Роль искусственного интеллекта и мультилингвальной коммуникации в цифровой реактуализации архитектурного наследия

Architectural heritage functions as a public reservoir of meaning by condensing the long arc of collective life into forms that endure and invite interpretation. Contemporary conditions complicate this process of interpretation. The problem is not the presence of new tools but the absence of an integrated account of how AI mediation, platform visibility, and multilingual communication jointly shape what becomes credible, discoverable, and discussable as heritage. Without such an account, translation ethics, provenance, and community correction remain treated as add-ons, even though they determine whether digital re-actualization stabilizes cultural memory or quietly simplifies it.

The geopolitical environment has fractured into competing spheres of exchange and control, platform infrastructures have stratified access while claiming universality, and public discourse has intensified around contested pasts and selective futures. Built heritage stands in these currents with vulnerability and strength, held in place by the weight of materials and pushed along by currents of memory and policy.

Artificial intelligence has entered heritage practice as a family of mediations that convert images and texts into recognizable patterns, generate plausible reconstructions, and scale multilingual communication beyond traditional limits. Computer vision identifies ornamental vocabularies, materials, and pathologies with remarkable speed [14]. Language technologies for text analysis structure descriptive corpora and user submissions by topic and improve search and retrieval across large collections. Machine translation lowers thresholds for cross-linguistic participation [8], and generative models produce visual and narrative possibilities that challenge established conservation sensibilities. These capacities recalibrate what can be seen, retrieved, and narrated, and they do so at scales previously reserved for centralized institutions.

Multilingual communication, in parallel, reconfigures audiences by acknowledging that heritage lives in multiple languages even within single neighborhoods. Architectural vocabularies intersect with legal phrases, artisanal idioms, and community speech, producing zones of partial overlap and necessary explanation. Where earlier practice tolerated translation as a service appended to curation, current conditions reveal it as a constitutive act within heritage interpretation [3; 11]. In digitally mediated settings, translation and explanation shape reception as directly as image resolution and interface design [15]. These developments create conditions for re-actualization, understood here as the reattachment of historical form to present concerns through careful mediation rather than simple display. Re-actualization defends plural narration and cultivates critical seeing. It resists reductive harmonization across languages and discourages algorithmic preferences for picturesque completeness over difficult complexity. It emphasizes transparency in technical operations and insists upon community participation as a precondition for legitimacy [4].

The analysis that follows situates digital re-actualization within scholarly debates, formalizes a conceptual approach attentive to semiotics and memory, describes a method aligned with interpretive rigor and public pedagogy, and explores empirical vignettes that illustrate promises and hazards. The study’s aim is to articulate academic and practical guidance for institutions and educators who carry responsibility for built memory in circumstances that reward speed but demand judgment.

Literature Review

Scholarship on digital heritage evolved from documentation-centric strategies into broader concerns with experience, access, and governance. Early work took as its mandate the preservation of fragile materials through digitization, emphasizing standards, metadata, and inter-institutional interoperability [7; 13]. The ethos of this phase centered on rescue and availability: to capture, catalog, and disseminate visual and textual traces before environmental, political, or economic pressures rendered the work impossible. As repositories expanded, attention shifted toward narrative curation and audience reception. Virtual tours, augmented overlays, and three-dimensional reconstructions promised engagement for publics unable to travel and proposed new modalities of encounter for local audiences [13; 4]. Critical studies responded by interrogating presence and authenticity, proposing that increased technical fidelity does not automatically produce meaningful experience and that curatorial rhetoric shapes interpretive outcomes as much as pixel density [6]. Urban semiotics contributed a complementary perspective by treating built form as a system of signs embedded in social practice. Foundational work described how space is produced through social relations and material practices that structure perception and use [10]. Memory studies tracked the creation of lieux de mémoire and the transformation of living memory into curated remembrance [12], while cultural memory theory articulated mechanisms through which societies stabilize narratives over time and ritual [1]. The linguistic landscape tradition sharpened urban semiotics by analyzing inscriptions, signage, and informal markings as part of an urban grammar that registers power, identity, and conflict [9; 15]. In this literature, heritage does not occupy an inert past but participates actively in present identity formation and contestation.

