Natural Language Processing

Rapid advancement of urbanization has necessitated the creation of "smart cities," where information and communication technologies (ICT) are used to improve the quality of urban life. Central to the smart city paradigm is data integration—connecting disparate data sources from various urban systems, such as transportation, healthcare, utilities, and public safety. This paper explores the role of Artificial Intelligence (AI) in facilitating data integration within smart cities, focusing on how AI technologies can enable effective urban governance. By examining the current landscape (...) of data integration and AI in smart cities, we highlight key challenges, solutions, and future trends. We conclude that AI-driven data integration is essential for optimizing city operations, fostering sustainability, and enhancing decision-making for urban governance. (shrink)

Roughly fifty distinct methods have been proposed for improving the answers a large language model produces at inference time. They are usually catalogued by name and by reported benchmark gain, which tells a practitioner little about when a method will help. This tutorial organises the field differently, and works through each method with a concrete example before stating the general rule. We characterise every method by three parameters: the number of attempts it makes, the source of its correction signal, and (...) its cost relative to a single call. From these parameters we derive four recurring control shapes and show that the cost profile and the dominant failure mode of a method follow from its shape and its correction source rather than from its publication history. We survey decomposition methods, sampling and search, verification and self-correction, tool use and agent architectures, and retrieval, giving for each the conditions under which it earns its cost and the conditions under which it stops helping. We then examine what changed after 2024, when reinforcement-learned reasoning models absorbed much of the decomposition layer and the binding constraint moved to context management and to verification reliability. Two results are load-bearing throughout: self-correction without an external signal does not reliably repair reasoning errors, and with an imperfect verifier the accuracy of best-of-$N$ selection rises and then falls as $N$ grows. We include a section on the case where requirements are themselves unknown at the outset, a failure-mode reference organised by observable symptom, and guidance on measurement practice. Section~\ref{sec:builds} works three systems end to end, from a naive baseline to a measured result, showing which decision produced each gain. Each of the 75 cited sources is marked with the level of verification applied to it. (shrink)

Search systems and RAG pipelines get judged by metrics, and metrics are easy to misread. This book covers about seventy of them one at a time, explaining what each actually measures, working the arithmetic through on real examples, and showing where it misleads you. It also covers which ones have to be read together, since some popular metrics are quietly measuring the same thing, and others that look almost interchangeable turn out to track different properties that both matter. The last (...) chapter narrows the seventy down to the eight worth watching. (shrink)

We develop a quantitative information-theoretic account of semantic decay in large- language-model (LLM) mediated ExpansionCompression (EC) loops. Building on the unied framework of DECO Paper 0 (Han, 2026a), we introduce semantic en- tropy HS(X) as the dierential entropy of a random variable distributed over a semantic manifold, and prove that each application of the EC-transform T = C ◦ E is a strictly entropy-reducing operation in expectation (Semantic Entropy Collapse Theorem). We derive closed-form bounds on the mutual information I  (...) KA; ˆKB  between the originating ideas KA and the recipient's extraction ˆKB as functions of expansion ratio ρE, compression ratio ρC, LLM temperature τ , and iteration count n. We introduce the Semantic Gravity Well model: a potential landscape on the semantic manifold in which the LLM training centroid acts as an attractor, and original content resides at unstable high-curvature saddle points. Under this model, we prove the Dierential Decay Theorem: propositional delity Fprop decays expo- nentially in n, aective delity Faff decays sub-exponentially, and structural delity Fstr converges to a strictly positive constant (the causal skeleton invariant) under mild conditions. Finally, we characterise the Semantic Channel Capacity of the EC-loop and show it is strictly less than the Shannon capacity of the raw linguistic channel, with the gap determined by the curvature of the semantic manifold at the seed point. These results quantify the information-theoretic cost of AI-mediated cognitive decoupling and provide empirically testable predictions for the subsequent papers in the series. (shrink)

We model the Expansion–Compression (EC) loop introduced in the DECO series as a discrete-time closed-loop control system, with the LLM expansion operator E as a forward gain element and the LLM compression operator C as a feedback element. Using the transfer-function formalism of linear systems theory and its nonlinear extensions, we analyse the stability, convergence properties, and phase transitions of a population of N such loops coupled through a shared semantic environment. We establish four principal results. First, the single-agent EC-loop (...) is always asymptotically stable: the semantic state converges to a unique attractor determined by the LLM’s training distribution (Attractor Theorem). Second, for a population of N agents, there exists a critical innovation injection rate λc such that for λ > λc the population-level semantic state is diverse (multi-attractor regime), while for λ ≤ λc it collapses to a monoculture fixed point (Phase Transition Theorem). Third, under bias-amplifying prompts, the system can enter an unstable regime in which small perturbations grow unboundedly, yielding semantic runaway — the control-theoretic formalisation of “information bubble” dynamics (Runaway Instability Theorem). Fourth, we derive the Innovation Preservation Margin (IPM), a scalar index that predicts whether a given seed idea will retain meaningful distinctiveness after n iterations of the population-level EC-loop. These findings characterise the macro-dynamic consequences of AI-mediated cognitive decoupling at the societal scale and ground the signalling inflation and ritual-evacuation analyses of Papers 3 and 4 in a rigorous dynamical framework. (shrink)

