Inference Time Style Control for Summarization
This addresses the need for flexible style control in summarization for users like content creators, though it is incremental as it builds on pre-trained models.
The paper tackles the problem of generating summaries in different styles without needing style-specific corpora or separate models, by introducing two novel inference-time methods that adjust decoder states and constrain word usage, resulting in simpler language summaries and headlines with ideological leanings that humans can distinguish.
How to generate summaries of different styles without requiring corpora in the target styles, or training separate models? We present two novel methods that can be deployed during summary decoding on any pre-trained Transformer-based summarization model. (1) Decoder state adjustment instantly modifies decoder final states with externally trained style scorers, to iteratively refine the output against a target style. (2) Word unit prediction constrains the word usage to impose strong lexical control during generation. In experiments of summarizing with simplicity control, automatic evaluation and human judges both find our models producing outputs in simpler languages while still informative. We also generate news headlines with various ideological leanings, which can be distinguished by humans with a reasonable probability.