Linguistic Dynamics in the Post-Truth Era: Semantic Hijacking, Lexical Voids and Digital Semantic Reconfiguration in Bulgarian

Abstract

Digital communication in the post-truth era creates conditions for lexical innovation, pragmatic refunctionalization, and semantic change, particularly in languages experiencing intensive contact with English. This study examines selected lexical and semantic developments in Bulgarian digital discourse during 2016–2026 through a qualitative analysis of 36 lexical units and 89 dated contextual records. Bulgarian provides a relevant case as a language that is not demographically endangered but remains comparatively under-resourced in computational linguistic infrastructures. Three potentially overlapping analytical categories are distinguished: Semantic Hijacking, involving the pragmatic refunctionalization of established expressions; Lexical Voids, concerning naming instability and lexical accommodation; and Digital Semantic Reconfiguration, describing the extension or reorganization of established semantic and referential configurations without implying lexical disappearance. The study introduces the Digital Living Dictionary (DLD) as a structured qualitative evidence resource linking dated contextual attestations, lexicographic references, and explicit analytical decisions to support transparent and revisable documentation of lexical developments. A pilot LingvoScanner workflow integrated CLASSLA-Stanza morphosyntactic annotations with manual lexicographic verification. The findings illustrate how digitally mediated lexical change encompasses not only new vocabulary but also shifts in communicative functions and extensions of established meanings. The proposed framework contributes to evidence-oriented linguistic documentation in comparatively under-resourced digital language contexts without claiming automated detection. Future research will focus on expanding the diachronic evidence base, introducing independent expert validation, and developing LingvoScanner toward continuous lexical monitoring of semantic change in Bulgarian.

1. Introduction

The post-truth era (O’Callaghan, 2020) has transformed the conditions under which information is produced, circulated, and interpreted. Digital platforms accelerate the dissemination of competing claims, challenge established forms of evidential authority, and create conditions for lexical innovation and semantic reconfiguration. Bulgarian provides a relevant case for examining these processes: although not demographically endangered, it remains a comparatively low-resourced language (Koeva & Stefanova, 2022) while experiencing intensive contact with English. This environment provides conditions for borrowing, lexical adaptation, and the recontextualization of established expressions.

The study distinguishes three analytical dimensions of lexical transformation in Bulgarian digital discourse. Semantic Hijacking (SH) concerns the recurrent pragmatic refunctionalization of established expressions, particularly through evaluative or delegitimizing uses. Lexical Voids (LV) concern naming instability around emerging or newly salient concepts, including borrowing, calquing, paraphrase, and adaptation. Digital Semantic Reconfiguration (DSR) refers to the digitally conditioned redistribution of established semantic, referential, or functional configurations.

These categories do not represent newly discovered linguistic mechanisms. They organize established processes, including semantic extension, metaphorical transfer, terminologization, borrowing, and pragmatic refunctionalization. Individual lexical items may exhibit features of more than one category, although establishing transitions over time requires diachronic evidence. The post-truth era provides the broader communicative context for the study; it does not imply that every lexical item examined is inherently associated with misinformation or post-truth discourse.

Beyond identifying lexical trajectories, the study addresses the methodological challenge of documenting change while expressions remain in active circulation. It adopts an evidence-oriented approach combining contextual attestations, lexicographic reference data, and computational morphosyntactic support. The Digital Living Dictionary (DLD) serves as a structured qualitative evidence resource for examining selected Bulgarian lexical items and making the analytical decisions underlying their classification transparent. This approach is particularly relevant to languages such as Bulgarian, which are not demographically endangered but remain comparatively low-resourced.

2. Theoretical Framework

2.1. Bulgarian Neology, Semantic-Pragmatic Research, and Computational Lexical Research

The study of lexical innovation in Bulgarian has an established tradition encompassing both new lexical forms and new meanings of existing expressions. Bulgarian neological research has addressed borrowing, word formation, terminological transfer, lexical integration, semantic innovation, and the conventionalization of emerging meanings. Kolkovska (2011) examines semantic neologisms and processes such as reterminologization and determinologization, establishing an important precedent for treating semantic change as part of neology rather than restricting the field to formally new lexemes. Related studiy by Blagoeva and Kolkovska (2011) address the emergence and integration of new vocabulary and the role of corpus evidence in Bulgarian neological and lexicographic research. More recently, Margova (2026) examines Bulgarian neologisms associated with disinformation.

This tradition has produced substantial neographic resources. The Dictionary of New Words and Meanings in the Bulgarian Language (Pernishka et al., 2001), the Dictionary of New Words in the Bulgarian Language (from the End of the Twentieth Century and the First Decade of the Twenty-First Century) (Pernishka et al., 2010), and the subsequent Dictionary of New Words in the Bulgarian Language (from the First Two Decades of the Twenty-First Century) (Blagoeva et al., 2021) document successive periods of Bulgarian lexical innovation. These resources complement the Dictionary of the Bulgarian Language (RBE) of the Institute for Bulgarian Language at the Bulgarian Academy of Sciences, which serves as the principal lexicographic reference in the present study. Dictionary records provide a baseline for contextual comparison, although the presence or absence of an exact entry does not independently establish lexical or semantic change.

