From Sentences to Documents: Extending Abstract Meaning Representation for Understanding Documents.

The overabundance of information and its heterogeneity requires new ways to access, process and generate knowledge according to the user's needs. To define an appropriate formalism to represent textual information capable to allow machines to perform language understanding and generation will be crucial for achieving these tasks. Abstract Meaning Representation (AMR) is foreseen as a standard knowledge representation that can capture the information encoded in a sentence at various linguistic levels. However, its scope only limits to a single sentence, and it does not benefit from additional semantic information that could help the generation of different types of texts. Therefore, the aim of this paper is to address this limitation by proposing and outlining a method that can extend the information provided by AMR and use it to represent entire documents. Based on our proposal, we will determine a unique, invariant and independent standard text representation, called canonical representation. From it and through a transformational process, we will obtain different text variants that will be appropriate to the users' needs.