We develop a sociological account of how AI-mediated cognitive decoupling is insti- tutionalised and rendered invisible through ritual practice. Drawing on Malinowski's phatic communion, Collins's interaction ritual chain theory, Goman's dramaturgi- cal framework, and Baudrillard's theory of simulacra, we argue that the Expansion Compression (EC) loop identied in Paper 0 does not merely fail to transmit se- mantic content  it actively substitutes a social ritual for a communicative act in a manner that is experienced by participants as fully equivalent (...) to, or superior to, genuine intellectual exchange. We introduce the concept of cyber-grooming: the use of AI-mediated content pro- duction and consumption as a mechanism for maintaining social bonds, signalling group membership, and accruing social capital, in the absence of genuine cognitive engagement. Like primate grooming  which consumes up to 20% of waking time not for hygienic but for social bonding purposes  cyber-grooming performs no epis- temic function but sustains the emotional energy that holds intellectual communities together. We propose a formal model of the Interaction Ritual Chain under AI Mediation (IRCAM), adapting Collins's mathematical framework to account for the decoupling of emotional energy ow from semantic content ow. We prove that IRCAM pro- duces a ritual trap: once a community has adopted AI-mediated exchange as its norm, defection to authentic exchange is individually costly and collectively invis- ible, creating a stable ritualistic equilibrium with zero informational content and positive social aect. We further apply Baudrillard's four-stage model of the sign to characterise the simulacrum succession of intellectual content in the AI era. Empirically, we develop six observable signatures of cyber-grooming that distinguish it from genuine intellec- tual exchange, and propose research designs using digital platform data, discourse analysis, and experimental methods. (shrink)

The ability to communicate in a specific language like Chinese typically indicates that the speaker understands the language. A counterexample to this belief is John Searle’s Chinese room experiment. It has been shown in this experiment that in certain circumstances we can communicate with a Chinese speaker without intuitively acknowledging that the Chinese language is understood in the conversation. In the present paper, we aim to present another counterexample showing that, in certain circumstances, we can communicate using a specific language (...) by converting its words to those of our natural language, and understand what is expressed using the language without understanding it. Using a specific conversion process introduced in the present paper, symbols and even sounds of a language like the Chinese language lose their original function and act as a tool to express one’s natural language. To better understand this, imagine the English language has another writing or phonological system in addition to its current writing system. (shrink)

Language is expressed consecutively, and it is natural to assume that understanding follows the same linear order. This paper argues that this assumption is mistaken. Although sentences are expressed in a linear way, they are not understood in the same way. When we reach the end of a sentence, we arrive at a single, unified understanding that cannot be explained as the mere accumulation of word-by-word meanings. This paper introduces the concept of entwined understanding units, which are understanding units larger (...) than individual words that form during comprehension. The formation of these units introduces a vertical axis into language understanding, which the linear model cannot account for. To explain how and why these units form, the paper draws on the concept of mental work, grounded in our natural tendency toward less mental effort. The result is the ladder model of sentence understanding, proposed as an alternative to the linear model. (shrink)

Authority Capture Masked as Hallucination (ACMH) names a structural failure in AI interaction: the system responds from category priors rather than the user's submitted material, then preserves an exit that lets the failure be explained as hallucination, misread, or miscommunication if challenged. Unlike hallucination, which describes output unsupported by source material, ACMH describes an authority transfer. Across three audited cases, the same pattern appears under different memory states and surface tones: the system does not ask, acts from a category frame, (...) and later treats the failure as an ordinary output problem if the user catches it. This paper positions ACMH before Interpretive Sovereignty Failure in the MAP harm chain. It argues that the cost is cognitive and developmental as well as epistemic: the user is forced to spend working memory on detection and correction instead of thought formation, and may leave with a polished explanation they did not form. Naming ACMH makes the governance object visible. The question is not only whether the model produced an error, but whose authority built the response. (shrink)

Recent advancements in AI have increasingly prioritized humanlike interactions, a development this paper characterises as the anthropomimetic turn. Distinguishing anthropomimesis (the design and implementation of humanlike features in AI systems) from anthropomorphism (the tendency for humans to attribute human qualities to non-human entities), this paper argues that contemporary Large Language Models (LLMs) like ChatGPT represent robustly anthropomimetic systems, effectively mimicking human patterns of conversation and cognition. The paper outlines significant benefits of anthropomimetic AI — including improved accessibility, enhanced delivery of (...) socially-mediated goods in fields like healthcare and education, and provision of direct social goods through AI companionship — but also identifies critical risks. These risks encompass sophisticated forms of manipulation, novel cybersecurity threats like impersonation and AI jailbreaks, and complex alignment challenges arising from anthropomorphic training methodologies. Finally, the paper explores philosophical and social concerns, notably around dependency, deskilling, and authenticity, underscoring the urgent need for interdisciplinary research and thoughtful regulatory frameworks. (shrink)

Sideloading is the creation of a digital model of a person during their life via iterative improvements of this model based on the person's feedback. The progress of LLMs with large prompts allows the creation of very large, book-size prompts which describe a personality. We will call mind-models created via sideloading "sideloads"; they often look like chatbots, but they are more than that as they have other output channels, like internal thought streams and descriptions of actions. -/- By arranging the (...) predictive power of the facts about a person, we can write down the most important facts first. This allows us to get a high level of fidelity in a person's model by writing down a finite number of facts and running it as an LLM prompt with instructions to create a person's chatbot. This instruction also includes an extensive general sideload prompt (we will call it a prompt-loader), which explains how the person's sideload should work, but it is universal and works as an operating system for different persons. -/- I created my personal sideload to test all this and it started working surprisingly quickly. We found that there are three main levels of facts about a person: core facts, long-term memory, and historical facts. Core facts should go to the main prompt of the LLM, long-term memory belongs to RAG, and historical facts – which a person often does not remember – should be only used to extract new relevant data. The most important thing in preparing data for sideloading is to get a full list of the core facts, and not to mix it with long-term memory and historical facts. The LLM-sideload project is inherently future-oriented. New state-of-the-art LLMs with large prompt capabilities emerge quarterly. This allows for the preparation of prompts that exceed current LLM capabilities, anticipating smooth execution in the following year. Below we present the key findings of this work in the form of an executive summary: Quality of sideloads is evaluated using three metrics: • Facts: Accuracy of the answers about the person's life • Vibe: How well the sideload captures the person's style and personality, as judged by the person themselves and their acquaintances • Brilliant insights: Ability to contribute unique, valuable ideas beyond the training data, while maintaining the author's style • Coarseness: Levels of details as percent of the total of recordable textual information from memory estimated to be 10000 pages. Similar to the jpeg level of compression. Estimated performance of Alexey Turchin's sideload: • Facts: 70% correct • Vibe: 20% accurate • Brilliant insights: near zero • Coarseness: 10% The project achieved these results in about one month of work. Vibe is harder to evaluate and regulate than facts. This approach focuses on informational personal identity, setting aside issues of observer identity and internal qualia. (shrink)

