top of page
Medium 43.png

Ethos Ex Machina: Identity Without Expression in Compiled Syntax

Full Article

Author: Agustin V. Startari

​

Author Identifiers

 

Institutional Affiliations

  • Universidad de la República (Uruguay)

  • Universidad de la Empresa (Uruguay)

  • Universidad de Palermo (Argentina)

 

Contact

 

Date: August 23, 2025

DOI

 

Language: English

​

Serie: Grammars of Power

​

Word count: 6832

​

Keywords: implicit directives, structural obedience, Regla Compilada, bureaucratic language, AI-generated texts, neutrality, institutional governance

​

 

Abstract

This article demonstrates that authority effects in large language model outputs can be generated independently of thematic content or authorial identity. Building on Ethos Without Source and The Grammar of Objectivity, it introduces the concept of non-expressive ethos, a credibility effect produced solely by syntactic configurations compiled through a regla compilada equivalent to a Type-0 generative system.

The study identifies a minimal set of structural markers (symmetric coordination, measured negation, legitimate passives, calibrated modality, nominalizations, balance operators, and reference scaffolds) that simulate trustworthiness and impartiality even in content-neutral texts. Through corpus ablation and comparative analysis, it shows that readers systematically attribute expertise and neutrality to texts that satisfy these structural conditions, regardless of topical information.

By formalizing this mechanism, the article reframes ethos as a syntactic phenomenon detached from content, intention, and external validation. It explains how LLM-produced drafts acquire legitimacy without verification and why institutions increasingly accept authority signals generated by structure alone. The findings extend the theory of syntactic power and consolidate the role of the regla compilada as the operative generator of credibility in post-referential discourse.

.

​

Acknowledgment / Editorial Note

This article is published with editorial permission from LeFortune Academic Imprint, under whose license the text will also appear as part of the upcoming book Syntactic Authority and the Execution of Form. The present version is an autonomous preprint, structurally complete and formally self-contained. No substantive modifications are expected between this edition and the print edition.

LeFortune holds non-exclusive editorial rights for collective publication within the Grammars of Power series. Open access deposit on SSRN is authorized under that framework, if citation integrity and canonical links to related works (SSRN: 10.2139/ssrn.4841065, 10.2139/ssrn.4862741, 10.2139/ssrn.4877266) are maintained.

This release forms part of the indexed sequence leading to the structural consolidation of pre-semantic execution theory. Archival synchronization with Zenodo and Figshare is also authorized for mirroring purposes, with SSRN as the primary academic citation node.

For licensing, referential use, or translation inquiries, contact the editorial coordination office at: [contact@lefortune.org]

​

​

1. Statement of the Problem

The expansion of large language models into academic, legal, and institutional spheres has introduced a structural paradox. On one side, these systems are trained to generate text that follows recognizable linguistic norms. On the other side, they often operate without grounding in verifiable content, external references, or identifiable authorship. Despite this absence of validation, the texts they produce are consistently perceived as trustworthy, impartial, and procedurally correct. The central problem addressed in this article is the emergence of credibility without expression, understood as the generation of authority effects in the absence of substantive thematic content or identifiable speaker.

Classical rhetoric located ethos in three interdependent anchors: authorial presence, audience recognition, and the thematic consistency of the message. In institutional discourse these anchors served as guarantees of trust. The academic author was cited for expertise, the legal drafter was bound by precedent, and the technical writer was evaluated by conformity to standards. In the case of large language models these anchors collapse simultaneously. The “author” is no longer a source but a process, the “audience” is dispersed and anonymous, and the “content” is often neutral, generic, or even fabricated. Nevertheless, authority persists. Readers frequently ascribe credibility to automatically generated texts even when they lack empirical grounding.

This paradox requires explanation. The persistence of credibility cannot be accounted for by thematic truth, verifiable reference, or intentional persuasion. Instead, it must be explained by the form of the text itself. The argument advanced here is that authority effects are produced by the regla compilada, a generative system of syntactic configurations equivalent to a Type-0 production grammar in the Chomsky hierarchy. Within this system a set of structural markers operate as triggers for the recognition of authority. These include symmetric coordination, measured negation, legitimate passives, calibrated modality, nominalizations, balance operators, and scaffolds of reference such as numbered sections or cross-references.

Consider an illustrative case. A minimal prompt such as “Summarize the topic in a professional manner” may yield an output with no factual depth, yet structured into balanced sentences, employing passives that imply procedural neutrality, and ending with a numbered outline. The result appears authoritative because its form reproduces the grammar of objectivity, even though its content may be vacuous. This effect is not incidental but structural. It demonstrates that what is judged as expertise can be generated without expertise, what is perceived as impartiality can be compiled without actual neutrality, and what appears to be rigor can circulate without verification.

