UnitedQA: A Hybrid Approach for Open Domain Question Answering (2101.00178v2)

Abstract: To date, most of recent work under the retrieval-reader framework for open-domain QA focuses on either extractive or generative reader exclusively. In this paper, we study a hybrid approach for leveraging the strengths of both models. We apply novel techniques to enhance both extractive and generative readers built upon recent pretrained neural LLMs, and find that proper training methods can provide large improvement over previous state-of-the-art models. We demonstrate that a simple hybrid approach by combining answers from both readers can efficiently take advantages of extractive and generative answer inference strategies and outperforms single models as well as homogeneous ensembles. Our approach outperforms previous state-of-the-art models by 3.3 and 2.7 points in exact match on NaturalQuestions and TriviaQA respectively.

Overview of ACL-IJCNLP 2021 Proceedings Instructions

The paper "Instructions for ACL-IJCNLP 2021 Proceedings" presents a comprehensive guide for authors intending to submit their manuscripts to the ACL-IJCNLP 2021 conference. It outlines detailed specifications for manuscript preparation, ensuring consistency and uniformity in the submissions for both review and final publication.

Manuscript Specifications

The guidelines delineate the precise requirements for formatting, including the use of a two-column layout, specific font sizes, and spacing. Authors are required to adhere to page limits: up to eight pages for long papers and four pages for short papers, excluding references and ethics statements. The final versions, upon acceptance, may extend by one additional page to incorporate reviewers’ feedback.

A crucial aspect of the instructions is ensuring that the submissions conform to the anonymity requirements of the double-blind review process by omitting any author-identifying information. This ensures an unbiased review and maintains the integrity of the evaluative process.

Submission Policies

The paper stipulates ACL-IJCNLP's multiple submission policy, prohibiting concurrent submissions of papers under review at other venues except for specific workshop submissions. Furthermore, it mandates that any overlapping content between submissions must not exceed 25%.

Authors are encouraged to submit supplementary materials, such as appendices or code, that can facilitate the replication of experiments. However, the main content should be self-contained, and reviewers are not obligated to consider supplementary files.

Technical Aspects

The document also details technical formatting instructions, particularly for \LaTeX users, addressing common issues such as PDF generation and cross-referencing. It emphasizes the importance of clear typography and layout, including the recommended use of Times Roman font for uniformity.

Additionally, to aid accessibility, authors are advised against relying solely on color distinctions in graphics, promoting grayscale readability for the benefit of color-blind individuals and those using print materials.

Implications and Future Developments

This instructional paper serves as a crucial resource for maintaining the high standards of publication and presentation associated with ACL-IJCNLP conferences. By rigorously enforcing these guidelines, the conference aims to ensure the clarity, accessibility, and professionalism of its proceedings.

Looking ahead, future iterations of these instructions might incorporate further technological advancements in manuscript preparation and submission processes, potentially integrating more automated tools for compliance checking. Additionally, as the field of NLP and computational linguistics continues to evolve, the conference guidelines might adapt to accommodate new types of content or methodologies pertinent to advanced research areas.

In conclusion, the comprehensive nature of these instructions underscores their role in facilitating a smooth submission process while upholding the academic rigour of the ACL-IJCNLP conference proceedings.