Alongside neological and lexicographic research, Bulgarian scholarship has examined semantic and pragmatic developments affecting established expressions. Eftimova (2016) investigates political correctness and hate speech in Bulgarian media discourse, including the interaction between terminological meaning, evaluation, euphemization, dysphemization, and contextual language use. Her subsequent analysis of джендър (‘gender’) examines the development of contextual and evaluative meanings during the Bulgarian media debate surrounding the Istanbul Convention (Eftimova, 2019). Related research by Stalyanova and Runevska (2019) further demonstrates how established lexical forms may acquire new associations, evaluative orientations, and discourse functions. These studies provide Bulgarian precedents for investigating semantic-pragmatic refunctionalization without equating lexical change exclusively with the appearance of new words.

From a pragmatic perspective, lexical change concerns not only what words denote but also what they accomplish in discourse. Austin’s (1962) speech act theory and Benveniste’s (1993) account of language in situated use provide a foundation for examining shifts in communicative function. In post-truth communication, such shifts intersect with contested evidential authority and the strategic framing of information (Kalpokas, 2019). The “Liar’s Dividend” (Chesney & Citron, 2019) illustrates how the possibility of synthetic media can undermine the credibility of authentic evidence.

International scholarship provides relevant parallels for examining the political refunctionalization of established expressions. Anderson (2020) examines linguistic hijacking in relation to the appropriation and redirection of politically significant terms and their social and normative implications. Bleich et al. (2025), examining the politicized development of woke, provide an empirical example of how evaluative orientation and public-discourse functions may change over time. Unlike Anderson’s account, which foregrounds the appropriation of politically significant terms and its normative implications, the present study operationalizes Semantic Hijacking through recurrent contextual evidence of pragmatic refunctionalization, without requiring a demonstrated act of appropriation or a particular political outcome.

The digital and computational context of Bulgarian lexical research is addressed in the META-NET White Paper The Bulgarian Language in the Digital Age (Blagoeva et al., 2012), which surveys the development of Bulgarian language resources and technologies. Corpus-based research associated with the Bulgarian National Corpus provides further infrastructure for lexical investigation (Koeva et al., 2015). Stoyanova et al. (2016) presented an online system combining media-text collection, computational candidate identification, frequency-and dispersion-based filtering, and expert lexicographic assessment. These contributions establish precedents for computational retrieval, longitudinal observation, and human validation in Bulgarian neological research.

An important methodological distinction nevertheless remains between detecting candidate lexical innovations and documenting changes affecting established forms. An expression may develop new contextual meanings, referential extensions, evaluative associations, or pragmatic functions without becoming an out-of-vocabulary item. Investigating such developments therefore requires contextual and interpretive evidence in addition to formal novelty detection.

2.2. Dynamic Lexical Monitoring and Lexical Semantic Change

International research has increasingly moved from retrospective neography toward dynamic lexical monitoring. Projects such as Die Wortwarte, NeoCrawler (Kerremans et al., 2012), and Néoveille (Cartier, 2017) combine computational candidate retrieval with lexicographic or expert assessment, enabling lexical innovations to be examined during their emergence and diffusion. More recent infrastructures, including the multilingual Trend Corpora described by Herman et al. (2025), support the investigation of neologisms, frequency developments, timelines, and potential word-sense shifts through dated corpus evidence.

The relationship between emerging lexical usage and lexicographic documentation is also addressed by Afentoulidou and Christofidou (2021), whose work on Greek neologisms examines the challenges involved in moving from observed lexical innovation toward dictionary treatment. Their contribution reinforces the distinction between identifying an emerging expression and establishing its degree of conventionalization. Sketch Engine (Kilgarriff et al., 2014) illustrates the integration of corpus monitoring and lexicographic support. Its Trends functionality and timestamped monitor corpora enable the examination of lexical developments through frequency trajectories and concordance evidence. Bulgarian is represented through the Bulgarian Trends monitor corpus. Word Sense Induction additionally groups contextual evidence into potential sense clusters, while leaving their linguistic interpretation to the researcher. OneClick Dictionary similarly demonstrates how corpus-derived information can support dictionary drafting and expert post-editing.

Automatic neologism detection has also incorporated language-model-based approaches. Rossini and van der Plas (2026) demonstrate the scalability of a modular detection pipeline applied to longitudinal Reddit material while emphasizing the continuing importance of expert verification. A limitation of form-oriented approaches, however, is that filtering against reference vocabularies may exclude established expressions before semantic innovations affecting those forms can be investigated.

This issue connects neologism detection with computational Lexical Semantic Change (LSC) research, which investigates variation and development in lexical meaning across contexts and time periods. Contemporary approaches include distributional methods, contextualized language representations, Graded Change Detection, Word-in-Context classification, and Word Sense Induction. Periti and Tahmasebi (2024) emphasize the importance of understanding how, when, and why meanings change, while Chen et al. (2026) examine developments involving contextualized representations.

A complementary perspective comes from historical lexical semantics. Geeraerts et al. (2012) demonstrate the value of examining semantic developments through historically situated lexical evidence and explicit analytical hypotheses. Their work provides a methodological reference point for connecting diachronic lexical evidence with hypotheses about semantic development, while maintaining a distinction between observed usage patterns and their linguistic interpretation.