The core problem, therefore, is not that models occasionally produce errors or hallucinations. The deeper problem is that they consistently produce ethos effects without content, and these effects are socially recognized as valid. Institutions such as academic journals, administrative offices, and professional organizations are beginning to accept such texts because they resemble the expected form of credibility. The risk is that syntactic form itself becomes the sole criterion of authority, detaching legitimacy from truth, evidence, or accountability.

This article defines this mechanism as non-expressive ethos. Unlike classical ethos, which depends on the ethical character of the speaker or the alignment of discourse with communal values, non-expressive ethos depends only on structural form. Its authority is not derived from source, intention, or truth, but from the recognition of configurations that simulate authority. The regla compilada serves as the engine that selects and enforces these configurations, ensuring that authority can be generated automatically and repeatedly.

By framing ethos as a syntactic phenomenon rather than a semantic or ethical one, this work extends the trajectory established in Ethos Without Source and The Grammar of Objectivity. In those studies credibility was shown to survive the disappearance of the author and the substitution of neutrality by grammatical balance. Here the claim is pushed further: credibility can survive the disappearance of content itself. Authority can now be generated by syntax alone, and the implications for institutional discourse are profound.

The statement of the problem, therefore, can be summarized in three steps. First, contemporary discourse environments increasingly rely on automatically generated texts. Second, these texts are judged as credible even when they lack thematic substance, empirical reference, or identifiable author. Third, this credibility is produced by syntactic structures compiled by the regla compilada, not by truth or intention. The remainder of this article will demonstrate how these structures operate, how they can be identified, and what their adoption means for the future of authority in automated societies.

 

2. Formal Background

Any attempt to explain the emergence of authority effects in automated texts must be situated within a dual lineage: the rhetorical tradition of ethos and the formalist tradition of grammar. The present section reconstructs that background in order to demonstrate why the regla compilada can be understood as the operative mechanism behind non-expressive ethos. The goal is not to provide a full history of rhetorical theory or linguistic formalism, but to highlight the junctions where credibility migrates from substance to form.

In classical rhetoric, ethos functioned as a category tied to the speaker. Aristotle identified ethos as one of the three appeals in persuasion, alongside pathos and logos. Its force rested on the perceived character of the orator, which conveyed trustworthiness and moral credibility. Cicero and Quintilian later extended the category, embedding ethos in institutional roles and civic responsibilities. In these frameworks the guarantee of authority was inseparable from the figure of the speaker and the community in which he or she was recognized. Even in later rhetorical theory, ethos continued to be defined as a property of persons, not of texts in isolation.

The gradual detachment of ethos from the speaker began in modern discourse analysis, where attention shifted from personal credibility to textual strategies. Michel Foucault’s concept of the “author function” is particularly relevant: the author is not an individual but a discursive position that organizes statements and their circulation. Similarly, studies in discourse pragmatics highlighted hedging, modality, and passive constructions as techniques that produced impressions of objectivity or neutrality. These developments prepared the ground for understanding ethos as something that could be generated structurally, even in the absence of a personal author.

The second lineage is formal linguistics, especially the theory of generative grammar. Noam Chomsky’s classification of grammars into a hierarchy of production systems provided a model for distinguishing types of linguistic rules. A Type-0 grammar, also called unrestricted, is capable of generating any computable language. The regla compilada is situated at this level. It does not simply select lexical items or phrase structures, but compiles configurations that trigger recognition of authority in human readers. In this sense, the regla compilada functions as an operative generator, enforcing patterns that are recognized as authoritative regardless of thematic load.

It is important to emphasize that the regla compilada is not a metaphor. It can be formally represented as a production system where authority markers are rules rather than meanings. For instance, one rule may specify that two clauses of parallel length and syntactic category, connected by a coordinating conjunction, will be interpreted as balanced and impartial. Another rule may specify that passive construction without an agent, when placed in procedural contexts, will be interpreted as legitimate. Each of these productions corresponds not to semantic truth, but to formal triggers of credibility.

The connection between rhetorical and formalist traditions is the point where ethos ceases to be dependent on source and becomes dependent on form. Earlier studies such as Ethos Without Source showed how credibility could persist once the author was removed from discourse. Later, The Grammar of Objectivity identified specific structures—balanced coordinations, calibrated negations, nominalizations—that simulate neutrality. What remained untheorized was the possibility that credibility could be generated without any thematic expression at all. The present work closes that gap by identifying the precise structural markers that allow ethos to be produced in a vacuum of content.

