Since the first CoNLL Shared Task on NP chunking in 1999, CoNLL shared tasks over the years have tackled increasingly complex natural language learning tasks. Early shared tasks focused on identifying text chunks or named entities that typically correspond to single words or short phrases within a sentence. Shared tasks on semantic role labeling are concerned with identifying arguments for individual predicates and characterizing the relationship between each argument and the predicate. Shared tasks on joint dependency parsing and semantic role labeling target the syntactic and semantic structure of the entire sentence, rather than the argument structure of individual predicates. More recently, shared tasks on coreference went beyond sentence boundaries and started to deal with discourse phenomena, and shared tasks on grammatical error correction dealt with detecting and correcting grammatical errors in texts.
CoNLL-2015 will continue the tradition of having a high-profile shared task in natural language processing. This year’s shared task will be Shallow Discourse Parsing (SDP). A participant system is given English newswire texts 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, a participant system needs to i) identify explicit discourse connectives (e.g., “because”, “however”, “and”) in a text, ii) predict the sense of the discourse relations (e.g., “Cause”, “Condition”, “Contrast”), both explicit and implicit, and iii) identify the spans of text that serve as the two arguments for each discourse relation. Understanding such discourse relations is an important step in understanding the discourse structure of a text and will potentially benefit a wide range of natural language applications.
Participating teams will be provided with common training data and auxiliary annotations and linguistic resources. Blind test data will be used to evaluate the outputs of the participating teams using a common scoring software and evaluation metric.
In order to receive further information about the shared task, participants should register their intent to participate, by sending an e-mail to conll15st [at] gmail.com or filling out the registration form at the shared task website (http://www.cs.brandeis.edu/~clp/conll15st). The website is open for registration on January 26, 2015. Although the deadline for registration is not until March 1, 2015, we recommend participants to register as early as possible, in order not to miss any information. Registered participants will be added to a discussion forum where further details of the shared task will be made available.
Questions about the shared task can be sent to conll15st [at] gmail.com. More information about the shared task is available at the shared task home page: http://www.cs.brandeis.edu/~clp/conll15st.
Important dates:
January 2, 2015: Announcement of the shared task and set up of the shared task website.
January 26, 2015: registration begins, and release of training set and scorer.
March 1, 2015: Registration deadline.
April 20, 2015: Test set available.
April 24, 2015: Submission of participating systems.
May 1, 2015: System results due to participants.
May 8, 2015: Shared task system papers due.
May 18, 2015: Reviews due.
May 21, 2015: notification of acceptance.
May 28, 2015: camera-ready version of system papers due.
July 30-31, 2015. CoNLL conference (Beijing China).