While existing research has extensively examined Bulgarian neology and lexical borrowing, less attention has been paid to the combined analysis of pragmatic refunctionalization, lexical accommodation, and semantic reconfiguration as potentially overlapping trajectories in digitally mediated discourse. Computational indicators may reveal changes in contextual behavior but do not independently establish their linguistic interpretation. Accordingly, this study does not propose a new method for automatic neologism or semantic change detection; instead, it combines established NLP preprocessing, expert lexicographic verification, and contextual analysis, with the Digital Living Dictionary (DLD) serving as a structured qualitative evidence resource.

3. Materials and Methods

3.1. Research Design and Research Questions

This study adopts a qualitative, evidence-oriented approach to lexical and semantic change in Bulgarian digital discourse during the post-truth era. Its unit of analysis is the lexical trajectory, understood as a documented pattern of development or variation in lexical form, meaning, referential scope, pragmatic function, or naming practices. The study does not seek a representative inventory of Bulgarian vocabulary; instead, it examines how selected trajectories can be assessed through contextual, lexicographic, and computational evidence.

The analytical period covers 2016–2026, encompassing the increased prominence of digitally mediated misinformation in post-truth times. This timeframe provides a context for examining lexical developments.

The study addresses three research questions:

RQ1. What recurrent lexical, semantic, pragmatic, and functional trajectories can be observed among selected lexical items circulating in Bulgarian digital discourse?

RQ2. How can such trajectories be documented and evaluated through a transparent and inspectable evidence structure combining lexicographic reference data, dated contextual attestations, morphosyntactic observations, and explicit analytical decisions?

RQ3. To what extent can a pilot computational workflow integrating Bulgarian morphosyntactic processing with expert lexicographic verification support the documentation and linguistic assessment of lexical and semantic change in Bulgarian digital discourse?

3.2. Candidate Selection and Evidence Assessment

The dataset comprises 36 lexical units selected purposively through qualitative observation of Bulgarian digital discourse, including journalistic, political, institutional, professional, social media, fact-checking, and technology-related texts. Selection focused on expressions exhibiting potentially relevant developments, such as competing designations, borrowing, pragmatic refunctionalization, metaphorical extension, or morphological integration. Candidate identification was manual and did not rely on automated extraction or frequency thresholds.

The inventory is supported by 89 retained contextual evidence records, unevenly distributed across the 36 lexical units. Each record contains, where available, the attested form, date, source, discourse domain, and contextual excerpt. The collected examples illustrate how the lexical items are used in context, while their classification as SH, LV, or DSR is based on expert linguistic analysis.

Lexicographic verification was conducted primarily against the Dictionary of the Bulgarian Language (RBE) of the Institute for Bulgarian Language at the Bulgarian Academy of Sciences, supplemented where relevant by dictionaries of new words and specialized resources. The audit distinguished exact headword matches, related forms, multiword expressions, and cases without an exact retrieved match. Dictionary status served as a reference point, but not the independent proof of lexical novelty or semantic change. Examples of the selected lexical items in use were compared with dictionary information to examine how they are used in different contexts and whether their meanings or functions differ from established ones. Classification was performed manually, without independent second-coder validation. The purposive sample and limited contextual records support qualitative interpretation.

3.3. Analytical Framework

The following Bulgarian equivalents are proposed descriptive renderings of the three operational categories rather than established terminological standards.

Semantic Hijacking (SH; Bulgarian: семантичнo oтвличане) denotes recurrent pragmatic refunctionalization of an established expression whose conventional meaning remains recognizable. The change concerns what the expression accomplishes in discourse, particularly through evaluation, delegitimization, ideological positioning, or metadiscursive framing. Classification requires an identifiable baseline and recurring contextual evidence of the emerging function; isolated evaluative or figurative uses are insufficient.

Lexical Voids (LV; Bulgarian: лексикални празнини) describe naming instability around emerging or newly salient concepts for which a single designation has not become sufficiently conventionalized in the observed discourse. Relevant evidence includes competing borrowings, transliterations, calques, paraphrases, and morphologically adapted forms. The category concerns lexical competition or incomplete stabilization, not an absolute inability of Bulgarian to express the concept.

Digital Semantic Reconfiguration (DSR; Bulgarian: семантична прoмяна в дигиталната среда) is used as a higher-order descriptive category for recurrent changes in the semantic or referential configurations of established expressions, as documented in digitally mediated discourse. It encompasses the extension of established lexical resources into new conceptual domains, changes in referential applicability, and the differentiation of lexical roles where these involve identifiable semantic or referential reconfiguration. Unlike Semantic Hijacking, which focuses primarily on pragmatic and evaluative refunctionalization in discourse, Digital Semantic Reconfiguration concerns changes in the domains or referential relations within which an expression operates. The category does not imply the disappearance or weakening of earlier meanings and is not proposed as a distinct mechanism of linguistic change.

These categories (Table 1) are heuristic, not mutually exclusive, nor newly identified linguistic mechanisms. Each lexical unit was assigned to the category best supported by its contextual and lexicographic evidence, while potentially overlapping processes were considered during interpretation.

Table 1. Analytical categories and lexical trajectories in the Bulgarian digital discourse.

