has been an active area within the linguistics and NLP communities for decades, the wave of optimism in the machine learning community has inspired increased interest from researchers, companies, and foundations.  The NLP community has enthusiastically embraced and innovated neural information processing systems, resulting in substantial relevant activity published outside of NIPS.  The goal of this forum is increased interaction (dialogue!) between these communities at NIPS to accelerate creativity and progress.

Call For Papers
The workshop will consist of a mixture of invited talks and contributed talks, with panel sessions.  To avoid the "mini-conference effect", there is no poster session. We anticipate a total of six contributed talks of 20 minutes each, distributed evenly over the following three high-level areas:

• Being data-driven.
  • What can and cannot be done with offline evaluation on fixed data sets?
  • How can we facilitate development of these offline evaluation tasks in the public domain?
  • What is the role of online evaluation (e.g. as a benchmark?), and how would we make it accessible to the general community?
  • What can be done with simulated environments, or tasks where machines communicate solely with each other?
• Build complete applications.
  • Do we need to build a irreducible end-to-end system, or can we define modules with abstractions that do not leak?
  • How do we ease the burden on the human designer of specifying or bootstrapping the system?
• Model innovation.
  • What are the requisite capabilities for learning architectures, and where are the deficiencies in our current architectures?
  • How can we beneficially incorporate linguistic knowledge into our architectures?

​​The papers should be typeset according to NIPS format.
The paper should not exceed more than 4 pages (including references).
The authors of all the accepted papers will be expected to give a 20 minute talk (15 for the talk + 5 min for questions) and participate in a panel session.
Accepted papers will be displayed on the website.
There will be no posters.

Key Dates

• 10/09/2016: Submissions Due
• 10/23/2016: Acceptance Notification

Organizers:

• Hal Daume III
• Paul Mineiro
• Amanda Stent
• Jason Weston

Paper submission and more information : http://letsdiscussnips2016.weebly.com/

Contact: Jason Williams