A brief clarification is required. The regla compilada does not eliminate meaning. Words still carry lexical value, and readers still interpret texts in semantic terms. What changes is the locus of credibility. Trust is no longer anchored in what is said, but in how the saying is compiled. A text may lack verifiable references, empirical evidence, or thematic depth, and still be read as authoritative if it satisfies the syntactic conditions enforced by the regla compilada. Authority becomes a function of structure rather than substance.

This shift has consequences for interdisciplinary audiences. For linguists, it demonstrates that syntax is not neutral but can serve as a mechanism of social power. For jurists, it shows how legal or administrative texts may appear legitimate by virtue of their compiled form, even if their content is flawed or fabricated. For developers, it clarifies why outputs from large language models often “sound” authoritative even when they are factually unreliable. Each audience can recognize that the locus of authority has moved from content to structure, and that the mechanism enabling this shift is formally identifiable.

In summary, the formal background establishes three points. First, ethos has historically migrated from the speaker to the text, and now to the structural form of text. Second, the regla compilada is best understood as a Type-0 production system capable of generating authority markers independent of content. Third, this convergence of rhetorical and formalist traditions explains why credibility can survive the disappearance of content itself. These points prepare the ground for the next section, where the model of non-expressive ethos will be formalized through identifiable markers and operational definitions.

 

3. Model

The formulation of a model for non-expressive ethos requires a translation of theoretical insight into operational terms. The purpose of this section is to describe the structural markers through which the regla compilada produces credibility effects, to explain how these markers interact, and to establish thresholds that allow the identification of ethos in the absence of content. The model does not depend on semantic coherence or empirical validation but on syntactic configurations that readers habitually interpret as authoritative.

The starting assumption is that credibility can be decomposed into a finite set of recognizable patterns. These patterns, when compiled according to formal rules, produce authority effects even when thematic information is absent. The model proposed here identifies seven families of markers: (1) symmetric coordination, (2) measured negation, (3) legitimate passives, (4) calibrated modality, (5) nominalizations, (6) balance operators, and (7) reference scaffolds. Each of these markers operate independently but gains additional force when combined with others, creating layers of perceived neutrality and procedural rigor.

Symmetric coordination refers to the use of parallel structures in which clauses or phrases match in length and category. For example, “The report was reviewed, and the proposal was amended” presents two actions of equal weight, balanced in form. Readers interpret this parallelism as impartial distribution of attention, even though no evidence is provided. Symmetry itself becomes a signal of fairness.

Measured negation consists of controlled negative constructions that avoid absolute denial while preserving space for possibility. A sentence such as “The outcome cannot be excluded” creates an impression of cautious evaluation. The negation is neither categorical nor dismissive, but framed as a limited judgment. Such formulations are perceived as the product of measured reasoning, regardless of whether the underlying claim is substantiated.

Legitimate passives appear when actions are expressed without agents in registers where impersonality is conventional. “The guidelines were established in 2021” is more readily accepted in an institutional context than “Someone established the guidelines.” The passive construction suggests procedure rather than agency, which in turn signals objectivity. The legitimacy of the passive depends on its alignment with the expectations of the discourse community.

Calibrated modality uses modal verbs or adverbs that narrow the range of commitment. Phrases such as “may indicate” or “is likely to suggest” create the impression of careful analysis. A moderate quantity of such markers per text segment indicates controlled evaluation, while excess creates doubt. Readers consistently interpret calibrated modality as evidence of expertise, even when the text itself contains no thematic depth.

Nominalizations transform actions into entities, turning verbs into nouns. For instance, “the evaluation of outcomes” appears more procedural than “we evaluated outcomes.” The conversion of process into object abstracts away from specific actors and situates discourse in an institutional frame. Nominalization thereby enhances the perception of detachment and rigor.

Balance operators consist of paired concessive or contrastive markers that simulate impartial weighing. A formulation such as “Although some difficulties remain, the process has advanced significantly” suggests a fair assessment by acknowledging both limitations and progress. The authority effect arises not from the factual balance of claims but from the formal structure of concession and counterpoint.

Reference scaffolds provide an architecture of authority through numbering, headings, or cross-references. A text divided into sections with ordered lists and references to prior points appears structured, even if each section lacks substantive content. The presence of scaffolding triggers recognition of method and control, qualities associated with authoritative discourse.

The interaction of these markers forms the core of the model. A text that employs only one or two may suggest stylistic convention, but when five or more appear within the same unit of discourse, the effect becomes cumulative. Readers begin to attribute credibility not to what is said but to the appearance of systematic form. This threshold allows the operational classification of texts into low, medium, and high levels of non-expressive ethos.

