Post date: Jul 6, 2015 3:22:22 PM
Due to popular demand, we are making public the predictions for each system submitted to the 2013 shared task. We feel this will be a great resource for researchers interested in developing bigger and better ensemble classifiers. A first use of this data was explored in the BEA10 paper "Oracle and Human Baselines for Native Language Identification" by Malmasi, Tetreault and Dras (2015): https://aclweb.org/anthology/W/W15/W15-0620.pdf
If you plan on using the data and reference it in a paper, we suggest citing the NLI Shared Task 2013 summary paper as always:
http://aclweb.org/anthology/W/W13/W13-1706.pdf
Special thanks to Shervin Malmasi for helping prepare the data.