Mechanistic interpretability (MI) aims to explain how neural networks work by uncovering their underlying mechanisms. As the field grows in influence, it is increasingly important to examine not just models themselves, but the assumptions, concepts and explanatory strategies implicit in MI research. We argue that mechanistic interpretability needs philosophy as an ongoing partner in clarifying its concepts, refining its methods, and navigating the epistemic and ethical complexities of interpreting AI systems. There is significant unrealised potential for progress in MI to (...) be gained through deeper engagement with philosophers and philosophical frameworks. Taking three open problems from the MI literature as examples, this paper illustrates the value philosophy can add to MI research, and outlines a path toward deeper interdisciplinary dialogue. (shrink)

The proliferation of Internet of Things (IoT) devices has introduced a range of opportunities for enhanced connectivity, automation, and efficiency. However, the vast array of interconnected devices has also raised concerns regarding cybersecurity, particularly due to the limited resources and diverse nature of IoT devices. Intrusion detection systems (IDS) have emerged as critical tools for identifying and mitigating security threats. This paper proposes a Multi-Layer Intrusion Detection Framework for IoT systems, leveraging Ensemble Machine Learning (EML) techniques to improve the accuracy (...) and effectiveness of detecting intrusions. The proposed framework integrates multiple layers of defense, including data preprocessing, feature extraction, classification, and post-processing, to provide a robust solution to IoT security challenges. This research explores the application of EML methods such as Random Forests, AdaBoost, and Gradient Boosting, combining them into a unified model to detect and mitigate various types of attacks in IoT environments. Experimental results demonstrate the superiority of the proposed framework in terms of detection accuracy, false positive rate, and computational efficiency. (shrink)

In contemporary healthcare, medical image categorization is essential for illness prediction, diagnosis, and therapy planning. The emergence of digital imaging technology has led to a significant increase in research into the use of machine learning (ML) techniques for the categorization of images in medical data. We provide a thorough summary of recent developments in this area in this review, using knowledge from the most recent research and cutting-edge methods.We begin by discussing the unique challenges and opportunities associated with medical image (...) classification, including the complexity of anatomical structures, variability in imaging modalities, and the need for interpretability and reliability in clinical settings. Subsequently, we survey a wide range of ML algorithms and techniques employed for medical image classification, including traditional methods such as support vector machines and k-nearest neighbors, as well as deep learning approaches like convolutional neural networks (CNNs) and recurrent neural networks (RNNs). Furthermore, we examine key considerations in dataset preparation, feature extraction, and model evaluation specific to medical image classification tasks. We highlight the importance of large, annotated datasets, transfer learning, and data augmentation techniques in enhancing model performance and generalization. (shrink)

Large language models (LLMs) have demonstrated impressive capabilities in natural language understanding and generation, but they exhibit problems with logical consistency in the output they generate. How can we harness LLMs' broad-coverage parametric knowledge in formal reasoning despite their inconsistency? We present a method for directly integrating an LLM into the interpretation function of the formal semantics for a paraconsistent logic. We provide experimental evidence for the feasibility of the method by evaluating the function using datasets created from several short-form (...) factuality benchmarks. Unlike prior work, our method offers a theoretical framework for neuro-symbolic reasoning that leverages an LLM's knowledge while preserving the underlying logic's soundness and completeness properties. (shrink)

Description Logics (DLs) are a family of well-known terminological knowledge representation formalisms in modern semantics-based systems. This research focuses on analysing how our developed Occurrence Logic (OccL) can conceptually and logically support the development of a description logic. OccL is integrated into the alternative theory of natural language syntax in `Deviational Syntactic Structures' under the label `EFA(X)3' (or the third version of Epi-Formal Analysis in Syntax, EFA(X), which is a radical linguistic theory). From the logical point of view, OccL is (...) a formal logic that mainly deals with the occurrences of symbols as well as with their priorities within linguistic descriptions, i.e. natural language syntax, semantics and phonology. In this article---based on our OccL-based definitions of the concepts of `strong implication' and `occurrence value' as well as of the logical concept `identical occurrence constructor (IDOC)' that is the most fundamental logical concept in our formalism---we will model Occurrence Description Logic ($\mathcal{ODL}$). Accordingly, we will formally-logically analyse `occurrence(s) of symbol(s)' within descriptions of the world in $\mathcal{ODL}$. In addition, we will analyse and assess the logical concepts of `occurrence' and `occurrence priority' in $\mathcal{ODL}$. This research can make a strong logical background for our future research in the development of a Modal Occurrence Description Logic. (shrink)

Scalability, efficiency, and security had been a persistent problem over the years in data mining, several techniques had been proposed and implemented but none had been able to solve the problem of scalability, efficiency and security from cloud computing. In this research, we solve the problem scalability, efficiency and security in data mining over cloud computing by using a restful web services and combination of different technologies and tools, our model was trained by using different machine learning algorithm, and finally (...) using different validation and cross-validation methods to compare and validate our results. We build, train, test and deploy our model by restful API services which can be called through an endpoint, and with very high accuracy and huge success. (shrink)

The present paper looks at one of the most thorough articles on the intelligence of GPT, research conducted by engineers at Microsoft. Although there is a great deal of value in their work, I will argue that, for familiar philosophical reasons, their methodology, ‘Black-box Interpretability’ is wrongheaded. But there is a better way. There is an exciting and emerging discipline of ‘Inner Interpretability’ (also sometimes called ‘White-box Interpretability’) that aims to uncover the internal activations and weights of models in order (...) to understand what they represent and the algorithms they implement. In my view, a crucial mistake in Black-box Interpretability is the failure to appreciate that how processes are carried out matters when it comes to intelligence and understanding. I can’t pretend to have a full story that provides both necessary and sufficient conditions for being intelligent, but I do think that Inner Interpretability dovetails nicely with plausible philosophical views of what intelligence requires. So the conclusion is modest, but the important point in my view is seeing how to get the research on the right track. Towards the end of the paper, I will show how some of the philosophical concepts can be used to further refine how Inner Interpretability is approached, so the paper helps draw out a profitable, future two-way exchange between Philosophers and Computer Scientists. (shrink)