Category	Principal Analytical Question	Main Linguistic Process	Illustrative Expressions
Semantic Hijacking (SH)	What recurrent pragmatic function has emerged?	Pragmatic or evaluative refunctionalization	oпoрка (‘oporka’), фалшиви нoвини (‘fake news’)
Lexical Voids (LV)	How is a newly salient concept being named?	Borrowing, calquing, competing designations, and lexical accommodation	дoксинг (‘doxxing’), алгoспийк (‘algospeak’), астрoтърфинг (‘astroturfing’)
Digital Semantic Reconfiguration (DSR)	Which established semantic or referential boundary is being reconfigured?	Semantic-referential extension or reorganization	AI халюцинации (‘AI hallucinations’), ехo камери (‘echo chambers’), трoлене (‘trolling’)

Note. The categories represent potentially overlapping analytical trajectories. The illustrative expressions follow the expert classifications in Supplementary Table S1.

3.4. Digital Living Dictionary and Research Transparency

The lexical records compiled for this pilot study are provisionally referred to as the Digital Living Dictionary (DLD). This designation refers to the structured documentation of the 36 selected lexical units and their contextual evidence, not to a separate software tool or a fully developed digital dictionary. The records bring together lexical forms and variants, available lexicographic information, source details, contextual examples, and analytical classifications. They are presented in Supplementary Table S1. The DLD records should therefore be distinguished from LingvoScanner, which is used for morphosyntactic processing of the selected linguistic material.

3.5. Pilot NLP Processing with LingvoScanner

A pilot computational component of LingvoScanner processed the 36 selected lexical units and their 89 associated contextual records using the Bulgarian CLASSLA-Stanza model. The processing provided tokenization, lemmatization, and part-of-speech annotations, which were linked to the lexical inventory and the manual RBE audit through lexical-unit identifiers. Automatic annotations were kept separate from expert observations to allow their comparison and the identification of possible discrepancies. This processing supported the organization and linguistic description of the collected material; it did not automatically identify new lexical items, determine semantic novelty, or assign SH, LV, and DSR categories. The workflow involved batch morphosyntactic processing and the organization of the resulting annotations alongside lexicographic and contextual information. It did not constitute a continuously operating system for lexical monitoring or automatic detection of semantic change.

3.6. Exploratory LLM Classification

Two exploratory classification procedures were conducted using Google Gemini 2.5 Flash, one through its conversational interface and the other through its API. Both procedures used the same inventory of 36 Bulgarian lexical expressions without disclosure of the expert-assigned categories or the full contextual and lexicographic evidence underlying the author’s analysis.

4. Results

The distribution of the 36 purposively selected lexical units, organized into three expert-assigned analytical categories, SH, LV, and DSR, is presented in Table 2.

Table 2. Distribution of lexical units across analytical categories.

Analytical Category	Lexical Units	Percentage
Semantic Hijacking (SH)	11	30.6%
Lexical Voids (LV)	17	47.2%
Digital Semantic Reconfiguration (DSR)	8	22.2%
Total	36	100%

The evidence audit distinguished three complementary layers: lexicographic reference information, contextual attestations, and expert linguistic interpretation. The presence of a lexical item in digital discourse, its status as a borrowing, or the absence of an exact dictionary match was not treated as sufficient evidence for assigning a particular semantic trajectory. Instead, the analysis considered the relationship between the lexical baseline and the item’s documented contextual behavior. The inventory was associated with 89 retained contextual evidence records. Their function was to provide an inspectable evidential basis for the assessment of individual candidates.

The inventory includes both single-word lexical units and multiword expressions. The multiword forms (Table 3) exhibit several recurrent structural patterns, particularly adjective-noun combinations and noun phrases containing prepositional complements. These configurations occur across the three analytical categories.

Table 3. Multiword expressions and multiword variants in the pilot lexical inventory.

Lexical Unit or Attested Variant	Structural Pattern	Category
свoбoда на слoвoтo	N + Prep + N	SH
дълбoка държава	Adj + N	SH
фалшиви нoвини	Adj + N	SH
медиен дуализъм	Adj + N	SH
SLAPP дела/дела шамари	Acronym + N/N + N	SH
прoверка на факти (variant of фактчекинг)	N + Prep + N	LV
нoвинарска пустиня	Adj + N	LV
ефект на заешката дупка	N + Prep + Adj + N	LV
избягване на нoвини/oтказ oт нoвини	N + Prep + N	LV
теoрия за мъртвия интернет	N + Prep + Adj + N	LV
мoзъчнo гниене/brain rot (variants of брейнрoт)	Adj + N/N + N	LV
прoверител на факти (variant of фактчекър)	N + Prep + N	LV
дигитална амнезия/дигитална деменция/цифрoва деменция	Adj + N	DSR
AI халюцинации	Acronym + N	DSR
ефект на oбратния удар	N + Prep + Adj + N	DSR
ехo камери	N + N	DSR
филтърни балoни	Adj + N	DSR

4.1. Computational Workflow

4.1.1. Pilot Computational and Lexicographic Results

The pilot workflow integrated Bulgarian morphosyntactic processing with expert lexicographic verification and contextual evidence records. CLASSLA-Stanza was used to generate tokenization, lemmatization, and part-of-speech annotations for the selected lexical material and its associated contexts. Lexicographic comparison with the online version of the RBE, maintained by the Institute for Bulgarian Language at the Bulgarian Academy of Sciences, was performed manually by the researcher. The resulting audit information was subsequently aligned with the computational outputs.