Digital layers add voices to an already polyphonic city, amplifying certain registers while potentially muting others. Platform studies introduced an infrastructural critique focused on proprietary architectures governing public culture. Research documented how recommendation engines, moderation policies, and advertising models shape visibility, legitimacy, and affective tone [5]. The promise of global reach collides with data localization, regulatory variance, and geopolitical interference [17; 16]. Cultural materials travel through pipelines optimized for engagement, and these optimizations carry normative assumptions about relevance and risk. Scholars have identified algorithmic amplification of dominant narratives, suppression of minority languages, and incentive structures that simplify complex histories to fit attention economies [5; 17]. Heritage practice within platforms thus demands political economy analysis alongside curatorial craft.

Artificial intelligence entered these debates both as technical possibility and epistemic challenge. Computer vision advanced tasks in style recognition, damage detection, and object retrieval across large image corpora [14]. Heritage-focused work demonstrated benefits of semantic enrichment, linked open data, and computer-assisted annotation for architectural archives [18]. Natural language processing improved clustering of themes and automatic summarization across multilingual sources. Machine translation began to produce acceptable drafts in contexts once deemed prohibitively complex [8]. Generative models introduced speculative reconstruction and simulation into workflows, raising questions about the epistemic status of synthetic imagery. The critical response emphasized training data provenance, model bias, hallucination risks, and interpretive consequences of automation [6]. Heritage scholarship underscored that labels, clusters, and reconstructions are not neutral outputs but proposals requiring social validation and contextualization.

Language pedagogy and applied linguistics recognized in digital heritage a fertile site for experiential, project-based learning. Translanguaging approaches invite learners to mobilize full repertoires rather than perform monolingual identities [3; 11]. Classroom interventions demonstrated that co-curation of microexhibits across languages promotes critical literacy, intercultural competence, and ethical reflexivity [15]. Treating machine translations and summaries as drafts subject to annotation, repair, and rejection positions learners as active mediators rather than passive consumers. This pedagogical strand contributes operational principles to heritage practice, including explicit attention to untranslatable terms, community glossaries, and deliberative processes foregrounding local idioms.

Across the literature, ethical concerns about consent, representation, and extraction recur with urgency. Data colonialism describes patterns in which cultural resources become raw material for distant value production without adequate community benefit or control [2]. Heritage contexts magnify these stakes because built memory bears intimate connections to identity and belonging. Scholars advocate governance models that share authority, transparent technical documentation, and licensing regimes preserving community values [4; 6]. The convergence of digital heritage, urban semiotics, platform critique, AI methods, and language pedagogy thus yields a composite field attentive to symbolic complexity and infrastructural realities.

Theoretical Framework

The conceptual foundation proceeds from cultural semiotics and memory studies, integrating computational mediation within a broader interpretive ecology. Architectural heritage is conceived as a living semiotic field where forms function as signs, settings, and scripts for social action [10; 12; 1]. Meanings emerge at the intersection of aesthetic codes, legal classifications, ritual practices, and narrative habits. Memory is treated as public performance rather than passive archive, characterized by repetition, contestation, and selective emphasis [12; 1]. Communities enact memory through ceremonies, disputes, and everyday movement through spaces that remind, admonish, and encourage reflection.

Artificial intelligence takes its place in this field as a translator of signals into proposals. Visual inputs become labels and embeddings; textual inputs become clusters, summaries, and cross-references; multilingual inputs become alignments and candidate translations [14; 8]. These operations, while computationally impressive, remain interpretive in consequence. Outputs propose categories and narratives that may align with local understanding or diverge from it. The framework rejects technocratic neutrality and emphasizes social validation. Proposals warrant acceptance, amendment, or refusal through processes that include expert review, community consultation, and pedagogical examination [4; 6]. Multilingual communication functions as the medium of shared interpretation. Heritage resides in translation not only across distinct languages but across registers within a single language, including professional jargon, vernacular speech, and ceremonial diction [15].

Translation ethics, therefore, assume central importance. Literal equivalence often fails where cultural density is high, and responsible mediation requires explanation, exemplification, and sometimes deliberate retention of untranslatable terms with notes [3; 11]. The framework privileges transparency about choices and invites audiences to see the work of translation rather than consume its products invisibly.

Pedagogy enters as a constitutive component of the framework. Educational practice is not an external application but a primary venue for cultivating civic literacy in digital heritage. Learners engage as co-cu-rators, scrutinizing model outputs, testing translations against community usage, and building narratives that honor plural perspectives [15; 3; 11]. This pedagogical stance emphasizes agency, critique, and collaboration. It produces not only classroom outcomes but public value by generating materials, glossaries, and practices transferable to institutional settings. In this sense, pedagogy becomes method and governance, training interpretive habits necessary for ethical curation.