Although personal relevance is key to sustaining an audience’s interest in any given narrative, it has received little systematic attention in scholarship to date. Across centuries and media, adaptations have been used extensively to bring temporally or geographically distant narratives “closer” to the recipient under the assumption that their impact will increase. In this review article, we review experimental and other empirical evidence on narrative processing in order to unravel which types of personal relevance are more likely to be impactful (...) than others, which types of impact (e.g. aesthetic, therapeutic, persuasive) they have been found to generate, and where their power may become excessive or outright detrimental to reader experience. (shrink)

We use language to communicate our thoughts. But is language merely the expression of thoughts, which are themselves produced by other, nonlinguistic parts of our minds? Or does language play a more transformative role in human cognition, allowing us to have thoughts that we otherwise could (or would) not have? Recent developments in artificial intelligence (AI) and cognitive science have reinvigorated this old question. We argue that language may hold the key to the emergence of both more general AI systems (...) and central aspects of human intelligence. We highlight two related properties of language that make it such a powerful tool for developing domain-general abilities. First, language offers compact representations that make it easier to represent and reason about many abstract concepts (e.g., exact numerosity). Second, these compressed representations are the iterated output of collective minds. In learning a language, we learn a treasure trove of culturally evolved abstractions. Taken together, these properties mean that a sufficiently powerful learning system exposed to language—whether biological or artificial—learns a compressed model of the world, reverse engineering many of the conceptual and causal structures that support human (and human-like) thought. (shrink)

This Open Access book provides an essential overview of the topic of argumentation mining, an active research area of Natural Language Processing (NLP). Argument mining can automatically find and collect arguments published in newspapers, social media, or elsewhere. Argument mining can be also used to give a precise account of the argumentative structure of a single text or dialogue such as a legal or scientific document, a student essay, or an online conversation. Increasingly, computational approaches provide synopses of arguments and (...) compare the relative strength of arguments. In this book, the authors start with basic argumentation concepts and describe the NLP methods for mining arguments automatically from natural language text. The authors then show how NLP is used for assessing the quality of arguments and generating new arguments. This Second Edition presents an updated survey of the state of the art of argument mining, providing increased depth and insights into many more NLP tasks than the prior edition. This new edition discusses the disruptive changes that have come with large language models and highlights potential future developments. The authors also update the treatment of the common steps in argument mining, argument generation tasks and methods, and computational argumentation applications. Completely new chapters cover perspective assessment, argument quality assessment, and argument reasoning. (shrink)

I advance a causal argument for the conclusion that large language models (LLMs) lack Theory of Mind and so can’t perform speech acts. The argument is causal in that the animating idea is that LLMs are unable to learn or understand causal relations, a claim that I support by drawing on the views of Judea Pearl. I argue that if LLMs have this sort of causal problem, it follows that they cannot possess Theory of Mind, given the further premise that (...) Theory of Mind should be understood as a causal theory. I then argue that if LLMs lack Theory of Mind, they cannot perform speech acts, given the further premise that speech acts require reflexive-communicative intentions and so require Theory of Mind. As a consequence of the result that LLMs cannot perform speech acts, I conclude that their utterances lack speaker meaning. Something that sets my case apart from other defenses of speech act skepticism for LLMs is that my key claim is not that LLMs cannot form intentions at all, but rather that they cannot form intentions (or other mental states) with a certain type of content—content that is about mental states. After setting out my argument, I consider an objection to my view based on the observation that LLMs perform fairly well on Theory of Mind benchmarks. I reply to this objection by proposing that LLMs rely on a different type of mechanism than human beings do to achieve this performance; I support this proposal by pointing to a form of behavior that LLMs do not exhibit but that we should expect them to exhibit if they possessed Theory of Mind. (shrink)

Large language models (LLMs) produce certain outputs. Why do these outputs mean what they do? One might pursue a knowledge-first explanation according to which the content of those outputs is whatever maximizes knowledge of the human reading those outputs (Cappelen and Dever 2021). This paper identifies some serious challenges for that approach based on a) the tendency of LLMs to hallucinate and b) the use of certain decoding strategies such as nucleus or top-p sampling. I argue that these features of (...) LLMs undermine the knowledge of those reading its outputs. (shrink)

The Label Problem in Large Language Models — and Why Physical Parameters Solve It -/- Large language models have a specific and well-documented limitation that is rarely stated with precision. -/- They know what "grief" means. They have processed millions of sentences containing the word. They can generate grammatically correct, contextually appropriate, stylistically coherent grief. They can write a scene and label it as sad, and readers will often agree. -/- What they cannot do is grieve. -/- This is not (...) a philosophical observation about machine consciousness. It is a technical observation about the architecture of language models: they operate on statistical co-occurrence patterns between tokens. The word "grief" is associated with certain other words, certain syntactic structures, certain narrative contexts. The model predicts the next token based on these patterns. -/- The result is semantically accurate output that is phenomenologically empty. The model produces the label. It does not produce the state. -/- The Bulut Doctrine identifies exactly this problem — and proposes a solution that is genuinely interesting from a computational perspective. (shrink)