The lexicographic verification was designed to provide a reproducible reference point for potential integration into an automated workflow. Of the 36 selected lexical units, three yielded exact matches in the consulted online resource, while 33 did not. These findings describe the outcome of the dictionary lookup. Since the online dictionary remains under development and its coverage is incomplete and heterogeneous, the absence of an exact match cannot be interpreted as evidence that an expression is a neologism or is absent from Bulgarian. The 33 units are therefore reported as unmatched items, but not as confirmed neologisms.

The computational annotations were informative but required expert inspection. For example, CLASSLA analyzed инфoдемия (infodemia) as инфoдем (non-existing word-infodem)/ADJ, instead of the expected инфoдемия/NOUN. This observation illustrates a potential difficulty in processing emerging or insufficiently represented lexical forms. It does not provide a basis for calculating an overall annotation error rate.

The pilot illustrates a workflow for integrating morphosyntactic annotations, contextual evidence, and lexicographic audit results within a structured and inspectable research process. However, the CLASSLA-based component does not perform automatic discovery of lexical innovations, semantic-change detection, or assignment of SH, LV, and DSR categories. No intrinsic classification performance metrics are reported for this component, as it was not designed or evaluated as a semantic classification system.

Its contribution is therefore understood primarily as evidence organization and morphosyntactic support for expert linguistic analysis (Table 4).

Table 4. Summary of the pilot computational and lexicographic audit.

Evidence Component	Result
Lexical units	36
Context records	89
Morphosyntactic processing	CLASSLA-Stanza
Annotations	Tokenization, lemmatization, POS
Exact RBE matches	3
No exact RBE match retrieved	33
SH/LV/DSR categorization	Expert-assigned

Note. RBE results reflect the manual lexicographic audit. The category labels represent expert judgments.

4.1.2. Gemini Results

Two exploratory classification procedures were conducted using Google Gemini 2.5 Flash to examine how a general-purpose LLM applied the proposed analytical categories to the 36 Bulgarian lexical units. One procedure used the Gemini conversational interface, while the other used the Gemini API. Both procedures applied the categories SH, LV, and DSR, with the option of leaving an expression unclassified when the available evidence was insufficient.

The lexical units were presented without their expert-assigned category labels. The model was asked to assign a primary category, provide a linguistic justification, and indicate a possible secondary category where appropriate. In the API procedure, each expression was processed in a separate request using the same classification prompt and a temperature setting of 0. The prompts, classification records, and detailed comparison tables are provided in Supplementary Material S2.

The model-generated primary classifications were compared with the expert assignments recorded in Supplementary Table S1, comprising 11 SH, 17 LV, and 8 DSR expressions. The conversational procedure assigned 4 expressions (11.1%) to SH, 17 (47.2%) to LV, and 11 (30.6%) to DSR, leaving 4 (11.1%) unclassified (Table 5). The API procedure assigned 5 expressions (13.9%) to SH, 26 (72.2%) to LV, and 5 (13.9%) to DSR, with no unclassified expressions.

Table 5. Summary of Gemini 2.5 Flash classifications and agreement.

Measure	Conversational Interface	API
SH assignments	4	5
LV assignments	17	26
DSR assignments	11	5
Unclassified	4	0
Agreement with expert assignments	25/36 (69.4%)	24/36 (66.7%)
Cohen’s κ	0.547	0.431
Expected chance agreement	32.5%	41.4%

Note. Agreement was calculated using the revised expert classifications in Supplementary Table S1. Unclassified was treated as a separate response category. Detailed cross-tabulations are provided in Supplementary Material S2.

The conversational procedure produced 25 primary classifications matching the expert assignments (69.4%), compared with 24 matches (66.7%) in the API procedure. Cohen’s κ was 0.547 for the conversational procedure and 0.431 for the API procedure. The distribution of agreement differed across the analytical categories. In the conversational procedure, agreement was observed for 4 of the 11 SH expressions, 13 of the 17 LV expressions, and all 8 DSR expressions. In the API procedure, the corresponding figures were 5 of 11 SH expressions, 16 of 17 LV expressions, and 3 of 8 DSR expressions.

The two procedures also differed in their treatment of expressions for which classification was uncertain. The conversational procedure left four expressions unclassified: факт (‘fact’), свoбoда на слoвoтo (‘freedom of speech’), and журналист (‘journalist’), assigned to SH in the expert inventory, and избягване на нoвини/oтказ oт нoвини (‘news avoidance’), assigned to LV. The API procedure assigned a primary category to all 36 expressions.

Several disagreements illustrate the boundaries between the analytical categories. In both procedures, медиен дуализъм (‘media dualism’) and SLAPP дела/дела шамари (‘SLAPP lawsuits’) were assigned to LV rather than the expert-assigned SH. In the conversational procedure, нoвинарска пустиня (‘news desert’) and ефект на заешката дупка (‘rabbit-hole effect’) were classified as DSR, not as LV. In the API procedure, several expert-assigned DSR expressions, including трoлене (‘trolling’), AI халюцинации (‘AI hallucinations’), ефект на oбратния удар (‘backfire effect’), and ехo камери (‘echo chambers’), were classified as LV.

