This workshop will highlight research that investigates: existing or novel RL techniques for text-based settings; what agents that solve text-based games learn about language; and more generally, whether text-based games provide a good testbed for research at the intersection of RL and NLP. The program will feature a collection of invited talks alongside contributed posters and spotlight talks. Panel discussions will highlight perspectives of influential researchers from the RL and NLP fields and encourage open dialogue. In addition to the exchange of ideas and the initiation of collaboration, an expected outcome is that text-based games emerge more prominently as a benchmark task to bridge RL and NLP research. For a more detailed description, see the NeurIPS website.
The workshop will also introduce a text-based games competition to drive AI research. For this we'll release a carefully designed collection of training/validation/test games built with the TextWorld framework. Stay tuned!
Call for Papers
We welcome papers describing original research related to text-based games and their unique challenges, with a slight preference for work at the intersection of RL and NLP. This includes, but is not limited to, papers that investigate the following topics in the context of text-based games:
RL in compositional, combinatorial action spaces
Use of linguistic priors to constrain RL policies [Andreas et al., 2017]
Learning representations of language suited to planning and decision-making, e.g., approximating Q-values while using the textual information in natural language action spaces [He et al., 2016]
Fundamental RL challenges that are prevalent in text-based games, like (sub)goal identification and efficient exploration
The use of gaming environments for RL research
Grounded language understanding and linguistic common sense
Language generation/evaluation in goal-oriented settings
Procedural generation of games and training environments
Automatic or crowdsourcing methods for linguistic diversity in simulation
Papers should not have been previously presented at other conferences, but follow-up work, reviews or summaries of prior work, and papers submitted but not yet accepted elsewhere are welcome. Papers should be formatted according to the current NeurIPS style guide and are required to be under 5 pages in length, excluding references. The workshop proceedings will be non-archival.
All accepted papers will be presented at the poster sessions, and the top submissions will be given spotlight presentations. Submissions should be sent to EasyChair before 11:59pm UTC-10 on 9 November. Submitted manuscripts should be anonymized.
Katja Hofmann, Microsoft Research
Karthik Narasimhan, Princeton
Lihong Li, Google
Maxime Chevalier-Boisvert, MILA
Jason Weston, FAIR
Nancy Fulda, BYU
Please email all workshop inquiries to wp2018 [at] easychair [dot] org