Text Segmentation by Cross Segment Attention
This work addresses document and discourse segmentation to aid downstream tasks like information retrieval, but it appears incremental as it builds on existing transformer methods.
The authors tackled the problem of text segmentation for NLP tasks by proposing three transformer-based architectures, achieving a new state-of-the-art with large reductions in error rates across three standard datasets.
Document and discourse segmentation are two fundamental NLP tasks pertaining to breaking up text into constituents, which are commonly used to help downstream tasks such as information retrieval or text summarization. In this work, we propose three transformer-based architectures and provide comprehensive comparisons with previously proposed approaches on three standard datasets. We establish a new state-of-the-art, reducing in particular the error rates by a large margin in all cases. We further analyze model sizes and find that we can build models with many fewer parameters while keeping good performance, thus facilitating real-world applications.