Competition

To evaluate the responses of a system participating in the challenge, we employ the F1 score, as in [1]. Let system A_i return a set S^t_Ai of character offsets, for parts of the post found to be toxic. Let G^t be the character offsets of the ground truth annotations of t. We compute the F1 score of system A_i with respect to the ground truth G for post t as follows, where |·| denotes set cardinality.

If S^t_G is empty for some post t (no gold spans are given for t), we set F1^t(A_i, G) = 1 if S^t_Ai is also empty, and F1^t(A_i, G) = 0 otherwise. We finally average F1^t(A_i, G) over all posts t of T to obtain a single score for system A_i.

References

[1] G. Da San Martino, S. Yu, A. Barr ́on-Cedeno, R. Petrov, and P. Nakov. 2019. Fine-grained analysisof propaganda in news article. In EMNLP-IJCNLP, pages 5640–5650.

Terms and Conditions

• By submitting results to this competition, you consent to the public release of your scores at this website and at the SemEval 2021 workshop and in the associated proceedings, at the task organizers' discretion. Scores may include, but are not limited to, automatic and manual quantitative judgments, qualitative judgments, and such other metrics as the task organizers see fit. You accept that the ultimate decision of metric choice and score value is that of the task organizers.
• You further agree that the task organizers are under no obligation to release scores and that scores may be withheld if it is the task organizers' judgment that the submission was incomplete, erroneous, deceptive, or violated the letter or spirit of the competition's rules. Inclusion of a submission's scores is not an endorsement of a team or individual's submission, system, or science.
• This task has a single evaluation phase. To be considered a valid participation/submission in the task's evaluation, you agree to submit a single (possibly empty) list of character offsets (as in the task overview) per test text (post), for every test text. 
• Each team must create and use exactly one CodaLab account.
• Team constitution (members of a team) cannot be changed after the evaluation phase has begun.
• During the evaluation phase, each team can submit as many as ten submissions; the top-scoring submission will be considered as the official submission to the competition.
• The organizers and the organizations they are affiliated with make no warranties regarding the datasets provided, including but not limited to being correct or complete. They cannot be held liable for providing access to the datasets or the usage of the datasets.
• Each task participant will be assigned at least one other teams' system description paper for review, using the START system. The papers will thus be peer reviewed.

Overview

We used posts (comments) from the publicly available Civil Comments dataset, which already comprises post-level toxicity annotations, i.e., annotations indicating which (entire) posts are toxic, instead of annotations of particular toxic spans in toxic posts. We retained only posts that had been found toxic (or severely toxic) by at least half of the crowd-raters from Borkan et al's annotation [1]. This left approximately 30k toxic posts, out of a total of 1.2M posts in the original dataset.

We selected a random 10k subset of the 30k posts for toxic spans annotation. We used a crowd-annotation platform, and we employed three crowd-raters per post, all of whom were warned for explicit content (only coders who allowed adult content were selected). Coders were selected from the smallest group of the most experienced and accurate contributors (i.e., `highest quality’). The annotators were given the following instructions: ``Extract the toxic word sequences (spans) of the comment, by highlighting each such span and then clicking the right button. If the comment is not toxic or if the whole comment should be annotated, check the appropriate box and do not highlight any span.'' 

Note that we do not claim it is possible to annotate toxic spans in all toxic posts. For example, in some toxic posts the core message being conveyed may be inherently toxic (e.g., a sarcastic post indirectly claiming that people of a particular origin are inferior) and, hence, it may be difficult to attribute the toxicity of those posts to particular spans. In such cases, the corresponding posts may have no toxic span annotations (see, for example, the fourth post of the following table).

Inter-annotator agreement

In an initial experiment, we employed five crowd raters per post for a sample of 35 posts, to measure inter-annotator agreement. We computed the mean pairwise Cohen's Kappa per post (using character offsets as instances being classified in two classes, toxic and non-toxic) and averaged over the 35 posts, which yielded a Kappa score of 0.61.

Ground truth

To obtain the ground truth of our dataset, we used the following process: for each post t, first we mapped each annotated span to its character offsets. Then we merged the annotated spans of each rater per post, to obtain a single set of character offsets per rater and post. We assigned a toxicity score to each character offset of t, computed as the fraction of raters who annotated that character offset as toxic (included it in their toxic spans). We then retained only character offsets with toxicity scores higher than 50%; i.e., at least two raters must have included each character offset in their toxic spans for the offset to be included in the ground truth. 

Trial dataset

A trial dataset of 690 texts has been released. Note that some texts do not include any annotations while others include one or more toxic spans. Please find the respective CSV file in the "Participate" tab. We suggest the use of Pandas (data loading/processing) and ast.literal_eval (restoring the lists after loading). Some useful lines of code follow:

References 

[1] D. Borkan, L. Dixon, J. Sorensen, N. Thain, andL. Vasserman. 2019. Nuanced metrics for measur-ing unintended bias with real data for text classifi-cation. In WWW, pages 491–500, San Francisco, USA.