The 3rd Computational Linguistics Scientific Document Summarization Shared Task (CL-SciSumm 2017)

Results will be presented at the BIRNDL 2017 Workshop, a workshop co-located after
the ACM SIGIR Conference on Research and Development in Information Retrieval, Tokyo, Japan
Task Sponsored by Microsoft Research Asia

Last updated: Fri Aug 11 01:32:16 SGT 2017
  • System registration deadline extended to June 4th.
  • We are happy to announce that Prof. Simone Teufel will deliver the keynote at the workshop!


Call for Participation

You are invited to participate in the CL-SciSumm Shared Task at BIRNDL 2017. The shared task will be on automatic paper summarization in the Computational Linguistics (CL) domain. This year also we will invite the authors of selected system papers at the CL-SciSumm Shared Task, to submit extended versions to a special issue in International Journal of Digital Libraries (IJDL). The special issue will revolve on themes associated with the BIRNDL workshop.

This task follows up on the successful CLScisumm-16 at the BIRNDL workshop co-located with JCDL 2016, Rutgers, NJ, USA and the CL Pilot Task conducted as a part of the BiomedSumm Track at the Text Analysis Conference 2014 (TAC 2014). Over the two editions, a training corpus of ten topics and a development corpus of ten more topics from CL research papers were released. Participants were invited to enter their systems in a task-based evaluation. We also released the manually annotated dataset, comprising of ACL Computational Linguistics research papers and summaries. The output summaries are of two types: faceted summaries of the traditional self-summary (the abstract) and the community summary (the collection of citation sentences ‘citances’). We also group the citances by the facets of the text that they refer to. In our proposed shared task, we will expand the corpus with a new test dataset of 10 topics (closed for evaluation).

The CL-SciSumm 2017 corpus is expected to be of interest to a broad community including those working in computational linguistics and natural language processing, text summarization, discourse structure in scholarly discourse, paraphrase, textual entailment and text simplification.

We have secured support for the costs of the shared task annotation from Microsoft Research Asia. The National University of Singapore will be primarily responsible for the task's oversight.


Given: A topic consisting of a Reference Paper (RP) and Citing Papers (CPs) that all contain citations to the RP. In each CP, the text spans (i.e., citances) have been identified that pertain to a particular citation to the RP.

Task 1A: For each citance, identify the spans of text (cited text spans) in the RP that most accurately reflect the citance. These are of the granularity of a sentence fragment, a full sentence, or several consecutive sentences (no more than 5).

Task 1B: For each cited text span, identify what facet of the paper it belongs to, from a predefined set of facets.

Task 2 (optional bonus task): Finally, generate a structured summary of the RP from the cited text spans of the RP. The length of the summary should not exceed 250 words.

Evaluation: Task 1 will be scored by overlap of text spans measured by number of sentences in the system output vs the gold standard created by human annotators. Task 2 will be scored using the ROUGE family of metrics between (i) the system output and the gold standard summary from the reference spans (ii) the system output and the asbtract of the reference paper.

Organizations wishing to participate in the CL Shared Task track at BIRNDL 2017 are invited to register on EasyChair

by May 31st. Participants are advised to register as soon as possible in order to receive timely access to evaluation resources, including development and testing data. Registration for the task does not commit you to participation - but is helpful to know for planning. All participants who submit system runs are welcome to present their system at the BIRNDL Workshop.

Dissemination of CL-SciSumm work and results other than in the workshop proceedings is welcomed, but the conditions of participation specifically preclude any advertising claims based on these results. Any questions about conference participation may be sent to the organizers mentioned below.




The corpus for the CL-SciSumm task has been created by randomly sampling ten documents from the ACL Anthology corpus and selecting their citing papers. The training, development and testing set will be made publicly available at the GitHub link above at dates specified below.

The training set of 10 articles is already available for download and can be used by participants to pilot their systems. The development set of 10 articles, a part of the same corpus, will be released in April, which participants can add to the training set to tune their system parameters. Finally the test set of 10 articles will be released in July 1st. The system outputs from the test set should be submitted to the task organizers, for the collation of the final results to be presented at the workshop.

The best performing system on Task - 1a, the core component of Shared Task this year is from NJUST. Here is a recording of their talk.



Important Dates

Please consult the BIRNDL Workshop for official dates for the workshop.

