Paper proposes Natural Language Programming, treating prompts as cognitive artefacts
An arXiv paper argues that large language models turn natural language from an object of analysis into an executable medium, and frames prompting as a form of cognitive programming.
In plain words
A researcher has published a theoretical paper arguing that prompting large language models should be studied as a new kind of programming. Instead of writing formal code, people express goals, constraints and reasoning steps in plain language, and the conversation itself acts like an external working memory. The paper is conceptual and does not present experiments or a working system.
What was announced
The paper, posted on arXiv, introduces Natural Language Programming (NLPg) as a theoretical extension of classical natural language processing. Where traditional NLP treated language as something to analyse, classify, translate or generate, the author argues that large language models let natural language act as an executable medium through which people structure reasoning, orchestrate problem solving and manage hybrid human/AI workflows.
Within this framing, prompts are described as dynamic cognitive artefacts that externalise goals, constraints, assumptions, validation criteria and intermediate reasoning states. Drawing on distributed cognition, extended mind, metacognition and dialogic learning, the paper proposes Dialogic Working Memory, in which prompts and conversational states serve as temporary external memory supporting reasoning continuity. It also maps existing memory architectures, named BTMA, SKGE and RMR, onto this framework, saying their episodic, semantic and metacognitive layers support long-horizon conversational coherence. The author presents this as a step toward a post-symbolic, post-interface paradigm in which language is simultaneously communication, executable specification, cognitive scaffold and distributed memory.
What it does not show
- The evidence is the paper's abstract only; the full article, including how the framework is argued or validated, is not available for review.
- The paper is theoretical and epistemological; the abstract reports no empirical evaluation, implementation or independent confirmation of the proposed framework.
- The memory architectures BTMA, SKGE and RMR are referenced by acronym without descriptions in the available text, so their nature cannot be assessed.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.