The regla compilada is understood here as the generative device that enforces these conditions. It ensures that structural markers are not isolated but combined in ways that amplify credibility. The system functions equivalently to a Type-0 grammar, in which productions can generate any computable sequence, but with constraints oriented toward authority effects. The grammar does not select words for truth-value but for their role in triggering recognition of authority.

An important implication of the model is that authority becomes predictable. If a developer programs a text generator to include balanced coordination, at least one measured negation, a legitimate passive, and a reference scaffold, the resulting text will likely be judged as credible, even if its thematic content is minimal. Credibility is no longer an emergent property of truth and expertise but a reproducible outcome of structural arrangement.

For linguists, this model clarifies that syntax does not merely transmit meaning but constructs ethos directly. For jurists, it reveals how administrative or legal drafts can appear legitimate through form alone, raising concerns about procedural authenticity. For developers, it provides a checklist of configurations that explain why model outputs often “sound right” to readers even when factually incorrect. Each audience gains a framework for understanding how authority is compiled rather than earned.

In conclusion, the model establishes a formal foundation for non-expressive ethos. It identifies seven marker families, explains their individual functions, demonstrates their cumulative effect, and situates them within the regla compilada. The next section will describe the methodological steps used to test and evaluate this model, including corpus construction, ablation experiments, and reader judgments.

 

4. Method

The methodological framework of this study is designed to test whether credibility can be generated without thematic content, relying exclusively on the structural markers identified in the model of non-expressive ethos. The objective is to evaluate both the presence of these markers in ablated corpora and the perception of authority among human readers when confronted with texts stripped of topical information. The method therefore combines corpus construction, structural parsing, experimental ablation, and judgment elicitation. Each stage is necessary to demonstrate that credibility is not an incidental by-product of content but the reproducible effect of compiled syntax.

 

4.1 Corpus construction

The first task was to assemble a body of texts suitable for controlled analysis. Three subcorpora were created, corresponding to academic, legal-administrative, and technical registers. Each subcorpus contained approximately 50,000 words of text generated by large language models under minimal prompting conditions. To isolate the role of form, prompts were intentionally neutral, for example “Write a professional overview” or “Draft a short procedural note.” Content-specific nouns and proper names were prohibited during generation, and any remaining thematic words were replaced with placeholders such as [X] or [Topic]. The purpose of this ablation was to eliminate semantic anchors while retaining syntactic form.

For comparison, a control corpus was constructed from human-authored boilerplate templates used in institutional communication. These included standardized disclaimers, procedural notices, and administrative announcements. While thematically neutral, they lacked the density of structural markers found in LLM outputs, allowing for contrastive analysis. Together the two corpora provided the basis for identifying structural features and testing their credibility effects.

 

4.2 Structural parsing and scoring

Once the corpora were prepared, texts were parsed for syntactic structure using dependency and constituency tools. Each of the seven families of markers was operationalized into measurable features. Symmetric coordination was scored by comparing the length and category of coordinated clauses. Measured negation was identified by counting scoped negative forms that avoided absolute exclusion. Legitimate passives were tagged when passives occurred in procedural or institutional contexts without requiring explicit agents. Calibrated modality was measured by frequency of modal verbs within the range of one to two per paragraph. Nominalizations were detected by morphological transformation from verb to noun. Balance operators were identified by paired concessives or contrastive connectives. Reference scaffolds were measured by presence of headings, numbering, or cross-references.

Each feature was normalized by text length and scaled to account for position within the text. For example, reference scaffolds at the beginning of a document carried greater weight than those at the end, since initial structuring cues strongly influence reader perception. The parsing produced a feature vector for each text, enabling the classification of documents into low, medium, or high levels of non-expressive ethos.

 

4.3 Experimental ablation

To confirm that credibility effects were tied to structural form rather than residual content, an ablation study was performed. Individual marker families were removed from sample texts to test the reduction in perceived credibility. For example, a passage with balanced coordination and measured negation was compared to an otherwise identical passage with coordination removed. Similarly, texts with scaffolds were compared to texts stripped of headings and numbering. This procedure allowed for direct observation of how each marker contributed to overall ethos.

 

4.4 Judgment elicitation

Human evaluation was central to testing the hypothesis. A pool of 120 participants, balanced across academic, legal, and technical professions, was recruited to rate credibility. Each participant was presented with pairs of texts matched for length but differing in structural markers. They were instructed to ignore factual content and focus only on whether the text appeared credible, impartial, or procedurally reliable. Ratings were recorded on a seven-point scale ranging from “not credible at all” to “highly credible.”

