基于Markov逻辑网的句子压缩方法

提出了一种基于Markov逻辑网通过删除单词进行英文句子压缩的方法。通过一阶逻辑公式表示单词的局部特征以及单词之间的相互依赖关系,确定单词是否应该删除。与现有方法相比,该方法把判别式学习和整数规划的优点相结合,既融合了丰富的句子特征,又以逻辑公式方便地表示全局约束。在书面和口语两个新闻数据集上的实验结果表明:该方法与L3和SVTL系统相比具有明显优势,在压缩率接近时,以人工压缩结果为评测标准,压缩后句子的依存关系F-score有较大提高。

Sentence compression with a Markov logic network

A method was developed to compress English sentences by removing unimportant words.The Markov logic network(MLN) incorporates local linguistic features and captures global dependencies between word deletion operations.The MLN based method combines the advantages of discriminative learning and integer linear programming by incorporating a rich set of features and expressing global constraints as logic formulae.Tests on both written and spoken news corpora show that this approach is superior to state-of-the-a...