Multilingual Shallow Discourse Parsing
This is the 2nd edition of the CoNLL Shared Task on Shallow Discourse Parsing, following
the first edition in 2015. A participant system is given a piece
of newswire text as input and returns discourse relations in the form of a discourse connective (explicit or implicit)
taking two arguments (which can be clauses, sentences, or multi-sentence segments). Specifically, the
participant system needs to
- locate both explicit (e.g., "because", "however") discourse
connectives in the text
- identify the spans of text that serve as the two arguments
for each discourse connective
- predict the sense of the discourse connectives (e.g., "Cause", "Contrast")
Recognizing such discourse relations is an important part of natural language
understanding, which benefits a wide range of natural language applications.
More detail and examples.
What's new this year?
There are a few things. More detail will be provided later.
- You can choose to do the task in English and/or Chinese.
- You can choose to do the supplementary task, which is sense classification using gold standard argument pairs
- Evaluation based on argument partial matching
Official blind test sets in English and Chinese
The task has already concluded. We have released the blind test sets used in
English and
Chinese.
Joining the shared task
The instructions are the same whether you would like to participate in both languages and/or just the supplementary task.
- Complete the registration form (one per team)
- Submit the license agreement form to LDC
- Download the data from the link, which LDC will send to you after a few days
- Check out the resources that might be useful
- Login to the evaluation platform on tira.io using your credential, which we will send to you
- Clone/fork from our github repo and familiarize yourself with the data format
- Start developing the parser
Stay updated
Updates and announcements will be made through the task
forum. You should also read our weekly
blog posts for tips, literature reviews, and interesting stuff related to discourse parsing.
Last year blog might also be of interest.
Timeline
- January 15, 2016: Task begins
- February 29, 2016: Last day to join the task
- By April 29, 2016: Test sets available
- May 8, 2016 (11:59PM UTC/GMT-10): System submission deadline
- May 16, 2016: Results announced to participants
- May 25, 2016 (11:59PM UTC/GMT-10) : System papers due.
- June 1, 2016: Reviews due.
- June 8, 2016: Notification of acceptance.
- June 15, 2016: Camera-ready version of system papers due.
- Aug 11-12, 2016: CoNLL conference (Berlin, Germany, colocated with ACL).