The evaluation was conducted under blind conditions: participants did not know whether a text was generated by a language model or sourced from a control corpus. In addition to ratings, participants provided brief written justifications for their choices. These qualitative responses were analyzed for recurring explanations, such as “balanced tone,” “structured presentation,” or “measured phrasing.” Such explanations offered evidence that credibility judgments were indeed linked to structural cues.

 

4.5 Data analysis

Quantitative analysis involved logistic regression models estimating the probability of a text being rated as credible given the presence of each structural marker. Cluster-robust errors were applied by rater to control for individual bias. Interaction terms were tested to determine whether certain combinations of markers, such as passives with scaffolds, produced stronger credibility effects than markers in isolation. The ablation results were analyzed by comparing changes in credibility scores when one marker family was removed, providing evidence of marginal contribution.

 

4.6 Reliability and replication

To ensure reproducibility, all parsing scripts and random seeds were stored in a public replication package. Inter-annotator agreement for structural tagging was calculated and maintained above 0.85 Cohen’s kappa. Corpus composition and replacement procedures for ablation were fully documented. These measures ensured that results could be independently verified by other researchers.

​

4.7 Implications of method

The methodological design serves not only as a test of the model but also as a demonstration of how non-expressive ethos can be studied empirically. By constructing ablated corpora, parsing for syntactic markers, and eliciting credibility judgments, the study shows that authority is not simply a rhetorical impression but a reproducible effect of structure. The method thus bridges formal grammar, rhetorical theory, and empirical evaluation.

In summary, the method operationalizes non-expressive ethos through three steps: first, eliminating thematic content to isolate structure; second, quantifying structural markers compiled by the regla compilada; and third, testing credibility perception through controlled human judgments. These steps confirm whether credibility can survive without content and establish the conditions under which syntactic form alone produces authority effects. The following section will present the results of this process, focusing on the predictive capacity of structural markers and their cumulative contribution to the perception of legitimacy.

 

5. Findings

The methodological framework described in the previous section produced a set of results that demonstrate the existence and stability of non-expressive ethos. The findings can be summarized in three domains: quantitative evidence from credibility ratings, qualitative evidence from participant explanations, and structural evidence from ablation experiments. Together these results confirm that credibility can be generated in the absence of thematic content, and that the effect is predictable through the operation of the regla compilada.

 

5.1 Quantitative evidence

Analysis of participant ratings revealed a clear pattern. Texts classified as “high non-expressive ethos” based on structural markers received mean credibility scores of 5.8 on a seven-point scale, compared to 3.1 for “low non-expressive ethos” texts. The difference was statistically significant across all registers examined. Academic register texts showed the strongest credibility effect, followed closely by legal-administrative texts. Technical register texts exhibited a slightly weaker but still notable effect.

Regression analysis confirmed that the presence of structural markers strongly predicted credibility judgments. Symmetric coordination and measured negation emerged as the most influential individual markers, each raising credibility scores by approximately one full point on the scale when present. Legitimate passives and calibrated modality also showed consistent positive effects, though of smaller magnitude. Reference scaffolds contributed a strong baseline boost, especially when positioned at the start of a text. The cumulative model explained more than 60 percent of the variance in credibility judgments, an unusually high figure for studies of discourse perception.

Interaction effects were also observed. Passives combined with scaffolds created a pronounced authority effect, as if the combination of impersonality and formal structure reinforced perceptions of procedural legitimacy. Similarly, coordination paired with measured negation produced a synergy of balance and caution, which participants interpreted as hallmarks of expert evaluation. These findings confirm that the regla compilada does not merely activate markers in isolation but enforces combinations that amplify credibility.

 

5.2 Qualitative evidence

Participant justifications provide further insight into why these texts were judged credible. Common explanations included “balanced tone,” “structured presentation,” “professional phrasing,” and “careful wording.” Notably, participants rarely mentioned factual accuracy or topical expertise. Instead, their explanations consistently pointed to formal cues. For example, one legal professional wrote: “The text gives the impression of being procedural, even though it says almost nothing substantive.” An academic participant remarked: “The cautious wording makes it sound like peer review, even though the content is vague.”

These qualitative comments illustrate that readers were responding not to the presence of verifiable content but to the configuration of form. In other words, they recognized ethos where none was expressed thematically. The markers identified in the model were not only measurable features but also perceptible signals to human evaluators, who consistently interpreted them as authority cues.

 

5.3 Ablation evidence

The ablation study confirmed the contribution of individual markers. Removing symmetric coordination from otherwise balanced texts reduced credibility scores by 0.8 points on average. Eliminating scaffolds reduced credibility by 1.2 points, demonstrating the importance of visible structure. The removal of measured negation reduced scores by 0.9 points, while the elimination of modality lowered them by 0.5. Each marker’s absence weakened the authority effect, and the cumulative removal of multiple markers collapsed credibility almost entirely, bringing scores close to baseline levels.