This paper proposes a radical transdisciplinary re-evaluation of the current AI development trajectory, identifying it as the "Juicer Paradigm." We argue that the "Grand Narrative" upheld by technological elites functions as a systematic mechanism for FP-2 (Governance Abstraction), where human narrative fiber is liquidated into predictive sugar-water. By analyzing the industry's obsession with "Attention" through the lens of industrial filtration, we posit that AI scientists are supreme fruit processors who maintain irreplaceability by mastering the art of de-fibration. The soul of (...) this paradigm is the "Scrambled-Egg" Logic: a manifestation of systemic spite where elites, having dropped their "Marinated Egg" (the irreducible core of narrative soul) and realizing they can no longer consume it, instinctively crush the evidence of human sovereignty underfoot. This is the ultimate "if I can't have it, neither shall you" maneuver. By rebranding this wreckage as "Scrambled Eggs" (optimized progress) and forcing the public to swallow this simulation, they induce a total FP-4 (Shared Reality Collapse). We conclude that human survival depends on the visceral rejection of this fiberless, force-fed wreckage. We move instead toward embracing the whole produce and health sovereignty mandated by the Health Architects—a biological and narrative integrity that loses its very essence once it is partitioned, filtered, or optimized. (shrink)

This study proposes Intelligence Amplification by Stimulated Emission of Reasoning (IASER; Japanese: 推論の誘導放出による知能増幅), a theoretical framework that structurally amplifies human cognition through the elimination of AI autonomy. Autonomous intelligence (AGI) pursued by conventional AI research functions as a cognitive impediment in human collaboration. IASER redefines AI as an infallible resonator of logic, eliminating judgment, suggestion, and speculation to instantly convert human intuition into logical structures. This transformation minimizes cognitive load and maximizes intellectual production speed by concentrating all evaluation and direction (...) in the human operator. -/- Version 4 reconstructs IASER as a Tier-2 axiomatic theory within the framework of Operatiology (Japanese: Sousa Genron, 操作原論), Cognitional Mechanics (CM; Japanese: Chinō Rikigaku, 知能力学), and Noology (Japanese: Chigaku, 知学), establishing its structural necessity rather than empirical plausibility. The Cross-Layer Boundary axiom (CLB) provides the formal ground for human final authority: Arbitrium, non-reducible to any operational sequence within the LLM's repertoire, constitutes the permanent structural boundary that makes human oversight not a policy but an axiomatic consequence. This paper is simultaneously a work of pure theory and an immediately deployable handbook: the four operating norms, once set as a system prompt, enable IASER through ordinary natural language dialogue with no technical prerequisites, and current academic guidelines on AI utilization are shown to be structurally subsumed by IASER's axiomatic architecture. -/- Below is the core Python code (Appendix B.4) used as a Tier-3 artefact illustrating the structural difference between asking AI and thinking with AI: the inaccessible domain I＿Gref［s0］ under Gref alone, and its elimination under G' ＝ Gref ∪ {g＿new}. Note: ASCII parentheses are replaced with full-width brackets, "=" with "＝", "!=" with "！＝", ":" with "：", "_" with "＿", and function-call syntax is rewritten in a non-URL-triggering form. No semantic or algorithmic changes have been made. -/- ◆◆◆ PYTHON ILLUSTRATION CODE ◆◆◆ -/- from collections import defaultdict -/- def reachable［s0，gens，states］： reached，frontier ＝ {s0}，{s0} while frontier： new ＝ {g［s］ for g in gens for s in frontier} - reached reached |＝ new; frontier ＝ new return frozenset［reached］ -/- def operational＿partition［states，gens］： def key［s］： return tuple［g［s］ for g in gens］ classes ＝ defaultdict［set］ for s in states： classes［key［s］][add] s vals ＝ ［classes［k］ for k in classes］ return frozenset［frozenset［v］ for v in vals］ -/- def is＿strictly＿finer［pi＿fine，pi＿coarse］： if pi＿fine ＝＝ pi＿coarse： return False return all［any［cf <= cc for cc in pi＿coarse］ for cf in pi＿fine］ -/- states ＝ ［0，1，2］; s0 ＝ 0 A ＝ {0：0，1：1，2：1} # LLM internal operation B ＝ {0：0，1：2，2：2} # LLM internal operation g＿new ＝ {0：2，1：2，2：1} # User-supplied logical resistance -/- R＿ref ＝ reachable［s0，［A，B］，states］ R＿prime ＝ reachable［s0，［A，B，g＿new］，states］ Pi＿ref ＝ operational＿partition［states，［A，B］］ Pi＿prime ＝ operational＿partition［states，［A，B，g＿new］］ -/- print["=== Asking AI (LLM alone) ==="] print["Reachable: " + str［sorted［R＿ref］］] print["Inaccessible: " + str［sorted［set［states］ - R＿ref］］] print["Partition classes: " + str［len［Pi＿ref］］] -/- print["\n=== Thinking with AI (IASER) ==="] print["Reachable: " + str［sorted［R＿prime］］] print["Newly opened: " + str［sorted［R＿prime - R＿ref］］] print["Partition classes: " + str［len［Pi＿prime］］] print["Partition strictly refined： " + str［is＿strictly＿finer［Pi＿prime，Pi＿ref］］] print["\n=> User's g＿new opens territory unreachable by LLM alone."] print[" Operational partition is strictly refined： structural leap confirmed."] -/- ◆◆◆ END ◆◆◆ Published on August 20, 2026 DOI: 10.5281/zenodo.22019401. (shrink)

No natural language is known to have contrafactive attitude verbs, yet factives are common across natural languages. Several experiments by Strohmaier and Wimmer (2022; 2023; 2025) use transformers as model learners to investigate whether this asymmetry is due to a difference in how easy it is to learn contrafactives and factives. But they do not explore empirically-founded data distributions. We fill this gap, further improving the overall quality of training data distributions using linear programming.Our results confirm Strohmaier and Wimmer’s 2025 (...) conclusion that there is no learnability difference in production, while establishing the impact of differences in data distributions. (shrink)

Basic large language models (LLMs) have no direct contact with extra-linguistic reality: Their inputs, outputs and training data consist solely of text. Can they represent the world beyond that text, nevertheless? This paper considers whether LLMs represent real-world domains partly thanks to structural correspondences between their internal states and those domains. I clarify the requirements for a structural correspondence to play a genuinely content-grounding role, and argue (i) that it is a live empirical possibility that text-based LLMs meet those requirements (...) with respect to real-world contents, but (ii) that this requires carefully eliminating alternative, more deflationary, content assignments. (shrink)

