STS 2017 Shared Task Paper

[Daniel Cer]

Hi everyone,

For folks who will be able to make it to the workshop, we look forward to seeing you in Vancouver soon! 

A draft of the SemEval 2017 proceedings is available at: http://nlp.arizona.edu/SemEval-2017/. The STS shared task paper is available here: http://nlp.arizona.edu/SemEval-2017/pdf/SemEval001.pdf. As with the earlier draft, the methods section includes a hopefully useful summary of the techniques used by well performing systems (Section 6.6, pg. 5-8) 

The final paper includes a new comparison of well performing participant systems with several state-of-the-art baselines on the STS benchmark task (Section 8, pg. 9-10). Additionally, the paper includes error analysis that describes and provides examples for types of errors still being made by top performing systems including issues with negation, meaning composition and semantic blending (Section 7, pg. 8-9).

We still welcome any feedback on the paper. If there is anything substantive that would be worth revising for archival purposes, I believe it is possible to post paper corrections to the ACL anthology.

Dan

-----

STS 2017 Co-Organizer