`Keep it Together': Enforcing Cohesion in Extractive Summaries by Simulating Human Memory

Anonymous

Abstract:

Extractive summaries are usually presented as lists of sentences with no expected cohesion between them.
In this paper, we propose a method to enforce cohesion whilst controlling for redundancy in summaries,
in cases where the input exhibits high redundancy.
The pipeline controls for content redundancy in the input as it is consumed, and balances informativeness and cohesion during sentence selection.
Our sentence selector simulates human memory to keep track of cohesive chains while building the summary, enforcing cohesive ties between noun phrases.
Extensive experiments, both automatic and human, revealed that it is possible to extract highly cohesive summaries that are as informative as summaries optimizing only for informativeness.
The extracted summaries exhibit a smooth topic transition between sentences as signaled by lexical chains, with
chains spanning adjacent or near-adjacent sentences.

Paper Type: long

Research Area: Summarization

Contribution Types: Model analysis & interpretability, NLP engineering experiment

Languages Studied: english

Consent To Share Submission Details: On behalf of all authors, we agree to the terms above to share our submission details.