Language Model Pre-training for Hierarchical Document Representations
This addresses the problem of capturing long-distance dependencies in documents for tasks like summarization and segmentation, but it is incremental as it builds on existing language model pre-training methods.
The paper tackles the challenge of learning hierarchical document representations from limited labeled data by proposing pre-training algorithms that integrate contextual information from entire documents. Experiments show effectiveness on tasks like document segmentation, question answering, and summarization.
Hierarchical neural architectures are often used to capture long-distance dependencies and have been applied to many document-level tasks such as summarization, document segmentation, and sentiment analysis. However, effective usage of such a large context can be difficult to learn, especially in the case where there is limited labeled data available. Building on the recent success of language model pretraining methods for learning flat representations of text, we propose algorithms for pre-training hierarchical document representations from unlabeled data. Unlike prior work, which has focused on pre-training contextual token representations or context-independent {sentence/paragraph} representations, our hierarchical document representations include fixed-length sentence/paragraph representations which integrate contextual information from the entire documents. Experiments on document segmentation, document-level question answering, and extractive document summarization demonstrate the effectiveness of the proposed pre-training algorithms.