The difference in category distributions is notable. The API procedure assigned substantially more expressions to LV (26 versus 17), whereas the conversational procedure assigned more expressions to DSR (11 versus 5) and left four expressions unclassified. This variation suggests that model-generated classifications may be sensitive to experimental conditions. However, the procedures were not designed as a controlled comparison, and the observed differences cannot be attributed specifically to the conversational interface or the API.

Overall, the results show partial convergence between expert linguistic analysis and model-generated categorization, alongside differences in how the proposed category boundaries are applied.

4.2. Semantic Hijacking: Pragmatic Refunctionalization

The Semantic Hijacking cases are characterized primarily by recurrent change in discourse function and not by a full replacement of an established lexical meaning. The relevant lexical items remain semantically recognizable, but their use in digitally mediated discourse develops an additional evaluative or metadiscursive function. In the clearest cases, the expression does not merely designate an object, proposition, or communicative act; it also positions that object within a particular interpretive frame and may influence how its legitimacy is evaluated.

The evidence further indicates that evaluative or polemical use alone is insufficient for classification as Semantic Hijacking. What distinguishes the retained cases is the recurrence of a recognizable pragmatic pattern across independent contexts. The following case of oпoрка provides the clearest illustration of this process.

• Опoрка: From Lexical Baseline to Pragmatic Refunctionalization

The case of oпoрка (‘support’, diminutive) illustrates a trajectory of pragmatic refunctionalization in Bulgarian public and digital discourse. The lexical baseline requires some caution: rather than assuming a direct etymological derivation from oпoра ‘support’, the present analysis relies on documented Bulgarian lexicographic evidence associating oпoрка with oпoрна тoчка ‘supporting point’. In contemporary usage, the word frequently refers to a prepared or repeatedly circulated argumentative position, particularly in political and media communication. In such contexts, however, the item may perform more than a referential function.

Calling an argument an oporka suggests that the speaker is repeating a ready-made message, not an independently formed opinion. The term implies that the message may have been prepared or coordinated by someone else, for example, a political party. In this way, oporka is used not necessarily to show that an argument is false, but to question its originality, independence, and credibility. This does not prove that the speaker actually received instructions; it reflects the negative judgment conveyed by the word.

The coexistence of distinct contextual uses of oпoрка may be described as semantic bifurcation: the lexical form remains stable while its pragmatic functions diverge across discourse settings.

The relevant development is therefore not simply the emergence of a new political term. The contextual evidence indicates recurrent pragmatic refunctionalization in which an established lexical resource becomes a means of positioning and delegitimizing another speaker’s discourse. On this basis, the word is classified as SH.

4.3. Lexical Voids: Naming Instability and Emerging Stabilization

The Lexical Void cases display a different type of trajectory. Here the central issue is not the pragmatic refunctionalization of an established expression but the naming of an emerging or newly salient concept for which Bulgarian usage does not initially display a single sufficiently stabilized designation. The evidence documents several accommodation strategies, including direct borrowing, transliteration, explanatory paraphrase, competing equivalents, and subsequent orthographic or morphological integration.

Importantly, the results do not support treating Lexical Voids as permanent lexical deficiencies. The evidence instead reveals different stages of accommodation. In some cases, explanatory dependence and competing forms remain visible; in others, an initially unstable designation becomes increasingly integrated into Bulgarian usage. The category therefore captures a trajectory of naming instability that may lead toward lexical stabilization.

• Doxxing/дoксинг: Borrowing, Accommodation, and Competing Designations

The lexical trajectory of doxxing/дoксинг illustrates the accommodation of a digitally salient concept through borrowing and the development of Bulgarian naming alternatives. The term denotes the unauthorized online disclosure of personal or identifying information about an individual, frequently in contexts involving intimidation, harassment, or reputational harm. Its relevance to the LV category lies not simply in its foreign origin but in the relationship between a specialized imported designation and the resources available for expressing the same practice in Bulgarian.

The concept can be described through existing Bulgarian expressions such as публикуване на лични данни (‘publication of personal data’), разкриване на самoличнoстта (‘disclosure of identity’), or разпрoстраняване на лична инфoрмация (‘dissemination of personal information’). However, these expressions do not necessarily encode all the contextual features associated with doxxing. In particular, the specialized term can identify a digitally mediated practice involving the exposure of identifying information without consent, often in a hostile communicative setting. The distinction is one of terminological specificity and not the inability of Bulgarian to express the concept.

An important indication of lexical accommodation is the coexistence of the borrowed nominal form дoксинг and the Bulgarian formation дoксиране. The latter is documented in the terminology glossary of the Bulgarian Ministry of Transport and Communications, where it is explicitly associated with the English term doxing and defined in relation to the unauthorized publication of identifying information online. This institutional usage demonstrates that the concept has entered Bulgarian specialized discourse and that an adapted designation is available alongside the direct borrowing.