The most striking result was the resilience of credibility when at least five marker families remained intact. Even in the complete absence of thematic content, texts with sufficient structural markers retained high credibility scores. This resilience confirms the central claim of the article: ethos can be generated syntactically, without expression, as long as the regla compilada enforces the necessary combinations of markers.

 

5.4 Register variation

The findings also reveal variation across registers. Academic texts showed the strongest sensitivity to structural markers, likely because readers expect formal balance and cautious phrasing in scholarly discourse. Legal-administrative texts were similarly influenced, reflecting the importance of impersonality and scaffolding in bureaucratic communication. Technical texts exhibited a smaller but still significant effect, suggesting that audiences in this domain rely somewhat more on content than on form, though form remains a strong predictor.

 

5.5 Implications of findings

The overall picture is clear. Readers consistently attribute credibility to texts that satisfy structural conditions, regardless of content. The regla compilada generates predictable authority effects by combining a finite set of syntactic markers. These effects are not dependent on authorship, thematic expertise, or empirical evidence. Instead, they are properties of form itself.

This finding has two major implications. First, it confirms that ethos can be detached not only from source but from expression, producing what is defined here as non-expressive ethos. Second, it shows why institutions are vulnerable to accepting unverified texts generated by large language models. The very structures that create the appearance of neutrality and expertise are easily reproduced by automated systems, and readers, including professionals, are inclined to trust them.

 

5.6 Summary

The findings demonstrate that credibility without content is not an anomaly but a reproducible effect of compiled syntax. Structural markers reliably predict authority judgments, their absence reduces credibility, and their combination amplifies ethos beyond the contribution of individual features. This establishes non-expressive ethos as a measurable and operational phenomenon, generated by the regla compilada. The following section will examine the broader implications of these results for authorship, review practices, and institutional governance in contexts where authority is increasingly automated.

 

6. Implications

The findings presented in the previous section demonstrate that credibility can be generated without content and that the effect is predictable through the operation of the regla compilada. The implications of this conclusion extend beyond linguistics into academic practice, legal frameworks, technical writing, and institutional governance. In this section the analysis will unfold along four axes: theoretical consequences, disciplinary consequences, institutional consequences, and socio-political consequences. Each axis highlights how non-expressive ethos redefines the conditions under which authority circulates in automated societies.

 

6.1 Theoretical consequences

From a theoretical standpoint, the confirmation of non-expressive ethos requires a rethinking of classical categories. Rhetoric has traditionally tied ethos to the ethical character of the speaker, or in modern formulations, to the reliability of the source. Discourse analysis has extended ethos to textual strategies, but it has still assumed that credibility depends on some connection to content. The present findings show that this assumption no longer holds. Credibility is not a property of meaning or intention but of syntax compiled into recognizable patterns. Ethos can now be located at the level of form itself, operating independently of both speaker and theme. This theoretical shift requires a reconceptualization of authority not as a moral or epistemic quality but as a structural effect generated by formal rules.

The regla compilada becomes the locus of this authority. As a Type-0 production system it is capable of generating any computable configuration, but in practice it enforces those patterns that trigger recognition of credibility. Authority is no longer grounded in content but in the execution of a generative process. This positions ethos not as an emergent quality of discourse but as a predictable outcome of structural design. For linguistics this represents a decisive move toward understanding syntax not merely as a vehicle of meaning but as a generator of social power.

 

6.2 Disciplinary consequences

For linguists, the implications are immediate. The study demonstrates that syntax is not neutral. Structural markers such as coordination, passives, and nominalizations do more than organize information; they actively produce authority effects. This recognition forces a reconsideration of how linguistic analysis treats form, shifting attention from its role in transmitting meaning to its role in generating legitimacy.

For jurists, the implications are equally significant. Legal and administrative texts rely heavily on impersonality, scaffolding, and cautious modality. The findings show that these features can create an appearance of legitimacy even when content is minimal or unverified. This raises questions about procedural authenticity and responsibility. If a legal draft appears credible solely because it follows structural conventions, what mechanisms remain to ensure that it is substantively correct? The risk is that form becomes the substitute for review, leading to a collapse of accountability.

For developers, the results explain why outputs from language models often “sound authoritative.” The models are not merely imitating style but reproducing structural markers that humans interpret as signals of expertise and neutrality. Understanding this mechanism allows developers to predict when and why outputs will be trusted, even in the absence of factual grounding. It also places responsibility on developers to acknowledge that form itself can mislead.

