Although dominant in natural language processing, transformer-based models still struggle with long-sequence processing, due to the computational costs of their self-attention operations, which increase exponentially as the length of the input sequence grows. To address this challenge, we propose a **Sim**ple framework to enhance the long-content processing of off-the-shelf pre-trained transformers via three steps: **C**hunk, **A**lign, and **S**elect (SimCAS). More specifically, we first divide each long-sequence input into a batch of chunks, then align the inter-chunk information during the encoding steps, and finally, select the most representative hidden states from the encoder for the decoding process. With our SimCAS, the computation and memory costs can be reduced to linear complexity. In experiments, we demonstrate the effectiveness of the proposed method on various real-world long-text summarization and reading comprehension tasks, in which SimCAS significantly outperforms prior long-sequence processing baselines. The code is at [https://github.com/xjw-nlp/SimCAS](https://github.com/xjw-nlp/SimCAS).

While dominant in abstractive summarization, transformer-based language models with the standard maximum likelihood estimation (MLE) training remain challenged by two discrepancies: the misalignment between token-level training and sequence-level evaluation, and the divergence between teacher-forcing training manner and auto-regressive generation behavior. Recent studies have shown that sequence-level contrastive learning, which utilizes the quality differences between multiple summaries as prior information, can effectively mitigate these issues. However, as certain evaluation metrics often determine the contrastive signals in existing methods, this leads to the model performance aligning with the preferences of these metrics being limited by the evaluation capabilities of these metrics. Inspired by prior works that treat the evaluation of generated text as a text generation problem, we propose a generative evaluation-driven contrastive learning framework, which leverages the semantic understanding capabilities of the abstractive model itself to evaluate summary in reference-based settings. In this way, our method establishes a connection between the model’s reference-based evaluation and reference-free generation scenarios, allowing them to share the benefits of model capability enhancements. Extensive experiments on four summarization datasets demonstrate that our method outperforms the previous state-of-the-art regarding comprehensive performance. Various empirical analyses further substantiate the effectiveness of our method.

Most Transformer based abstractive summarization systems have a severe mismatch between training and inference, i.e., exposure bias. From diverse perspectives, we introduce a simple multi-level contrastive learning framework for abstractive summarization (SimMCS) and a tailored sparse decoder self-attention pattern (SDSA) to bridge the gap between training and inference to improve model performance. Compared with previous contrastive objectives focusing only on the relative order of probability mass assigned to non-gold summaries, SimMCS additionally takes their absolute positions into account, which guarantees that the relatively high-quality (positive) summaries among them could be properly assigned high probability mass, and further enhances the capability of discriminating summary quality beyond exploiting potential artifacts of specific metrics. SDSA simulates the possible inference scenarios of deviation in the training phase to get closer to the ideal paradigm. Our approaches outperform the previous state-of-the-art results on two summarization datasets while just adding fairly low overhead. Further empirical analysis shows our model preserves the advantages of prior contrastive methods and possesses strong few-shot learning ability.