The process of annotating a discourse tree with RST consists of three stages: first segmentation, followed by detection of relations and, finally, the construction of discursive trees. The criteria to be employed in each stage of the annotating process in the framework of the RST Spanish Treebank are explained in the next paragraphs.
The relations used for the annotation of texts should be:
Nucleus-Satellite Relations Multinuclear Relations Elaboration Union Evidence Contrast Enablement List Background Sequence Result Disjunction Justify Conjunction Concession Means Purpose Otherwise Antithesis Cause Condition Motivation Interpretation Circumstance Evaluation Reformulation Summary Solution Preparation Unconditional Unless
Regarding the "overlapping" units, the Same-Unit relation will be used, in the manner shown by the following example (extracted from da Cunha and Iruskieta, 2010):
To obtain a list of examples for all the relations here listed, we recommend the following article:
da Cunha, Iria: Iruskieta, Mikel (2010). «Comparing rhetorical structures of different languages: The influence of translation strategies». Discourse Studies 12(5). 563-598.
For the construction of a discourse tree it must be noted that relations must be binary, meaning that a satellite can only depend directly on a nucleus and vice versa. Firstly, EDUs are related on an intrasentential level. Secondly, sentences are related. Thirdly, paragraphs are related.
For the annotation of the discourse trees, the RSTtool (O’Donnell, 2000) interface must be used; it can be downloaded without charge in the next electronic address: www.wagsoft.com/RSTTool/
O’Donnell, M. (2000). «RSTTOOL 2.4 – A markup tool for rhetorical structure theory». In Proceedings of the International Natural Language Generation Conference. 253-256.
The file containing the relations to be employed, that should be incorporated to the RSTtool by the option Use Master, can be downloaded link.
