PFI-TT: Using Artificial Intelligence (AI) and Natural Language Processing (NLP) to Enhance Writing for Publication

“创新伙伴关系-技术翻译”（PFI-TT）项目的更广泛影响/商业潜力是，为新研究作者（例如，职业生涯早期的研究人员和学生）提供更多机会，让他们获得高水平的、响应性的科学写作培训，从而为接受资助申请和发表同行评议的期刊文章提供一个更加多样化、公平和包容的竞争环境。研究机构往往难以为他们的新手作家提供有意义的支持。这种支持通常需要极其昂贵的人力资源来提供关于科学写作的技术期望的培训，而且至关重要的是，还要对写作进行评估和提供反馈。然而，研究机构经常为解决社区内主要痛点的新教育技术提供开放预算，为包含基于语言的科学写作教学课程的软件创造强大的产品市场。所提出的技术经过训练，可以识别并提供论证级反馈。因此，预计这项技术将提高促进写作的能力，抵消培训负担，并增加代表性不足的人口获得资源的机会，所有这些都集中在一个工具中，通过提供反应迅速的按需指导和关于出版写作的反馈。该项目涉及两个主要的研究机会：实现高级自动化写作反馈和使用神经网络进行文本分类和反馈生成。到目前为止，现有的提供自动反馈的写作支持工具只能处理表层的语法和机制。从理论的角度来看，神经网络在文本分类和反馈生成方面的研究还不够充分，尤其是在学术写作方面。该项目预计将以两种主要方式推进知识的发展：(1)通过专注于科学论证和描述的特定语言，展示如何在表面语法和机制的反馈之外生成自动反馈；(2)通过强调如何使用模式标记和机器学习的深度神经网络方法来生成特定于同行评审期刊文章和拨款申请的反馈。这些目标将通过开发一个自动写作反馈引擎来实现，该引擎将集成到技术中，然后使用基于论证的框架进行验证，以评估和提高引擎的准确性和可用性。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this Partnerships for Innovation - Technology Translation (PFI-TT) project is greater access for novice research writers (e.g., early-career researchers and students) to high-level, responsive training in scientific writing and, as a result, a more diverse, equitable and inclusive playing field for acceptance of grant applications and publication of peer-reviewed journal articles. Research institutions often struggle to provide meaningful support for their novice writers. Such support typically requires prohibitively expensive human resources to provide training on the technical expectations of scientific writing and, crucially, to evaluate and provide feedback on writing. However, research institutions frequently have open budgets for new educational technologies that address major pain points within their community, creating a strong product-market fit for a software that includes a linguistically-grounded instructional curriculum for scientific writing. The proposed technology is trained to recognize and provide argument-level feedback. As such, the technology is expected to increase the capacity to advance writing, offset training burden, and enhance access to resources for underrepresented populations, all in one tool, by providing responsive, on-demand instruction and feedback on writing for publication. The project addresses two main research opportunities: the implementation of advanced automated writing feedback and the use of neural networks for text classification and feedback generation. To date, existing writing support tools that provide automated feedback address only surface-level grammar and mechanics. From a theoretical perspective, neural networks are underexplored for text classification and feedback generation, especially for scholarly writing. This project is expected to advance knowledge in two primary ways: (1) by demonstrating how automated feedback can be generated beyond feedback on surface-level grammar and mechanics through a focus on language specific to scientific argumentation and description and (2) by highlighting how pattern markup and a deep neural network approach to machine learning can be used to produce feedback specific to peer-reviewed journal articles and grant applications. These aims will be achieved by developing an automated writing feedback engine that will be integrated into the technology and then validated using an argument-based framework to evaluate and enhance the engine’s accuracy and usability.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.