The paper introduces Abstract Meaning Representation (AMR), a semantic representation language designed to capture the meanings of English sentences. The authors aim to create a sembank of simple, whole-sentence semantic structures to inspire new work in statistical natural language understanding and generation, similar to how the Penn Treebank revolutionized statistical parsing. AMR is a rooted, labeled graph that abstracts away from syntactic idiosyncrasies, focusing on logical meanings. It uses PropBank framesets and a set of relations to represent various linguistic phenomena, including frame arguments, general semantic relations, co-reference, and modal expressions. The paper outlines the format, content, and limitations of AMR, and discusses the development of an AMR editor and evaluation metrics. It also compares AMR with other semantic representation systems and outlines future work, including the expansion of the sembank and the application of AMR in machine translation.The paper introduces Abstract Meaning Representation (AMR), a semantic representation language designed to capture the meanings of English sentences. The authors aim to create a sembank of simple, whole-sentence semantic structures to inspire new work in statistical natural language understanding and generation, similar to how the Penn Treebank revolutionized statistical parsing. AMR is a rooted, labeled graph that abstracts away from syntactic idiosyncrasies, focusing on logical meanings. It uses PropBank framesets and a set of relations to represent various linguistic phenomena, including frame arguments, general semantic relations, co-reference, and modal expressions. The paper outlines the format, content, and limitations of AMR, and discusses the development of an AMR editor and evaluation metrics. It also compares AMR with other semantic representation systems and outlines future work, including the expansion of the sembank and the application of AMR in machine translation.