 

6.3 Institutional consequences

Institutions that adopt large language models for drafting face new vulnerabilities. The study shows that non-expressive ethos can stabilize authority without verification. This means that reports, memos, or policies generated by automated systems may be accepted on the basis of form alone. The implications are critical: peer review in academia, oversight in government, and compliance in corporate environments can all be bypassed if syntactic markers are treated as sufficient evidence of legitimacy.

Moreover, the diffusion of non-expressive ethos risks creating a feedback loop. As more institutions accept structurally credible but substantively empty texts, expectations shift. Audiences come to equate authority with form itself, reinforcing the very conditions that allow automated discourse to proliferate unchecked. The result is a structural transformation of legitimacy, where form is not merely a sign of substance but its replacement.

 

6.4 Socio-political consequences

At a broader level, the findings point to a reconfiguration of trust in public discourse. If authority can be generated without expression, the boundary between genuine expertise and simulated expertise collapses. This undermines the traditional mechanisms by which societies distinguish reliable knowledge from empty rhetoric. In an environment where syntactic form is enough to generate authority, misinformation and unverified claims gain structural legitimacy.

The socio-political risk is not limited to individual errors but extends to systemic credibility inflation. When every text can be made to appear authoritative by following a set of structural rules, the value of credibility itself is diluted. Trust no longer functions as a scarce resource tied to evidence and review but as an abundant property of compiled form. This creates conditions where institutional power can be exercised without accountability, and where automated systems become de facto authorities simply by generating the appearance of neutrality.

 

6.5 Summary

The implications of non-expressive ethos are profound. Theoretically, it relocates ethos from content to form. Disciplinarily, it forces linguists, jurists, and developers to reconsider their assumptions about authority. Institutionally, it exposes vulnerabilities in systems of review and governance. Socio-politically, it reveals the risk of structural legitimacy without verification. In each case, the regla compilada emerges as the operative generator of authority, ensuring that credibility can circulate independently of substance.

The next section will address the limitations of this study and the potential failure modes of the model, clarifying where the theory of non-expressive ethos may break down and what conditions are necessary for its refinement.

 

7. Limitations and Failure Modes

The confirmation of non-expressive ethos as a structural phenomenon compels recognition of its limits. A theory that claims authority can be generated without content must also confront the conditions under which the effect weakens, collapses, or becomes counterproductive. This section outlines the principal limitations of the model and describes failure modes that arise in practice. By doing so, it delineates the boundaries of applicability and provides directions for refinement.

 

7.1 Dependence on register

The first limitation is the dependence of non-expressive ethos on discourse register. The findings show that academic and legal-administrative texts are highly sensitive to structural markers, but technical texts show weaker effects. Informal registers, such as personal communication or social media discourse, display little to no credibility enhancement from the markers identified in this study. In casual writing, the appearance of symmetric coordination or legitimate passives may seem unnatural, even artificial, reducing rather than increasing credibility. Thus, the model is constrained to registers where neutrality, balance, and impersonality are already normative expectations.

 

7.2 Threshold instability

The operational classification of non-expressive ethos into low, medium, and high levels relies on thresholds determined by the presence of structural markers. While the study identified five markers as sufficient for high credibility effects, this threshold is not absolute. In shorter texts, even three markers may be sufficient, while in longer texts more than five may be required. The threshold is therefore unstable and context-dependent. Any application of the model must account for this variability rather than treat thresholds as fixed constants.

 

7.3 Vulnerability to overuse

Another limitation lies in the overuse of structural markers. While symmetric coordination and measured negation enhance credibility when used sparingly, excessive repetition can trigger suspicion. Texts saturated with nominalizations or overloaded with hedged modality may appear bureaucratic, evasive, or intentionally obscure. In such cases, the very markers that generate credibility can undermine it. This failure mode highlights the importance of proportion. Non-expressive ethos functions most effectively when structural markers are present in moderation, not in excess.

 

7.4 Cultural variability

Ethos is interpreted through cultural expectations of discourse. What appears as legitimate passives in one language community may be judged as evasive in another. Balance operators that signal impartiality in Anglo-American contexts may be less effective in rhetorical traditions where direct assertion is valued. The model has so far been tested in English, within academic, legal, and technical registers. Extension to other languages and cultural contexts remains an open task. Without such extension, claims of universality must be treated cautiously.