Methodology

The methodological approach combines interface reading, comparative observation, and classroombased intervention in a manner consistent with interpretive social research. The approach is designed to make the link between procedures and reported results explicit. Each method targets a different layer of the same phenomenon: how digital heritage becomes publicly legible and socially authorized under AI mediation and multilingual conditions. Interface reading captures the visible grammar of curation embedded in design and interaction; comparative observation tests whether the same mechanisms recur across contexts; classroom-based intervention provides documented evidence of how AI-assisted drafts are evaluated, corrected, and justified when multilingual narration is produced as accountable public work.

Interface reading treats digital heritage platforms as texts with rhetorical structure. Layout, navigation, labeling, recommendation carousels, and translation features are analyzed for the ways they stage meaning and distribute attention [5; 17]. This reading extends to moderation cues, community contribution prompts, and documentation of technical processes, with particular emphasis on legibility of algorithmic operations for non-specialist audiences.

Comparative observation examines multiple urban contexts in which AI-assisted heritage curation is active. The focus falls on recurring patterns rather than exhaustive cataloging. Platforms are selected based on evidence of computer vision tagging, NLP summarization, machine translation layers, and generative reconstruction features [14; 18]. Observations include the balance between official narration and user-generated content, the treatment of contested histories, and the visibility accorded to minority languages [15]. Attention is paid to geopolitical constraints such as data localization and regional feature disparities that alter user experience and curatorial capacity [16; 17]. Classroom-based intervention functions as empirical and pedagogical test bases. Learners assemble micro-collections for specific buildings or districts using images, plans, archival texts, and oral histories. AI tools assist with labeling, retrieval, and draft translation [8; 14]. Outputs are treated as proposals subject to iterative human revision. Public-facing micro-exhibits are constructed with bilingual or trilingual narration, community glossaries, and notes on contested terms [3; 11]. Feedback sessions with residents and practitioners provide qualitative evaluation of accuracy, tone, and ethical adequacy. The intervention yields insight into audience reception and provides operational protocols for participatory curation [4].

Accordingly, the methods do not function as parallel descriptions but as a single chain of inference. Interface analysis produces claims about attention distribution, legitimacy cues, and the treatment of translation and uncertainty; comparative observation supports cautious generalization beyond a single platform or city; classroom intervention adds process data that reveal typical failure points and repair strategies in labeling, retrieval, and translation. The subsequent Case Studies section demonstrates these mechanisms in situated form, while the Findings and Discussion sections consolidate them into transferable principles for multilingual, AI-supported heritage curation.

Data sources include platform interfaces and documentation, publicly accessible archives, classroom artifacts such as drafts and annotations, and recorded feedback from community sessions. Analysis proceeds through thematic coding of interface elements, narrative patterns, and translation choices, supported by descriptive statistics where appropriate to indicate attention distributions or language presence without reducing interpretation to quantification. The methodological stance privileges transparency about limits and emphasizes reflexivity regarding institutional positioning while maintaining formal academic register.

Case Studies

The cases below are presented as analytical vignettes that operationalize the framework and methods described above. They are not offered as exhaustive descriptions of particular cities, but as typologically distinct situations repeatedly encountered in platform-mediated heritage environments, where AI support and multilingual access are already embedded in ordinary user experience. The purpose of the section is to show, in concrete configurations, how the same tensions identified in the theoretical part become observable at the level of interface rhetoric, visibility allocation, translation practice, and the marking of uncertainty in reconstruction.

A waterfront district characterized by industrial warehouses illustrates how AI-assisted multilingual curation can widen access while quietly narrowing the story. Digitized elevations, interior panoramas, and mobile tours enable users to move between historical photographs and current views, while computer-vision tagging supports the retrieval of recurring architectural elements and details across large image sets [14]. A multilingual layer provides guides in several languages and may be supplemented by volunteer glossaries for artisanal terms that do not travel well across languages. User submissions in text and voice expand the record beyond official narration by adding labor histories, informal practices, and neighborhood sentiments.

