Compositional Generalization and Natural Language Variation: Can a Semantic Parsing Approach Handle Both? (2010.12725v2)

Abstract: Sequence-to-sequence models excel at handling natural language variation, but have been shown to struggle with out-of-distribution compositional generalization. This has motivated new specialized architectures with stronger compositional biases, but most of these approaches have only been evaluated on synthetically-generated datasets, which are not representative of natural language variation. In this work we ask: can we develop a semantic parsing approach that handles both natural language variation and compositional generalization? To better assess this capability, we propose new train and test splits of non-synthetic datasets. We demonstrate that strong existing approaches do not perform well across a broad set of evaluations. We also propose NQG-T5, a hybrid model that combines a high-precision grammar-based approach with a pre-trained sequence-to-sequence model. It outperforms existing approaches across several compositional generalization challenges on non-synthetic data, while also being competitive with the state-of-the-art on standard evaluations. While still far from solving this problem, our study highlights the importance of diverse evaluations and the open challenge of handling both compositional generalization and natural language variation in semantic parsing.

Overview of the Instructions for ACL-IJCNLP 2021 Proceedings

The paper "Instructions for ACL-IJCNLP 2021 Proceedings" serves as a comprehensive guide for authors preparing manuscripts intended for submission to the ACL-IJCNLP 2021 conference, including both initial submissions and final versions of accepted papers. Authored by a distinguished panel, this document incorporates adaptations from prior guidelines used in various ACL, NAACL, and EMNLP proceedings, demonstrating a refined iteration of established best practices for academic publications in the computational linguistics community.

Core Sections and Guidelines

Document Formatting and Length

A critical component of the guidelines is document formatting. Authors must adhere to a stringent two-column format specific to A4 paper dimensions, ensuring uniformity across all submissions. The document stipulates precise margin measurements, typeface specifications, and font sizes reflective of standard academic practices:

• Body text in 11-point Times Roman.
• Section and subsection titles are formatted in bold, with the former in 12-point and the latter in 11-point font.
• Titles should be centered in 15-point bold font.

Manuscript content is confined to predefined page limits: long papers can contain up to eight pages, with short papers limited to four, excluding references and statements on ethics or broader impact. Notably, an additional content page may be included for revisions following peer review feedback. This ensures that authors have the flexibility to address reviewers' comments adequately.

Submission Protocols

The paper outlines the multi-tiered submission structure, emphasizing compliance with double-blind review protocols that require anonymity—authors must omit personal identifiers and avoid self-referential citations that could reveal their identities. It highlights the potential rejection of non-compliant submissions, underscoring the importance of adherence to these guidelines.

Additional Material and Anonymity

Authors are encouraged to submit supplementary materials pertinent to their research, such as detailed appendices or additional datasets. However, the document cautions against integrating these materials within the primary manuscript, reserving detailed investigational nuances for external documentation. Supplementary materials must also conform to anonymity standards, a critical aspect of maintaining the integrity of the double-blind review process.

Formatting Integrity and Tool Support

The provision of \LaTeX{} style files significantly aids authors in maintaining compliance with formatting requirements. The document advises on specific \LaTeX{} commands essential for visual consistency, including the integration of page numbers and document structure for initial submissions. This typographic and structural guidance ensures that the visual presentation is standardized across manuscripts, which is crucial for peer review and subsequent publication.

Implications and Future Directions

The stipulations set forth in this instructional paper are crucial for fostering academic discipline and order in the presentation of research at ACL-IJCNLP 2021. These guidelines help ensure clarity, professionalism, and accessibility in academic writing. By adapting and refining established practices, authors are encouraged to present their research with precision and rigor.

Furthermore, the procedural rigor enforced by these guidelines suggests future enhancements in manuscript preparation and submission processes, potentially integrating more automated formatting checks and user-friendly template tools, to facilitate compliance and reduce the administrative burden on authors.

In conclusion, the meticulous attention to style and structure embodied in this document is an essential aspect of academic publishing, reflecting not only on the individual works submitted but also on the overall quality and professionalism of the field at large. These guidelines prepare authors to engage with global standards in academic discourse while contributing innovative ideas and findings to the scientific community.