ACL主席Emily M. Bender在泰国曼谷举办的ACL2024会议上发表了一篇演讲，她提出ACL不应被视为一个专注于人工智能的会议，而是一个专注于计算语言学和自然语言处理的学术平台。Bender强调了ACL的核心在于语言技术和计算语言学，而不是人工智能。她认为，尽管机器学习，包括深度学习，为这些领域提供了许多有用的技术，但当焦点转移到人工智能时，可能会导致研究实践的扭曲。
Bender批评了人工智能领域的一些不良研究实践，如不当使用基准测试、要求与最先进技术（SOTA）的封闭模型进行比较评估、数据集过大导致缺乏保留数据等。她认为，如果研究问题集中在“如何证明我的机器具有智能”上，可能会扭曲研究实践。她还指出，人工智能领域的焦点导致了糟糕的审稿实践，例如不使用大型语言模型（LLMs）或不提供最先进大小的LLM结果的论文可能会被认为不感兴趣。
相反，Bender认为计算语言学和自然语言处理领域的研究最佳实践包括对技术的适用性、对人类语言行为的理解、良好界定的评估、内在和外在的评估、坚实的基线、保留的测试数据和详细的错误分析。她强调，计算语言学和自然语言处理研究建立在对数据的理解上，包括对语言如何工作的知识（即语言学）和数据集文档。
Bender的演讲在社交平台上引起了很大争议。有人认为这是一种不够包容的表现，因为NLP历史上最好的时刻发生在人们对其他学科的想法持开放态度。还有人认为，做这种分割完全没有必要，因为人工智能和计算语言学已经有机地融合在一起。不过也有人表示理解，毕竟人工智能风头太盛，一旦某个会议被AI论文“承包”，其他领域的研究势必受到冷落。
对于这个演讲传递的信号，大家也开始猜测：这是不是意味着ACL不欢迎人工智能论文了？然而，Bender并没有明确表示ACL不欢迎人工智能论文，而是强调了会议应是一个关注语言技术的场所，一个促进跨学科研究的社区，一个关心语言群体的研究领域，并且是一个我们可以理性讨论我们研究和技术对社会影响的空间。因此，ACL将继续作为一个包容性强的平台，支持计算语言学和自然语言处理领域的多样性和创新。
英语如下：
News Title: “ACL Chair Questions: Is AI Conference Closed or a Wake-Up Call?”
Keywords: ACL Chair, AI Conference, Closed Nature
News Content:
At the ACL2024 conference held in Bangkok, Thailand, Emily M. Bender, the Chair of ACL, delivered a speech in which she proposed that ACL should not be seen as an AI-focused meeting but rather as an academic platform dedicated to computational linguistics and natural language processing. Bender emphasized that the core of ACL lies in language technology and computational linguistics, rather than artificial intelligence. She argued that while machine learning, including deep learning, has provided numerous useful techniques to these fields, shifting the focus to artificial intelligence could lead to a distortion of research practices.
Bender criticized some unfavorable research practices in the AI field, such as improper use of benchmark tests, requiring comparisons and evaluations with closed models against state-of-the-art (SOTA) technology, and the use of overly large datasets that lack retained data. She believed that if research questions were centered around “how to prove my machine is intelligent,” it could distort research practices. She also pointed out that the focus in the AI field has led to poor peer review practices, such as papers that do not use large language models (LLMs) or do not provide results from state-of-the-art-sized LLMs being considered uninteresting.
On the contrary, Bender believed that the best practices in research for the fields of computational linguistics and natural language processing include applicability of technology, understanding of human language behavior, well-defined assessments, internal and external evaluations, solid baselines, retained test data, and detailed error analysis. She stressed that research in computational linguistics and natural language processing is built on an understanding of data, including knowledge of how language works (i.e., linguistics) and documentation of datasets.
Bender’s speech sparked significant controversy on social platforms. Some saw it as an unacceptably exclusionary stance, noting that the best moments in NLP history have occurred when the field has been open to ideas from other disciplines. Others felt that such a division was unnecessary, as AI and computational linguistics have already organically integrated. Nonetheless, there were those who understood the sentiment, given that AI’s prominence could lead to the neglect of other fields’ research if a conference became dominated by AI papers.
As for the signal the speech was sending, speculation began: Was it signaling that ACL was not welcoming to AI papers? However, Bender did not explicitly state that ACL was not welcoming to AI papers, but rather emphasized that the conference should be a venue focused on language technology, a community that promotes interdisciplinary research, a field that cares about the language community, and a space where we can rationally discuss the impact of our research and technology on society. Therefore, ACL will continue to serve as a platform that is inclusive and supportive of the diversity and innovation in the fields of computational linguistics and natural language processing.
【来源】https://www.jiqizhixin.com/articles/2024-08-15-3
Views: 1