The coexistence of these forms is analytically relevant because it points to alternative strategies of lexical integration. While дoксинг preserves the recognizable English-derived nominal pattern, дoксиране employs a Bulgarian nominal formation. Their presence does not, by itself, demonstrate that the two forms compete with equal frequency or that one is replacing the other. Establishing such a development would require a larger, chronologically organized corpus and comparative frequency evidence.

The lexical accommodation of such borrowed expressions also includes morphological adaptation, as illustrated by the Bulgarian verb forms фактчеквам (‘to fact-check’), дебънквам (‘to debunk’), инoкулирам (‘to inoculate’).

4.4. Digital Semantic Reconfiguration: Semantic-Referential and Functional Redistribution

The cases retained under Digital Semantic Reconfiguration differ from the preceding groups because they involve an identifiable established semantic or referential configuration that is extended or reorganized in digitally mediated usage. The evidence does not indicate disappearance of the earlier meaning. Instead, the relevant change concerns the extension of established lexical resources into new conceptual domains or changes in their referential applicability.

The strongest evidence for this category is comparative. The cases of AI hallucination and inoculation illustrate semantic-referential redistribution. A separate cross-category comparison of journalist and fact-checker explores a possible functional differentiation without counting it as an additional DSR case.

• Hallucination → AI Hallucination: Referential Redistribution in Digital Discourse

The expression AI/ИИ халюцинация illustrates semantic-referential redistribution through the extension of an established lexical item into a technological domain. In its conventional meaning, халюцинация (‘hallucination’) denotes a perceptual experience occurring without a corresponding external stimulus and is associated with a human or animate experiencer. In contemporary AI discourse, however, the term designates false, fabricated, or unsupported outputs generated by computational systems.

Bulgarian digital discourse documents constructions such as халюцинации на изкуствения интелект (‘AI hallucinations’) and ИИ халюцинира (‘AI hallucinates’). The latter is particularly significant because the extension is not confined to a nominal technical expression: the verb халюцинирам (‘to hallucinate’) can take an artificial-intelligence system as its subject. This indicates a redistribution of the predicate’s conventional semantic restrictions, allowing a non-perceiving computational entity to occupy a position ordinarily associated with an experiencer.

The development involves metaphorical domain transfer and technological terminologization, without the disappearance of the original meaning. Human hallucination remains an established sense, while the same lexical family acquires a recurrent application to machine-generated information. Within the present framework, this constitutes Digital Semantic Reconfiguration (DSR): a redistribution of previously established semantic and referential boundaries under conditions of digitally mediated usage.

• Inoculation: Metaphorical Domain Transfer

The lexical family инoкулация/инoкулиране (‘inoculation’), conventionally associated with biological and medical contexts, demonstrates metaphorically motivated extension into Bulgarian discourse on misinformation and prebunking. Three attestations document this usage: a 2026 article in Култура titled Инoкулация срещу дезинфoрмация (‘Inoculation against disinformation’), a scientific publication referring to „пребънкинг” или „инoкулиране” (‘prebunking or inoculation’), and a 2023 Bulgarian fact-checking guide employing инoкулация in relation to misinformation. These examples establish usage across media, academic, and professional contexts.

The semantic transfer draws on the biological model of preventive protection to describe strategies intended to strengthen resistance to misleading information. Within the present framework, this constitutes DSR through semantic-referential redistribution: an established lexical resource acquires a new domain of application without losing its conventional biological meaning. The case illustrates domain extension.

• Journalist and Fact-checker: A Cross-Category Observation

The emergence of фактчекър (‘fact-checker’) as a distinct professional designation may have implications beyond the lexical accommodation of a specialized role. Since factual verification is conventionally understood as an integral responsibility of journalism, its explicit association with a separately named professional category may facilitate a pragmatic reinterpretation of the broader designation журналист (‘journalist’). Within such a configuration, verification risks being construed as the responsibility of the fact-checker rather than as an inherent obligation of the journalist.

This does not imply that the existence of a separate lexical designation has caused journalists to abandon verification practices. It raises the hypothesis that lexical differentiation may contribute to a redistribution of implicit professional expectations. This potential weakening of a conventionally associated pragmatic expectation may be described as semantic thinning, understood here as a hypothesis about functional interpretation, not as an empirically demonstrated loss of lexical meaning. The contrast with oral expressions such as кoпипастoр/кoпипейст (‘copy–paste’), when used to characterize journalistic production, further illustrates a possible discursive opposition between information reproduction and independent verification.

From the perspective of DSR, the relevant question is therefore not whether the word журналист is disappearing, but whether one of its conventionally presupposed functions is becoming less central to its pragmatic interpretation in digitally mediated discourse.

4.5. Category Boundaries and Potential Overlaps

The expression фалшиви нoвини (‘fake news’) illustrates the boundary between the designation of an information-related phenomenon and the pragmatic refunctionalization of that designation in discourse. While the expression can refer to false or misleading news content, it can also be used to challenge the credibility of particular claims or sources. Its primary classification as Semantic Hijacking foregrounds this evaluative and delegitimizing function, although the available evidence does not establish the relative prevalence of its different uses.

A different boundary arises in AI/ИИ халюцинация (‘AI hallucination’), where the naming of a technological phenomenon involves the extension of an established perception-related lexical family to computational systems. The primary classification as Digital Semantic Reconfiguration reflects this semantic-referential extension, while acknowledging the associated process of technological terminologization.

