Can one size fit all?: Measuring Failure in Multi-Document Summarization Domain Transfer
This work addresses the challenge of domain transfer failure in summarization for researchers and practitioners, highlighting incremental insights into model limitations and metric issues.
The paper tackled the problem of multi-document summarization models failing when applied to new domains without retraining, finding that models trained on one domain (e.g., News) show decreased factuality, quality, and target deviation when summarizing documents from other domains (e.g., Science or Conversation).
Abstractive multi-document summarization (MDS) is the task of automatically summarizing information in multiple documents, from news articles to conversations with multiple speakers. The training approaches for current MDS models can be grouped into four approaches: end-to-end with special pre-training ("direct"), chunk-then-summarize, extract-then-summarize, and inference with GPT-style models. In this work, we evaluate MDS models across training approaches, domains, and dimensions (reference similarity, quality, and factuality), to analyze how and why models trained on one domain can fail to summarize documents from another (News, Science, and Conversation) in the zero-shot domain transfer setting. We define domain-transfer "failure" as a decrease in factuality, higher deviation from the target, and a general decrease in summary quality. In addition to exploring domain transfer for MDS models, we examine potential issues with applying popular summarization metrics out-of-the-box.