Scholarlink

<p>Legal sign systems operate on a foundational semiotic assumption: that legal meaning can be constituted, communicated, and enforced through propositional language. The patent claim is among the most precise instantiations of this assumption, functioning as a legal sign that simultaneously signifies an invention and delimits the legal monopoly it attracts. For most technologies, this sign-object relationship functions adequately. For trained artificial neural networks (ANNs), however, it breaks down in a structurally revealing way. The technically operative core of a trained ANN, its weights, emerges stochastically through training and resists capture in propositional legal language. The claim can signify the architecture but not the trained model. A semiotic gap opens between what the legal sign can say and what the invention actually is. This article diagnoses that gap through two philosophical frameworks. Kant’s distinction between <i>phenomena</i> and <i>noumena</i> identifies the weights as epistemologically inaccessible to propositional knowledge before and during training. Wittgenstein’s distinction in the <i>Tractatus Logico-Philosophicus</i> between what can be said and what can only be shown identifies the weights as resistant to propositional signification even after training. Together, these frameworks reveal the weights gap not as a failure of legal drafting but as a structural limit of legal sign systems encountering a <i>noumenal</i>-unsayable object. The article traces this semiotic breakdown across four jurisdictions, the United Kingdom, the European Patent Office, the United States, and India, demonstrating that the same sign-system failure appears in each, wearing different doctrinal clothing. It concludes with implications for how legal semiotics might reframe patent law’s response to technological objects that can be shown but not said.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

What Patent Claims Cannot Say: A Kant–Wittgenstein Framework for the Patentability of Neural Networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月19日

查看全文
  • Gaurav Dahiya

摘要

Legal sign systems operate on a foundational semiotic assumption: that legal meaning can be constituted, communicated, and enforced through propositional language. The patent claim is among the most precise instantiations of this assumption, functioning as a legal sign that simultaneously signifies an invention and delimits the legal monopoly it attracts. For most technologies, this sign-object relationship functions adequately. For trained artificial neural networks (ANNs), however, it breaks down in a structurally revealing way. The technically operative core of a trained ANN, its weights, emerges stochastically through training and resists capture in propositional legal language. The claim can signify the architecture but not the trained model. A semiotic gap opens between what the legal sign can say and what the invention actually is. This article diagnoses that gap through two philosophical frameworks. Kant’s distinction between phenomena and noumena identifies the weights as epistemologically inaccessible to propositional knowledge before and during training. Wittgenstein’s distinction in the Tractatus Logico-Philosophicus between what can be said and what can only be shown identifies the weights as resistant to propositional signification even after training. Together, these frameworks reveal the weights gap not as a failure of legal drafting but as a structural limit of legal sign systems encountering a noumenal-unsayable object. The article traces this semiotic breakdown across four jurisdictions, the United Kingdom, the European Patent Office, the United States, and India, demonstrating that the same sign-system failure appears in each, wearing different doctrinal clothing. It concludes with implications for how legal semiotics might reframe patent law’s response to technological objects that can be shown but not said.