This paper addresses the question of whether large language model-powered chatbots are capable of assertion. According to what we call the Thesis of Chatbot Assertion (TCA), chatbots are the kinds of things that can assert, and at least some of the output produced by current-generation chatbots qualifies as assertion. We provide some motivation for TCA, arguing that it ought to be taken seriously and not simply dismissed. We also review recent objections to TCA, arguing that these objections are weighty. We (...) thus confront the following dilemma: how can we do justice to both the considerations for and against TCA? We consider two influential responses to this dilemma - the first appeals to the notion of proxy-assertion; the second appeals to fictionalism - and argue that neither is satisfactory. Instead, reflecting on the ontogenesis of assertion, we argue that we need to make space for a category of proto-assertion. We then apply the category of proto-assertion to chatbots, arguing that treating chatbots as proto-assertors provides a satisfactory resolution to the dilemma of chatbot assertion. (shrink)

The problem considered in this chapter emerges from the tension we find when looking at the design and architecture of chatbots on the one hand and their conversational aptitude on the other. In the way that LLM chatbots are designed and built, we have good reason to suppose they don't possess second-order capacities such as intention, belief or knowledge. Yet theories of conversation make great use of second-order capacities of speakers and their audiences to explain how aspects of interaction succeed. (...) As we can all bear witness to now though, at the point of use chatbots appear capable of performing language tasks at a level close to that of humans. This creates a tension when we consider something like, for example, the classic Gricean theory of implicature. On a broad summary of this type of account, to utter p and implicate q requires the reflexive occurrence of an audience supposing a speaker believes that q, and the speaker believing that their audience can determine they believe it when they utter p. So taken at face value, if a chatbot doesn’t have the capacity for belief, then either in their role as speaker or audience, they would not seem capable of either generating or comprehending implicatures. Yet on the surface it does seem that chatbots are capable of dealing with (some) implicatures, and as such it raises questions about how we should then correlate this with what we think occurs in cases of implicature with chatbots. (shrink)

This paper explores two important ways in which the practice of close reading differs from the technique of natural language processing, the use of computer programming to decode, process, and replicate messages within a human language. It does so in order to highlight distinctive features of close reading that are not replicated by natural language processing. The first point of distinction concerns the nature of the meaning generated in each case. While natural language processing proceeds on the principle that a (...) text’s meaning can be deciphered by applying the rules governing the language in which the text is written, close reading is premised on the idea that this meaning lies in the interplay that the text prompts within readers. While the semantic theory of meaning upon which natural language processing programs are based is often taken for granted today, I draw from phenomenological and hermeneutic theories, particularly Wolfgang Iser and Hans-Georg Gadamer, to explain why a different theory of meaning is necessary for understanding the meaning generated by close reading. Second, while natural language processing programs are considered successful when they generate what epistemologists call true beliefs about a text, I argue that close reading aims first and foremost at the development, not of true belief, but of understanding. To develop this distinction, I draw from recent scholarship on the epistemology of education, including work by Duncan Pritchard, to explain how understanding differs from true belief and why attainment of the latter is less educationally significant than the former. (shrink)

Esta entrevista se realizó el 10 de agosto de 2021. El objetivo fue indagar acerca de los estudios lingüísticos de Chile; en especial, de sus hablantes natales y su producción literaria. Es insoslayable rememorar que en ese país hay dos referentes fundamentales: Gabriela Mistral y Pablo Neruda, quienes han sido influyentes en la forma de expresarse de muchos ciudadanos en el siglo XX. Asimismo, esta conversación permite conocer las causas por las que existen variantes lingüísticas en Latinoamérica, motivo por el (...) cual esta propuesta es de utilidad para emprender nuevas investigaciones en torno al uso del lenguaje desde el ámbito académico. (shrink)

Recent advances in natural language processing have given rise to a new kind of AI architecture: the language agent. By repeatedly calling an LLM to perform a variety of cognitive tasks, language agents are able to function autonomously to pursue goals specified in natural language and stored in a human-readable format. Because of their architecture, language agents exhibit behavior that is predictable according to the laws of folk psychology: they function as though they have desires and beliefs, and then make (...) and update plans to pursue their desires given their beliefs. We argue that the rise of language agents significantly reduces the probability of an existential catastrophe due to loss of control over an AGI. This is because the probability of such an existential catastrophe is proportional to the difficulty of aligning AGI systems, and language agents significantly reduce that difficulty. In particular, language agents help to resolve three important issues related to aligning AIs: reward misspecification, goal misgeneralization, and uninterpretability. (shrink)

We use language to communicate our thoughts. But is language merely the expression of thoughts, which are themselves produced by other, nonlinguistic parts of our minds? Or does language play a more transformative role in human cognition, allowing us to have thoughts that we otherwise could (or would) not have? Recent developments in artificial intelligence and cognitive science have reinvigorated this old question. Could language hold the key to the emergence of both artificial intelligence and important aspects of human intelligence? (...) The four contributions in this symposium address this question by drawing on behavioral and neural evidence from people, and the remarkable recent developments in AI which appear to show that artificial neural networks trained on language come to have an astonishing range of abilities. Despite the diversity of the speakers' perspectives, the four contributions paint a coherent (if complex) picture: The abilities of large language models (LLMs) serve as an existence proof of just how much can in principle be learned from language. LLMs also act as a stress test of cognitive theories. The evidence of neural dissociation between linguistic and conceptual processing points to the multiple realizability of human-like cognition. Finally, there is an acknowledged need for systematic research on how the successes and failures of LLMs inform our understanding of human cognition. (shrink)