Interface reading, however, shows a stable bias in how attention is distributed. Recommendation surfaces and “nearby/related” pathways tend to privilege picturesque vantage points and visually coherent sequences, which makes maintenance histories, conflict-ual episodes, and traces of disrepair less discoverable unless the user searches deliberately. In multilingual mode the same asymmetry becomes semantic: automated drafts handle technical descriptors reliably, yet flatten idiomatic expressions and occupational speech into neutral paraphrase, reducing local texture and, with it, interpretive credibility [8]. Classroom-based co-curation tasks, where machine translations were treated as drafts requiring transparent correction, consistently indicated that short translation notes and curated glossaries reduce this loss because they make non-equivalence visible rather than pretending it does not exist. In feedback discussions, audiences tended to trust multilingual narration more when it acknowledged such decisions explicitly, even when that acknowledgement made the text less “smooth.”

A district marked by wartime damage or prolonged neglect foregrounds tensions around generative reconstruction and the public legibility of conjecture. Institutions increasingly employ models to propose plausible completions for missing elements and ornamental sequences, and they may add provenance notes and watermarking to prevent confusion between synthetic and archival imagery. The critical point observed at interface level is that conjectural outputs often receive privileged placement in the visual hierarchy, while evidentiary framing is placed downstream and becomes optional reading. In multilingual presentation this problem is amplified when provenance cues and certainty markers are translated unevenly or simplified, producing different degrees of epistemic clarity for different language audiences.

Across observed environments, the most robust practice is not to refuse reconstruction but to stage it as hypothesis at first contact, so that uncertainty is legible in captions, thumbnails, and default views rather than hidden in secondary documentation. Classroom exercises that required parallel captions in two or three languages under a strict constraint – equal visibility of certainty and uncertainty across versions – exposed how easily languages drift into different strength of claim, especially when a draft translation “improves”

rhetorical confidence. The pedagogical setting proved useful precisely because it forces the repair of that drift and turns certainty markers into an explicit part of curatorial responsibility.

A modernist housing complex provides a third configuration in which co-curation is not a supplement but a governance necessity. Here heritage is lived-in and debated daily, and narration is produced in competing registers that include professional heritage language, administrative speech, and resident experience. AI-assisted clustering and draft translation can support navigation across archives and community materials [18], but the observed risk is a forced coherence in which plurality is compressed into a single authoritative storyline. The analytical signal of this compression is visible both in interface pathways and in language layers: minority languages may appear only as secondary translations of a dominant narrative, which grants access while withholding interpretive agency.

Where multilingual co-curation is treated as a structured practice, plurality becomes visible without collapsing into noise. Divergent claims can be placed side by side with clear speaker positions, and culturally dense terms can remain present with brief explanation rather than being replaced by over-general equivalents. In classroom-based prototypes, audiences repeatedly preferred branching narrative pathways over linear tours because branching mirrors the actual social life of the place. The digital presence of the complex became more usable and more legitimate when it refused to simulate harmony where experiences diverge, and when it provided correction mechanisms anchored in community ownership of key terminology.

The case vignettes are therefore used not as self-contained narratives but as empirical lenses: each makes visible how the same curatorial tensions are produced by a combination of interface design, algorithmic visibility, and multilingual mediation. What varies across situations is the surface form of the conflict, while the underlying mechanism remains stable: platforms and AI tools shape what becomes discoverable, credible, and translatable, and the cost of “smooth” access is often paid by local texture, contested memory, and the legibility of uncertainty.

Findings

The findings below consolidate the recurring patterns observed across the vignettes and treat them as design- and governance-relevant regularities rather than as properties of a single site. In other words, the section translates situated observations into transferable claims about how AI-supported heritage curation operates under platform conditions: how visibility is allocated, how multilingual equivalence is negotiated, and how degrees of evidential certainty are either maintained or lost in public presentation.

Patterns across cases indicate that algorithmic curation inscribes priorities interacting with platform incentives and geopolitical constraints. Scenic preference, engagement prediction, and safety thresholds tilt visibility toward certain narratives and away from others [5; 16]. Multilingual layers improve access but require deliberative treatment of idiom, register, and untranslatable terms to preserve local texture [3; 11].

Generative reconstruction expands imagination but must be framed within evidentiary humility to avoid conflating conjecture with verification [6]. Co-curation distributes authority and yields richer interpretation, yet demands time, training, and governance structures capable of sustaining participation beyond project cycles [4].Effective practices include transparent documentation of recommendation logic [5], explicit labeling of synthetic imagery and conjectural models [6], and public articulation of translation choices supported by community glossaries [3; 11].

