RECORD
Infusing Commonsense World Models with Graph Knowledge
Computation and Language · Submitted 13 Jan 2023
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim We find it is possible to improve the consistency of action narration models by training on graph contexts and targets, even if graphs are not present at test time.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
While language models have become more capable of producing compelling language, we find there are still gaps in maintaining consistency, especially when describing events in a dynamically changing world. We study the setting of generating narratives in an open world text adventure game, where a graph representation of the underlying game state can be used to train models that consume and output both grounded graph representations and natural language descriptions and actions. We build a large set of tasks by combining crowdsourced and simulated gameplays with a novel dataset of complex actions in order to to construct such models. We find it is possible to improve the consistency of action narration models by training on graph contexts and targets, even if graphs are not present at test time. This is shown both in automatic metrics and human evaluations. We plan to release our code, the new set of tasks, and best performing models.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | While language models have become more capable of producing compelling language, we find there are still gaps in maintaining consistency, especially when describing events in a dynamically changing world. | background 0.56 | 0.83 | 0.04 | asserted 1.00 |
| s1 | We study the setting of generating narratives in an open world text adventure game, where a graph representation of the underlying game state can be used to train models that consume and output both grounded graph representations and natural language descriptions and actions. | contribution 0.78 | 0.38 | 0.05 | asserted 0.73 |
| s2 | We build a large set of tasks by combining crowdsourced and simulated gameplays with a novel dataset of complex actions in order to to construct such models. | method 0.96 | 0.93 | 0.04 | asserted 0.95 |
| s3 | We find it is possible to improve the consistency of action narration models by training on graph contexts and targets, even if graphs are not present at test time. | result 1.00 | 0.96 | 0.84 | hedged possibility 0.56 |
| s4 | This is shown both in automatic metrics and human evaluations. | result 0.95 | 0.89 | 0.09 | asserted 0.99 |
| s5 | We plan to release our code, the new set of tasks, and best performing models. | future work 0.97 | 0.26 | 0.03 | asserted 0.54 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
While language models have become more capable of producing compelling language, we find there are still gaps in maintaining consistency, especially when describing events in a dynamically changing world. We study the setting of generating narratives in an open world text adventure game, where a graph representation of the underlying game state can be used to train models that consume and output both grounded graph representations and natural language descriptions and actions. We build a large set of tasks by combining crowdsourced and simulated gameplays with a novel dataset of complex actions in order to to construct such models. We find it is possible to improve the consistency of action narration models by training on graph contexts and targets, even if graphs are not present at test time. This is shown both in automatic metrics and human evaluations. We plan to release our code, the new set of tasks, and best performing models.
Human review
How the literature uses it
What this paper does with the arXiv papers it cites
What it cites · 1 sentence without a confident label
Uses the method or tool of 1 paper
-
Thanks to the ParlAI framework [ref] , which made it easy to implement the tasks for this work.
ParlAI: A Dialog Research Software Platform · Acknowledgements · 1.00
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.