2017Q3 Reports: ACL Anthology

From Admin Wiki
Jump to: navigation, search

This is a preliminary report -- needs to be edited for current information.

The ACL Anthology is a digital archive of research papers in computational linguistics, sponsored by the CL community, and freely available to all. As of 2016, we employ a Creative Commons Attribution license for materials published by ACL. This makes our content usable by the general public with attribution to the ACL (although it is not mandatory for any user to inform us of their use of our materials). Dual licensing for a fee is presumably possible (although not exercised currently).

The Anthology now contains over 42,100 (up from 37,000 papers in the last report in Q3) The new ACL Anthology is now active and will be switched to the primary Anthology site around ACL this year, once production issues (detailed below, are fixed and the solutions judged maintainable). However, we know a portion of our membership will want to still use the older version, so we are going to maintain both sites at least until the end of 2017. [Current version] [Legacy version]

Mailing List. The Anthology mailing list's (http://groups.google.com/group/acl-anthology) membership pool has grown, now consisting of 763 members (up from 634 from a year ago). This is an announcement-only list, where we notify members of newly listed released materials online.

Auxiliary Ingestion. The Anthology now has ingestion workflows for software, datasets, general attachments, slides and posters that are hosted with the ACL Anthology. Hyperlinks to videos are also saved; typically proceedings chairs, will ask videos to be saved to the third party techtalks.tv site, which has thus far, been happy to absorb ACL content.

Digital Object Identifiers. We have assigned DOIs to all ACL materials in 2015 and have assigned ones to all current material, minus TACL. With our current practice of assigning DOIs to all materials, our costs are likely to escalate to at least US$ 3K as we digitally publish at least this amount of scholarly articles.

ACL Anthology Reference Corpus version 2 (ACL ARC 2). We have released a newer updated version of the ACL Anthology Reference Corpus, which standardizes a release of the scholarly articles in the Anthology, alongside a machine readable processed version using open-source software derived from Min-Yen Kan's research group. The release was made in 1 March 2016, and distribution is currently solely by the ACL ARC website (http://acl-arc.comp.nus.edu.sg). Distribution by LDC is being sought for a newer version 3, as re-distribution rights that conflict with the earlier CC licensing have been revisited with the help of LDC. Once Min's ACL 2017 work is finished we are likely to pursue this in addition to the handover and sourcing for a replacement editor.

Work Queue. The current state of ingestion and development of the ACL Anthology is publicly available on the ACL Anthology's footer. https://docs.google.com/spreadsheets/d/166W-eIJX2rzCACbjpQYOaruJda7bTZrY7MBw_oa7B2E/pubhtml

Anthology Steering Committee. The ASC currently consists of Jing-Shin Chang, Min-Yen Kan and Paola Merlo. The ASC met virtually on 14 Jul 2016. The ASC also discussed supplementary materials in the ACL Anthology and other authority networks (for papers, authors, etc.) that may be being used, proposed by other institutions (e.g., MIT Press for CL journal), and revisions and better placement of the contributor's instructions for new material Anthology ingestion. Due to Min-Yen's involvement in ACL 2017, the ASC has yet to meet for its next meeting but may be able to do so physically soon.

Plans, Prioritized

  1. While the new Anthology is live, it lives on a university virtual machine in Singapore, and does not scale well to provide adequate bandwidth when faced with the full access from the ACL membership and general public. We have investigated which service to take our work towards as it likely requires a virtual private server (VPS) account. We have also investigated its cost, as we need to install certain software and libraries that usually requires root privileges. We solicited and received volunteers to help with this process. Martin Villaba and Christoph Teichmann of Saarland have duplicated a set up of the Anthology and have gotten a more, high-powered virtual machine to serve the Anthology. The ACL Anthology is now containerized, and in Docker. We are still working on this as the ACL 2017 conference is happening. We have on our agenda to set up a mirroring system to allow several worldwide mirrors to serve the Anthology materials and volunteers to help with provisioning a local Anthology copy are welcomed to contact us
  1. A large current problem with the ACL Anthology is with indexing (specifically Google Scholar). We have known about the problem for a while and it is being addressed along with the virtual machine.
  2. Min-Yen Kan, current Anthology Editor, will relinquish editorship of the ACL Anthology in 2018. It is time to begin searching for qualified individuals to fulfil this important, voluntary role to ensure that service to the community will not be interrupted, while enjoying the benefits of having new leadership rejuvenate the Anthology with new ideas. The ASC concurs that this process needs to start, and recommends that the ACL Exec come up with a selection process, adding that the existing Nominating Committee could be asked to help with the process, but that the NC should be enlarged to add at least the current Anthology Editor, and/or a member that could advise on the technical expertise of candidates. The ASC additionally recommends that the NC:
    • Insist that prospective Editors have access to volunteers (i.e., students) who have the technical ability to help with the infrastructural maintenance work.
    • Conduct an open call for (self-) nominations that might dovetail with a call for general volunteers.
  3. A previous discussion (with Ken Church) proposed that we create a single BibTeX file for all Anthology materials. The beta Anthology can generate such information fairly easily with its database backing; we plan to have this file available during the ACL 2016 conference.
  4. We are capturing abstracts (albeit with some noise) from START, but have yet to successfully integrate this into the ACL Anthology's index.
  5. For long-term preservation, to create a XML representation of all of the metadata used to create the Anthology. This is similar in nature to the XML dump of DBLP or Wikipedia. It allows a clean separation of the underlying data in the Anthology from the code used to present it.
  6. Collaboration with START (also may involve the Conference Officer's work) to integrate user accounts in their system. This would allow START to have authority records for authors such that new paper submissions might start with correct, canonical forms of author names. The ASC is aware of ORCIDs and other name authority systems that might also be useful in this process.
  7. Collaboration with ELRA to allow the categorization of papers against the LRE Map and ISLRNs.
  8. To allow third-party applications to automatically annotate articles with new metadata on existing papers via an API. Such an API is a production API, allowing third-parties to add auto-analyzed materials to the Anthology (e.g., auto-extracted keywords, summaries). This will raise the visibility of the Anthology as a object of study, complementing work on the ACL ARC.