Using Dominance Chains to Detect Annotation Variants in Parsed Corpora

In this paper, some results on the detection of variation in annotation in parsed corpora or tree banks are presented. Tree banks are generally built by means of using both automatic tools (i.e., taggers and parsers) and human intervention. In this process, inconsistencies (and, thus, variation) in the annotation arise, caused by a number of factors, for instance, disagreement in interpretation, incomplete or unclear annotation guidelines, etc. In this study, the algorithm for automatic detection of variation proposed in [1] is evaluated against the Tycho Brahe Corpus (TBC, [2]) and compared to an alternative implementation where variants of annotation are characterized by means of "dominance chains". Experimental results demonstrate that the modified version has better relative precision and recall than the original method.