Narrative pathways that permit branching and juxtaposition reveal minority voices without requiring virality [15]. Moderation policies written in local idiom strengthen respectful participation and resist tone-deaf enforcement. Pedagogical involvement introduces disciplined revision and reflection, enhancing interpretive quality of public-facing materials.

Discussion

The analysis supports a conception of digital re-actualization as ethical and interpretive enterprise rather than purely technical upgrade. AI systems operate as powerful instruments whose outputs require social validation; treating these outputs as drafts clarifies institutional and community responsibilities and slows decision-making sufficiently to admit reflection [4; 6].

Multilingual communication functions not as accessory but as central channel through which heritage gains or loses its meaning; translation ethics, when made visible, increases audience trust and encourages co-production of knowledge [3; 11; 15].

Platform infrastructures complicate practice by embedding cultural materials in engagement economies [5; 17]. Recommendation systems and moderation models reduce complexity under pressure to retain attention and avoid conflict. Geopolitical fragmentation intensifies these dynamics by restricting data flows, segmenting features across jurisdictions, and altering availability of language technologies [16].

Institutions respond by diversifying hosting arrangements, adopting open standards, and negotiating platform constraints with explicit policy positions [13; 4]. Community partnerships provide legitimacy and local expertise that platforms cannot supply. The convergence of technical capacity, institutional governance, and public pedagogy becomes the decisive zone in which heritage retains dignity and interpretive richness.

Pedagogical Implications

Educational practice grounded in heritage curation fosters civic literacy and cross-linguistic competence. Learners who engage with AI tools as proposal generators develop habits of evaluation and repair [3; 11]. Project-based work culminating in public micro-exhibits introduces accountability and audience awareness [15].

Translanguaging approaches allow learners to mobilize full repertoires, confront untranslatable terms, and write notes that honor cultural specificity. Assessment models emphasizing revision, reflection, and ethical reasoning align with heritage’s interpretive demands more closely than speed or output volume. Institutions benefit from formal ties to educational programs through co-created materials, community glossaries, and internship pathways [4]. Audiences encounter heritage narratives crafted with explicit attention to plural meaning, learning to navigate complexity without retreat to simplistic consensus.

Ethics and Governance

Ethical practice in AI-mediated, multilingual heritage curation rests on consent, provenance, transparency, and shared authority. Consent must attach to images, texts, and oral histories with persistence surviving copying and distribution [6]. Provenance documentation should remain inseparable from synthetic reconstructions and algorithmically derived summaries. Transparency includes clear explanation of recommendation logics, translation strategies, and model limitations in accessible language [5; 8].

Shared authority requires governance bodies including community representatives empowered to influence decisions rather than merely advise [4]. Data colonialism concerns necessitate licensing regimes aligned with community values, mirroring strategies across jurisdictions to protect archives against geopolitical shocks, and adoption of open standards to reduce dependency on single vendors [2; 13].

Bias audits for models become part of curatorial routine, with findings publicly reported. Moderation policies should be written and enacted in local idioms to avoid tone mismatch and inadvertent suppression [15]. Institutions must invest in capacity building for co-cu-ration, recognizing that participatory practices demand time and sustained resource allocation.

Conclusion

Digital re-actualization of architectural heritage unfolds today inside platform environments that do not merely host narratives but actively shape their visibility, tempo, and tone. Across the cases considered, AI support enlarges what can be seen, grouped, and translated, yet the same enlargement produces predictable distortions: the interface invites users along the most legible routes, recommendation logics reward coherence and visual ease, and multilingual “smoothness” can erase the local grain of speech, the discomfort of contested memory, and the careful distinction between evidence and conjecture.

The article’s contribution is to describe this situation as a governed interpretive process rather than as a technical upgrade. When AI outputs are treated as proposals – useful precisely because they remain corrigible – translation choices, moderation norms, and certainty markers become part of curatorial responsibility and can be designed, taught, and audited.

The combination of interface reading, comparative platform observation, and classroom-based co-cura-tion clarifies how such responsibility can be organized in practice: not as a promise of neutral automation, but as a workflow that leaves room for correction, preserves multilingual non-equivalence where it matters, and makes public legitimacy depend on transparent mediation.

Further research should follow how specific sites “live” across platforms over time, test how different forms of translation transparency influence trust and participation, and examine which institutional arrangements keep co-curation viable when projects move from pilot enthusiasm to routine maintenance.