ON THE ROBUSTNESS OF AUTHORSHIP ATTRIBUTION 439
authorship attribution.^26 Finally, a missing block in the
authorship attribution research that is necessary to use this
technology as evidence in court is the ability to explain the
automatically derived decisions. In the case of attribution models
based on low-level information, like character n-grams, that
seem to be the most robust and effective approach, what is
needed is a way to associate this highly dimensional information
to some human interpretable high-level features.
(^26) See Shlomo Argamon & Patrick Juola, Overview of the International
Authorship Identification Competition at PAN-2011 (Sept. 19–22, 2011),
http://www.uni-weimar.de/medien/webis/research/events/pan-11/pan11-
papers-final/pan11-authorship-identification/juola11-overview-of-the-
authorship-identification-competition-at-pan.pdf; Patrick Juola, An Overview
of the Traditional Authorship Attribution Subtask Notebook for PAN at CLEF
2012 (Sept. 17–20, 2012), http://www.uni-weimar.de/medien/webis/research/
events/pan-12/pan12-papers-final/pan12-author-identification/juola12-overview-
of-the-traditional-authorship-attribution-subtask.pdf.