Controllable Summarization with Constrained Markov Decision Process
This work addresses the problem of generating summaries with specific attributes for users in NLP, but it is incremental as it builds on existing controllable summarization methods.
The authors tackled controllable text summarization by proposing a Constrained Markov Decision Process (CMDP) framework to enforce user-specified attributes like length, entity coverage, and abstractiveness, achieving improved compliance with requirements in experiments on benchmarks.
We study controllable text summarization which allows users to gain control on a particular attribute (e.g., length limit) of the generated summaries. In this work, we propose a novel training framework based on Constrained Markov Decision Process (CMDP), which conveniently includes a reward function along with a set of constraints, to facilitate better summarization control. The reward function encourages the generation to resemble the human-written reference, while the constraints are used to explicitly prevent the generated summaries from violating user-imposed requirements. Our framework can be applied to control important attributes of summarization, including length, covered entities, and abstractiveness, as we devise specific constraints for each of these aspects. Extensive experiments on popular benchmarks show that our CMDP framework helps generate informative summaries while complying with a given attribute's requirement.