Training Set ReleaseMay 1
Deadline for Registration and Short System Descriptions DueMay 20 May 31 June 4
Test Set ReleasedJuly 1
System Runs DueJuly 15 July 22
Preliminary System Reports Due in EasyChairJuly 30
Camera-Ready Contributions Due in EasyChairTBD
Participant Presentations at BIRNDL WorkshopAugust 11 in Tokyo, Japan

All deadlines for the CL-SciSumm shared task are calculated as 11:59pm Baker Island Time (BIT: UTC/GMT-12).




The CL-SciSumm task provides resources to encourage research in a promising direction of scientific paper summarization, which considers the set of citation sentences (i.e., "citances") that reference a specific paper as a (community created) summary of a topic or paper (Nanba, Kando and Okumura, 2011; Qazvinian and Radev, 2008). Citances for a reference paper are considered a synopses of its key points and also its key contributions and importance within an academic community. The advantage of using citances is that they are embedded with meta-commentary and offer a contextual, interpretative layer to the cited text. The drawback, however, is that though a collection of citances offers a view of the cited paper, it does not consider the context of the target user (Sparck Jones, 2007; Teufel and Moens, 2002; Nenkova and McKeown, 2011; Jaidka, Khoo and Na, 2013a), verify the claim of the citation or provide context from the reference paper, in terms the type of information cited or where it is in the referenced paper.

CL-SciSumm explores summarization of scientific research, for the computational linguistics research domain. An ideal summary of computational linguistics research papers would be able to summarize previous research by drawing comparisons and contrasts between their goals, methods and results, as well as distil the overall trends in the state of the art and their place in the larger academic discourse. Literature surveys and review articles in CL do help readers to gain a gist of the state-of-the-art in research for a topic. However, literature survey writing is labor-intensive and a literature survey is not always available for every topic of interest. What are needed, are resources which automate the synthesis and updating of automatic summaries of CL research papers.

Existing scientific summarization systems have automatically generated related work sections for a target paper by instantiating a hierarchical topic tree (Hoang and Kan, 2010), generating model citation sentences (Mohammad et al., 2009) or implementing a literature review framework (Jaidka et al., 2013). However, the limited availability of evaluation resources and human-created summaries constrains research in this area. The goal of the TAC 2014 CL Shared Task is to highlight the challenges and relevance of the scientific summarization problem, support research in automatic scientific document summarization and provide evaluation resources to push the current state of the art.



BIRNDL 2017 Workshop

The BIRNDL 2017 workshop will be held in Tokyo, , Japan. The workshop is a forum both for presentation of results (including failure analyses and system comparisons), and for more lengthy system presentations describing techniques used, experiments run on the data, and other issues of interest to NLP researchers. TAC track participants who wish to give a presentation during the workshop will submit a short abstract describing the experiments they performed. As there is a limited amount of time for oral presentations, the abstracts will be used to determine which participants are asked to speak and which will present in a poster session.



Organising Committee

Muthu Kumar Chandrasekaran -

He is broadly interested in natural language processing and its applications to information retrieval; specifically, in retrieving and organising information from asynchronous conversation media such as scholarly publications, discussion and debate forums. He was on the organizing committee of the CL-SciSumm 2016 Shared Task, the CL-SciSumm 2014 Pilot Task and the BIRNDL workshop. He also reviews for BIR, ACL and EMNLP conferences. He believes communication of scholarly research needs to be summarized to avoid redundant or outdated research and ensure faster progress to pressing problems. He is currently doing his Ph.D. research on a similarly motivated problem on Massive Open Online Course (MOOC) discussion forums on recommending salient student discussions for instructors to intervene given their limited bandwidth.

Kokil Jaidka -

Dr Kokil Jaidka is a postdoctoral researcher in Computer Science and Chief Technology Officer for the World Wellbeing Project at the University of Pennsylvania. She has been the lead coordinator of all aspects of the CL-SciSumm Shared Task since 2014, and she also co-organized the 1ST BIRNDL workshop. She has expertise working on large datasets using machine learning and unsupervised approaches on textual data, and in the specific areas of multi-document summarization and applied linguistics. She is a reviewer for Scientometrics, Applied Linguistics and Aslib journal of Information Processing \& Management. Her PhD dissertation involved the development of a literature review framework for the summarization of research papers. Currently, she is conducting social media analyses and user language modeling for opinion mining, behavioral profiling and health outcomes.

Min-Yen Kan -

My research interests fall under the areas of digital libraries, natural language processing, information retrieval, human-computer interaction. Specifically, they include document structure acquisition, verb analysis, digital library resource annotation and and applied text summarization. My research goal aims to investigate how natural language processing and information retrieval can be applied to improve scholarly publication and knowledge discovery.