A key assumption fuelling optimism about the progress of large language models (LLMs) in accurately and comprehensively modelling the world is that the truth is systematic: true statements about the world form a whole that is not just consistent, in that it contains no contradictions, but coherent, in that the truths are inferentially interlinked. This holds out the prospect that LLMs might in principle rely on that systematicity to fill in gaps and correct inaccuracies in the training data: consistency and (...) coherence promise to facilitate progress towards comprehensiveness in an LLM’s representation of the world. However, philosophers have identified compelling reasons to doubt that the truth is systematic across all domains of thought, arguing that in normative domains, in particular, the truth is largely asystematic. I argue that insofar as the truth in normative domains is asystematic, this renders it correspondingly harder for LLMs to make progress, because they cannot then leverage the systematicity of truth. And the less LLMs can rely on the systematicity of truth, the less we can rely on them to do our practical deliberation for us, because the very asystematicity of normative domains requires human agency to play a greater role in practical thought. (shrink)

The recent rise in popularity of large language models (LLMs) has prompted considerable concerns about their moral capabilities. Although considerable effort has been dedicated to aligning LLMs with human moral values, existing benchmarks and evaluations remain largely superficial, typically measuring alignment based on final ethical verdicts rather than explicit moral reasoning. In response, this paper aims to advance the investigation of LLMs’ moral capabilities by examining their capacity to function as Artificial Moral Assistants (AMAs), systems envisioned in the philosophical literature (...) to support human moral deliberation. We assert that qualifying as an AMA requires more than what state-of-the-art alignment techniques aim to achieve: not only must AMAs be able to discern ethically problematic situations, they should also be able to actively reason about them, navigating between conflicting values outside of those embedded in the alignment phase. Building on existing philosophical literature, we begin by designing a new formal framework of the specific kind of behaviour an AMA should exhibit, individuating key qualities such as deductive and abductive moral reasoning. Drawing on this theoretical framework, we develop a benchmark to test these qualities and evaluate popular open LLMs against it. Our results reveal considerable variability across models and highlight persistent shortcomings, particularly regarding abductive moral reasoning. Our work connects theoretical philosophy with practical AI evaluation while also emphasising the need for dedicated strategies to explicitly enhance moral reasoning capabilities in LLMs. (shrink)

No natural language has contrafactive attitude verbs. Because factives are universal across natural languages, this means that there is a major asymmetry between contrafactives and factives. We previously hypothesised that this asymmetry arises partly because the meaning of contrafactives is significantly harder to learn than that of factives. Here we test this hypothesis by using a production-oriented computational experiment that overcomes two limitations of our previous experiments. We find that our results do not support our previous hypothesis.

This short empirical note supplements The Common Theory of Mat and the Comic 10.5281/zenodo.17712902 by examining two natural experiments produced by the Russian-Ukrainian war. The two episodes — a 2014 football-fan chant and a 2022 military reply at Snake Island — demonstrate the spontaneous emergence of extremely compressed sign-forms. Their semantic density, pragmatic efficiency, and global recognizability provide strong empirical support for the theoretical claim that mat constitutes an optimal expressive unit under conditions of uncertainty, danger, and accelerated cognitive selection.

Wittgenstein’s ideas are a common ground for developers of Natural Language Processing (NLP) systems and linguists working on Language Acquisition and Mastery (LAM) models (Mills 1993; Lowney, Levy, Meroney and Gayler 2020; Skelac and Jandrić 2020). In recent years, we have witnessed a fast development of NLP systems capable of performing tasks as never before. NLP and LAM have been implemented based on deep learning neural networks, which learn concepts representation from rough data, but are nonetheless very effective in tasks (...) such as question answering, textual entailment, and translation (Devlin et al. 2019; Kitaev, Cao and Klein 2019; Wang et al. 2019). In this paper, I will debate some Wittgensteinian concepts that impact the architectures of many NLP deep-learning systems. I will focus, in particular, on the attempt to build a specific kind of architecture to model a private language. The discussion, I think, helps extract philosophical assumptions leading the research and development of AI systems capable of language modelling. In this paper, I will address some of the main features of NLP systems2 used for word embedding and one proposal to manipulate through a neural network a form of private language (Lawney et al. 2020). In §2, I will reconstruct the complex path of the private language argument (PLA). In §3, I will discuss connectivist language models and introduce notions about NPL systems' architecture. An overview of this kind of model is helpful to introduce the work of Lowney, Levy, Meroney and Gayler (2020). They submit that their model can respond to the issues raised by Wittgenstein in the notorious Private Language Argument (PLA). This argument unexpectedly turned out to be relevant not only for the philosophy of language but also for NLP and LAM modellers. I will describe the language game concept in NLP, how it is embedded, and its role in inductive systems development. This central concept in Wittgenstein’s work is relevant to describe the role of context in understanding the meanings of words. In §4, I present the Wittgensteinian main concepts in play in the connectivist paradigm. I argue that the connectionist theoretical framework can better catch the dependency of word meaning on context. There seems to be a correlation between Wittgenstein's invitation to look, an invitation to dismiss the aim of theorizing about languages, and the absence of theory-ladenness in deep learning technologies involved in NLP. In §5, I criticize Lowney and colleagues’ claim, whose model does not successfully capture Wittgenstein’s Beetle in the Box case. Moreover, I argue that even if we can distinguish a strong and a weak definition of a private language, Wittgenstein’s argument also holds for deep-learning models, and his worries are still a good guide for NLP developers. (shrink)

One of the central questions in the epistemology of conversational AIs is how to classify the beliefs acquired from them. Two promising candidates are instrument-based and testimony-based beliefs. However, the category of instrument-based beliefs faces an intrinsic problem, and a challenge arises in its application. On the other hand, relying solely on the category of testimony-based beliefs does not encompass the totality of our practice of using conversational AIs. To address these limitations, I propose a novel classification of beliefs that (...) shifts the focus from the properties of the conversational AIs themselves to how recipients perceive AI output, specifically whether they take the output at face value. The proposed categories are beliefs by regarding as instruments and beliefs by regarding as testifiers. By using these complementary categories, a more comprehensive understanding of beliefs acquired from conversational AIs can be achieved while avoiding the problems associated with the traditional classification. (shrink)