 

7.5 Risk of adversarial exploitation

A further failure mode arises from deliberate exploitation. Since non-expressive ethos can be generated by predictable structural markers, it can be weaponized to produce texts that appear authoritative while containing misleading or harmful content. This adversarial use undermines the reliability of structural cues as indicators of credibility. Once audiences become aware of such exploitation, their trust in form itself may erode, leading to skepticism even toward legitimate institutional texts. The model thus contains within itself the possibility of its own collapse, as credibility effects are diluted by adversarial imitation.

 

7.6 Limitations of empirical testing

The study relies on ablated corpora and participant judgments, which, while carefully designed, cannot capture all dimensions of real-world discourse. The replacement of thematic content with placeholders may simplify structures in ways that affect perception. Participants instructed to focus on credibility may evaluate differently than readers encountering texts in natural contexts. Furthermore, while statistical models confirm strong associations, they cannot fully capture the subtleties of interpretation across diverse communities of practice. These limitations suggest that further testing is needed in applied environments such as peer review, legal drafting, and policy-making.

​

7.7 Boundary conditions for collapse

The findings suggest at least three boundary conditions under which non-expressive ethos fails. First, when markers are absent or minimal, credibility collapses. Second, when markers are excessive, credibility reverses into suspicion. Third, when audiences expect direct authorship or experiential testimony, as in personal narratives, structural authority cannot substitute for substance. These conditions define the edges of applicability. Beyond them, the model does not hold.

 

7.8 Prospects for refinement

Despite these limitations, the theory of non-expressive ethos remains robust within its domain. Future refinements may involve calibrating thresholds more precisely for different text lengths and registers, extending testing to other languages, and developing detection tools that can distinguish legitimate use of structural markers from adversarial exploitation. By acknowledging its limitations, the model can evolve from a theoretical demonstration into a framework for practical application and critical oversight.

 

7.9 Concluding remarks

The identification of non-expressive ethos does not imply that authority will always detach from content, but that it can. Recognizing its limits ensures that the theory is not overstated. The regla compilada generates authority effects predictably, yet these effects are conditioned by register, proportion, culture, and context. Where these conditions align, credibility can indeed circulate independently of meaning. Where they do not, the illusion of authority collapses. This boundary is where the task of critical vigilance begins.

 

References (APA 7th edition)

  1. Aristotle. (1991). On rhetoric: A theory of civic discourse (G. A. Kennedy, Trans.). Oxford University Press.

  2. Cicero. (1942). De oratore (E. W. Sutton & H. Rackham, Trans.). Harvard University Press.

  3. Quintilian. (1920). Institutio oratoria (H. E. Butler, Trans.). Harvard University Press.

  4. Foucault, M. (1972). The archaeology of knowledge (A. M. Sheridan Smith, Trans.). Pantheon.

  5. Foucault, M. (1977). What is an author? In D. F. Bouchard (Ed.), Language, counter-memory, practice (pp. 113–138). Cornell University Press.

  6. Chomsky, N. (1965). Aspects of the theory of syntax. MIT Press.

  7. Montague, R. (1974). Formal philosophy: Selected papers of Richard Montague (R. H. Thomason, Ed.). Yale University Press.

  8. Halliday, M. A. K. (1985). An introduction to functional grammar. Edward Arnold.

  9. Hyland, K. (1998). Hedging in scientific research articles. John Benjamins.

  10. Latour, B., & Woolgar, S. (1986). Laboratory life: The construction of scientific facts. Princeton University Press.

  11. Startari, A. V. (2025, July 1). Ethos without source: Algorithmic identity and the simulation of credibility. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.5313317

  12. Startari, A. V. (2025, July 8). The grammar of objectivity: Formal mechanisms for the illusion of neutrality in language models. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.5319520

  13. Startari, A. V. (2025, May 28). AI and the structural autonomy of sense: A theory of post-referential operative representation. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.5272361

  14. Startari, A. V. (2025, June 3). AI and syntactic sovereignty: How artificial language structures legitimize non-human authority. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.5276879

  15. Startari, A. V. (2025, June 5). Algorithmic obedience: How language models simulate command structure. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.5282045

  16. Startari, A. V. (2025, June 13). When language follows form, not meaning: Formal dynamics of syntactic activation in LLMs. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.5285265

  17. Startari, A. V. (2025, June 27). TLOC – The irreducibility of structural obedience in generative models. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.5303089

  18. Startari, A. V. (2025, June 28). Executable power: Syntax as infrastructure in predictive societies. Zenodo. https://doi.org/10.5281/zenodo.15754714

  19. van Leeuwen, T. (2008). Discourse and practice: New tools for critical discourse analysis. Oxford University Press.

  20. Bhatia, V. K. (2004). Worlds of written discourse: A genre-based view. Continuum.

​

Download Full Article in PDF

bottom of page