论文发表在2018 ACL上,论文地址:https://arxiv.org/abs/1805.02473
本篇文章主要详细叙述了该论文的主要内容,全部都由自己总结,若文章有地方不正确或与读者想法不一致,欢迎在评论区中提出讨论.
AMR与AMR-to-Text Generation
AMR(Abstract Meaning Representation):a semantic formalism that encodes the meaning of a sentence as a rooted, directed graph.
AMR-to-Text Generation:to recover a text representing the same meaning as an input AMR graph.
sequence-to-sequence model
对AMR图作深度优先遍历,获得一个序列v1, . . . , vN,N为tokens的个数,例如上述AMR图的例子遍历后的序列为:“describe :arg0 ( person :name ( name :op1 ryan ) ) :arg1 person :arg2 genius”. (PS:我们可以看到,在AMR原图上“describe” and “genius”直接相连,而在序列上的距离为14