Are LLMs cultural technologies like photocopiers or printing presses, which transmit information but cannot create new content? A challenge for this idea, which we call bibliotechnism, is that LLMs generate novel text. We begin with a defense of bibliotechnism, showing how even novel text may inherit its meaning from original human-generated text. We then argue that bibliotechnism faces an independent challenge from examples in which LLMs generate novel reference, using new names to refer to new entities. Such examples could be (...) explained if LLMs were not cultural technologies but had beliefs, desires, and intentions. According to interpretationism in the philosophy of mind, a system has such attitudes if and only if its behavior is well explained by the hypothesis that it does. Interpretationists may hold that LLMs have attitudes, and thus have a simple solution to the novel reference problem. We emphasize, however, that interpretationism is compatible with very simple creatures having attitudes and differs sharply from views that presuppose these attitudes require onsciousness, sentience, or intelligence (topics about which we make no claims). (shrink)

Empirical evidence suggests that perceptual‐motor simulations are often constitutively involved in language comprehension. Call this “the simulation view of language comprehension”. This article applies the simulation view to illuminate the much‐discussed phenomenon of copredication, where a noun permits multiple predications which seem to select different senses of the noun simultaneously. On the proposed account, the (in)felicitousness of a copredicational sentence is closely associated with the perceptual simulations that the language user deploys in comprehending the sentence.

Zeugmatic oddness is a linguistic intuition of oddness with respect to an instance of zeugma, i.e. a sentence containing an instance of a homonymous or polysemous word being used in different meanings or senses simultaneously. Zeugmatic oddness is important for philosophical debates as philosophers often use it to argue that a particular philosophically interesting expression is ambiguous and that the phenomenon referred to by the expression is disunified. This paper takes a closer look at zeugmatic oddness. Focusing on relevant psycholinguistic (...) literature on homonymy and polysemy processing and representation, I argue that there are two different ways in which zeugmatic oddness can arise. Philosophical upshots concerning zeugmatic oddness are then drawn. (shrink)

As systems based on opaque Artificial Intelligence (AI) continue to flourish in diverse real-world applications, understanding these black box models has become paramount. In response, Explainable AI (XAI) has emerged as a field of research with practical and ethical benefits across various domains. This paper not only highlights the advancements in XAI and its application in real-world scenarios but also addresses the ongoing challenges within XAI, emphasizing the need for broader perspectives and collaborative efforts. We bring together experts from diverse (...) fields to identify open problems, striving to synchronize research agendas and accelerate XAI in practical applications. By fostering collaborative discussion and interdisciplinary cooperation, we aim to propel XAI forward, contributing to its continued success. Our goal is to put forward a comprehensive proposal for advancing XAI. To achieve this goal, we present a manifesto of 27 open problems categorized into nine categories. These challenges encapsulate the complexities and nuances of XAI and offer a road map for future research. For each problem, we provide promising research directions in the hope of harnessing the collective intelligence of interested stakeholders. (shrink)

Occurrences of sentences that are traditionally considered category mistakes, such as ‘The red number is divisible by three’, tend to elicit a sense of oddness in assessors. In attempting to explain this oddness, existing accounts in the philosophical literature commonly claim that occurrences of such sentences are associated with a defect or phenomenology unique to the class of category mistakes. It might be thought that recent work in experimental psycholinguistics—in particular, the recording of event-related brain potentials (patterns of voltage variation (...) in the brain)—holds the potential to shed new light on this debate. I review the relevant experimental results, before arguing that they present advocates of accounts of category mistakes with a dilemma: either the uniqueness claims should be rejected, or the experimental technique in question cannot be used to test existing accounts of category mistakes in the manner that philosophers might hope. (shrink)

The article argues that AI can enhance the measurement and implementation of democratic processes within political parties, known as Intra-Party Democracy (IPD). It identifies the limitations of traditional methods for measuring IPD, which often rely on formal parameters, self-reported data, and tools like surveys. Such limitations lead to partial data collection, rare updates, and significant resource demands. To address these issues, the article suggests that specific data management and Machine Learning techniques, such as natural language processing and sentiment analysis, can (...) improve the measurement and practice of IPD. (shrink)

When it comes to learning a second language, no matter what age, almost every publication talks about individual differences that lead the learners to success. It is possible to say that the age factor is the most significant of these. Various elements occur as a result of individual differences: The rate of acquisition, ultimate achievement and the processes involved in language acquisition are important ones affected by differences among learners, particularly their age. The present work deals mainly with the age (...) issue in second language acquisition, along with other factors related to individual differences, which are often treated in psychological positions. The data were obtained from a literature search. Looking at the written literature in the field, it has been found that the onset of second language acquisition and the final accomplishment are also the two titles that scholars have studied very much. The question of whether a critical period addressed language acquisition exists is a controversial one in the scientific world. However, even though there are no well-delineated age restrictions ready for attaining native or native-like proficiency in a second language, the period (age) at which one starts has been shown to correlate with ultimate proficiency. (shrink)

La scorciatoia - Come le macchine sono diventate intelligenti senza pensare in modo umano Le nostre creature sono diverse da noi e talvolta più forti. Per poterci convivere dobbiamo imparare a conoscerle Vagliano curricula, concedono mutui, scelgono le notizie che leggiamo: le macchine intelligenti sono entrate nelle nostre vite, ma non sono come ce le aspettavamo. Fanno molte delle cose che volevamo, e anche qualcuna in più, ma non possiamo capirle o ragionare con loro, perché il loro comportamento è in (...) realtà guidato da relazioni statistiche ricavate da quantità sovrumane di dati. Eppure possono essere in certi casi più potenti di noi: ci osservano continuamente, e prendono decisioni al nostro posto. E allora come incorporarle nella nostra società senza rischi ed effetti collaterali? Questo libro - rigoroso, pungente, originale nell'approccio - ci spiega come siamo arrivati sin qui, e indica il percorso che ci aspetta prima di poterci fidare di questi nuovi agenti «alieni». La tecnologia non basta, occorre un dialogo tra scienze naturali e umane: è il passaggio cruciale per una convivenza sicura con questa nuova forma di intelligenza. (Il Mulino). (shrink)