Learning Dense Representations for Entity Retrieval (1909.10506v1)

Abstract: We show that it is feasible to perform entity linking by training a dual encoder (two-tower) model that encodes mentions and entities in the same dense vector space, where candidate entities are retrieved by approximate nearest neighbor search. Unlike prior work, this setup does not rely on an alias table followed by a re-ranker, and is thus the first fully learned entity retrieval model. We show that our dual encoder, trained using only anchor-text links in Wikipedia, outperforms discrete alias table and BM25 baselines, and is competitive with the best comparable results on the standard TACKBP-2010 dataset. In addition, it can retrieve candidates extremely fast, and generalizes well to a new dataset derived from Wikinews. On the modeling side, we demonstrate the dramatic value of an unsupervised negative mining algorithm for this task.

Overview of CoNLL 2019 Proceedings Instructions

This paper serves as a comprehensive guide for authors preparing manuscripts for the CoNLL 2019 proceedings. It provides detailed instructions regarding the formatting and submission requirements for manuscripts intended for both review and final publication. As an exemplar, the document adheres to the specifications it prescribes.

Key Specifications

The paper establishes several key formatting requirements for submissions:

• Document Type and Format: Manuscripts must be produced in PDF format and adhere to a two-column layout for text, except for the title and author information. The documents should be formatted for printing on A4-sized paper.
• Font and Layout: Authors are instructed to use a uniform font, specifically Adobe's Times Roman, with prescribed font sizes for different sections (e.g., 15 pt bold for the title, 12 pt for author names, etc.).
• Page Limits: The length of the submission is restricted to eight pages of content, with unlimited pages for references. Upon acceptance, authors may expand their papers to nine pages of content to incorporate reviewer feedback.
• Double-Blind Review Process: The review process is double-blind; hence, initial submissions must exclude any author-identifying information to maintain anonymity.

Review and Submission Process

Submissions to CoNLL 2019 undergo a double-blind review process, necessitating the removal of authors' names and affiliations for initial submissions. Authors are to ensure that space for author information is allocated but kept empty until the final version. The style guide offers specific commands for LaTeX users to manage the anonymization and subsequent unveiling of author details.

Availability of Resources

The document’s authors provide electronically-available resources such as LaTeX style files and templates. These are made accessible to facilitate adherence to conference paper guidelines, ensuring that the manuscripts align with the required format and structure.

Implications

The standardization of formatting and submission practices as outlined in this guide has several implications:

• Consistency: Uniformity in paper presentation helps ensure proceedings are visually consistent, aiding both readability and professional appeal.
• Accessibility: By promoting standards in typography and layout, the instructions aim to enhance the accessibility and dissemination of research findings within the computational linguistics community.

Future Outlook

While the document specifically addresses the requirements for CoNLL 2019, its structured approach and recommendations may continue to inform future conferences' proceedings guidelines. As conferences evolve, modifications and updates will likely be incorporated to accommodate new publication technologies and methodologies, reflecting broader shifts in the field of computational linguistics.

In summary, the CoNLL 2019 proceedings instructions document plays a crucial role in guiding authors through the submission and formatting process, ensuring high standards of research dissemination whilst facilitating a rigorous and anonymous peer-review process.