Four knowledgeable referees reviewed the paper and overall they are in favor of acceptance. Key concerns include the large error bars in the tables, as well as the lack of experiments with more labeled data. Rebuttal generally does a good job at addressing the reviewers' concerns, including the second one above (but not the first one). All in all, I recommend acceptance, but the authors are strongly encouraged to address the reviewers' concerns in the final version of the paper.