| Attachment | Size |
|---|---|
 cfp.txt | 3.16 KB |
Call for Participation:
TextGraphs-14 Shared Task on Explanation Regeneration
https://competitions.codalab.org/competitions/23615
We invite participation in the 2nd Shared Task on Explanation Regeneration
associated with the 14th Workshop on Graph-Based Natural Language Processing
(TextGraphs 2020).
All systems participating in the shared task will be invited to submit system description papers.
Each system description paper will be peer-reviewed by (two) other participating teams and will be
presented as a poster at the main workshop.
Overview
========
This shared task focuses on explanation reconstruction, a stepping-stone
towards general multi-hop inference over language. In particular, the inputs
to this task consist of questions and their correct answers. Participating
systems must extract and rank explanation sentences from a provided
structured knowledge base such that the top-ranked sentences provide
a complete explanation for the given answer.
While large language models (BERT, XLNet) achieved the highest performance in
the 2019 shared task, substantially advancing the state-of-the-art over
previous methods, absolute performance remains modest, highlighting the
difficulty of generating detailed explanations through multi-hop reasoning.
This 2020 instantiation of the shared task includes approximately three times
more data than the original 2019 shared task. Submissions that evaluate
how well existing models designed on 2-hop multihop question answering datasets
(e.g. HotPotQA, QASC, etc) perform at many-fact multi-hop explanation regeneration
are encouraged.
Example
=======
For example, for the question: "Which of the following is an example of an
organism taking in nutrients?" with the correct answer: "a girl eating an
apple", an ideal system would rank the following explanatory statements
at the top of its extracted sentences:
1. A girl means a human girl.
2. Humans are living organisms.
3. Eating is when an organism takes in nutrients in the form of food.
4. Fruits are kinds of foods.
5. An apple is a kind of fruit.
Important Dates
===============
2020-03-06: Training data release
2020-04-06: Test data release; Evaluation start
2020-05-06: Evaluation end
2020-05-20: System description paper deadline
2020-06-10: Deadline for reviews of system description papers
2020-06-24: Author notifications
2020-07-11: Camera-ready description paper deadline
2020-09-14: TextGraphs-14 workshop
Data
====
The data used in this shared task contains approximately 5,100 questions drawn
from the AI2 Reasoning Challenge (ARC) dataset (Clark et al., 2018), together
with explanation sentences for their correct answers drawn from the
WorldTree V2 corpus (Xie et al., 2020, Jansen et al., 2018).
The knowledge base supporting these questions contains approximately 9,000 facts.
To encourage a variety of solving methods, the knowledge base is available both as
plain-text sentences (unstructured) as well as semi-structured tables. Facts are a
combination of scientific knowledge as well as common-sense/world knowledge.
Please see the shared task website for more details:
https://competitions.codalab.org/competitions/23615
https://github.com/cognitiveailab/tg2020task
References
=========
Jansen P. and Ustalov D. TextGraphs 2019 Shared Task on Multi-Hop Inference for Explanation Regeneration. Proceedings of the Thirteenth Workshop on Graph-Based Methods for Natural Language Processing (TextGraphs-13). Hong Kong: Association for Computational Linguistics, 2019, pp. 63–77.