Annotation Guidelines Sample Clauses
Annotation Guidelines. The annotation guidelines were created in an iterative process. A first draft was created, containing general guidelines as well as specific examples of difficult situations. Two archaeologists used the guidelines to annotate around 100 sen- tences, and these annotations were compared to our own desired annotations to see where problems and inconsistencies were encountered. This information was then used to update the guidelines, after which they were tested again. This led to an IAA (F1 score, further explained in section 3.5.1) of 0.94 between the two testers, which we consider sufficient for this task. During the annotation process itself, whenever one of the annotators ran into a situation that was unclear, this was added as an example to the guidelines. The annotation guidelines (in Dutch) can be downloaded as part of the data set (Brandsen, 2019).
Annotation Guidelines. The current revised version of the annotation guidelines of the TDB was distributed to the project members in March 2010. In the guidelines, general annotation principles, and exceptional annotation issues are described mostly following the PDTB annotation guidelines (PDTB-Group, 2006 & 2008). Major principles include what Turkish discourse connectives are, and where the arguments (ARG1 and ARG2) shall be searched. Annotation issues concerning the following exceptional cases of the TDB are also described in the annotation guidelines (TDB-Group, 2010): • Minimality principle (annotations shall contain minimum text spans enough to fully cover the discourse) • Annotation connectives with a listing function • Annotation of texts whose sentences/clauses are interrupted by punctuation marks • Annotation of shared arguments • Structures that shall not be annotated because of the absence of abstract object interpretation In brief, the TDB annotation guidelines form a sound basis for annotation efforts.