This paper introduces a novel neural network model for abstractive text summarization, which addresses the issue of repetitive and incoherent phrases in longer documents and summaries. The model features a new intra-attention mechanism that separately attends over the input and continuously generated output, and a training method combining supervised word prediction with reinforcement learning (RL). The model is evaluated on the CNN/Daily Mail and New York Times (NYT) datasets, achieving a 41.16 ROUGE-1 score on the CNN/Daily Mail dataset, outperforming previous state-of-the-art models. Human evaluation also confirms that the model generates more readable summaries. The paper discusses the benefits of the intra-attention mechanism and the mixed learning objective function, which combines maximum-likelihood training and RL to reduce exposure bias and improve readability. The results demonstrate the effectiveness of the proposed model in generating high-quality, abstractive summaries.This paper introduces a novel neural network model for abstractive text summarization, which addresses the issue of repetitive and incoherent phrases in longer documents and summaries. The model features a new intra-attention mechanism that separately attends over the input and continuously generated output, and a training method combining supervised word prediction with reinforcement learning (RL). The model is evaluated on the CNN/Daily Mail and New York Times (NYT) datasets, achieving a 41.16 ROUGE-1 score on the CNN/Daily Mail dataset, outperforming previous state-of-the-art models. Human evaluation also confirms that the model generates more readable summaries. The paper discusses the benefits of the intra-attention mechanism and the mixed learning objective function, which combines maximum-likelihood training and RL to reduce exposure bias and improve readability. The results demonstrate the effectiveness of the proposed model in generating high-quality, abstractive summaries.