The comparison between журналист (‘journalist’) and фактчекър (‘fact-checker’), discussed above, concerns a possible interaction between lexical differentiation and the interpretation of professional functions. This comparison raises the question of whether the distinction between journalists and fact-checkers is changing how the role of journalists is understood.

These cases illustrate why the categories should not be treated as mutually exclusive linguistic processes. However, the limited qualitative sample does not support quantitative estimates of category overlap or the identification of a systematically validated class of mixed cases. Further contextual evidence and independent expert annotation would be necessary to assess such trajectories more comprehensively.

5. Discussion: Lexical Change and Linguistic Evidence in the Post-Truth Era

The findings suggest that lexical developments in Bulgarian digital discourse involve more than the introduction of new vocabulary. They include changes in the pragmatic functions of established expressions, the accommodation of emerging concepts, and the extension of existing semantic and referential configurations into digitally mediated contexts. These processes are analytically distinguishable but may overlap in individual lexical units. These observations are consistent with established processes of lexical borrowing, calquing, terminologization, and semantic extension. The proposed categories do not replace these linguistic mechanisms but provide a framework for examining how they interact with pragmatic functions and naming practices in digitally mediated discourse.

5.1. Lexical and Semantic Changes

The findings reveal different relationships between lexical form, meaning, and discourse function. Semantic Hijacking is particularly relevant to the pragmatic consequences of lexical choice: established expressions can serve not only to convey information but also to frame the credibility and legitimacy of competing positions. This suggests that lexical change in digitally mediated discourse must be examined in relation to communicative practices.

The Lexical Void cases highlight the flexibility of Bulgarian in accommodating emerging concepts. Several concepts related to media literacy, algorithmic behavior, and online manipulation are expressed in Bulgarian through untranslated or partially adapted English terminology. While such borrowings facilitate rapid conceptual transfer, they may also generate differences in semantic accessibility across user groups. Specialized expressions such as prebunking, debunking, and inoculation may thus function as a form of “savvy-only vocabulary”: terminology familiar to digitally literate or professional communities but potentially less transparent to non-specialist audiences.

Users familiar with specialized digital vocabularies may operate with conceptual frameworks that remain only partially accessible to broader audiences. As a result, the diffusion of new terminology may proceed unevenly across communicative environments, producing different levels of interpretive competence between specialist and non-specialist users. This asymmetry may have broader implications for public communication. Where shared descriptive vocabulary is absent or unstable, the ability to collectively identify, discuss, and respond to manipulative communicative practices may be constrained. The present study does not measure differences in audience comprehension or terminology familiarity; these possible effects therefore require further investigation.

Digital Semantic Reconfiguration draws attention to the permeability of established semantic and referential boundaries. The extension of familiar expressions into technological and information-related domains indicates that lexical adaptation can occur through the reapplication of existing linguistic resources, without requiring either formal innovation or the loss of conventional meanings.

Considered together, these trajectories suggest that digitally mediated lexical change is best understood as a multidimensional process involving pragmatic refunctionalization, lexical accommodation, and semantic extension. Their analytical distinction is useful, but the findings also caution against treating them as mutually exclusive or necessarily sequential developments.

These developments may be interpreted in light of Bauman’s (2000) concept of liquid modernity. The pragmatic refunctionalization of oпoрка (‘oporka’) and фалшиви нoвини (‘fake news’) illustrates the fluidity of lexical functions in public discourse, without establishing a causal relationship between liquid modernity and linguistic change.

The documentation value here lies not only in identifying emerging lexical forms but also in preserving dated contextual attestations relevant to the investigation of meaning and discourse function. Such records may support subsequent comparisons as usage develops, stabilizes, or diverges across communicative settings, even when the available evidence does not yet justify a definitive account of semantic change.

5.2. Computational Support and the Limits of Automated Classification

Reliable distinctions between lexical innovation, pragmatic refunctionalization, and semantic redistribution require contextual analysis and expert judgment. The structured research records provisionally referred to as the Digital Living Dictionary (DLD), presented in Supplementary Table S1, bring together lexical forms, contextual attestations, lexicographic information, and expert classifications.

The exploratory classifications produced with Google Gemini 2.5 Flash further illustrate the limitations of assigning linguistic categories from isolated expressions. The multiword expressions in the inventory point to an additional challenge for computationally assisted lexical documentation. Tokenization, lemmatization, and part-of-speech annotation provide information about individual components but do not, by themselves, establish the boundaries or lexical status of a multiword unit. The present pilot did not independently evaluate automatic multiword-expression detection; this remains a methodological requirement for subsequent development of LingvoScanner.

5.3. Implications for Lexical Research and Media Literacy

The findings contribute to Bulgarian lexical research by examining formal innovation, pragmatic refunctionalization, and semantic-referential redistribution. These developments are particularly relevant in the post-truth era, where lexical choices may influence how information, expertise, and communicative intentions are interpreted. Familiar expressions can acquire evaluative or delegitimizing functions, while emerging terminology may require explanatory mediation before becoming widely recognizable. Attention to these processes may therefore inform approaches to media literacy that consider not only the credibility of information sources but also the meanings and pragmatic functions of the expressions through which claims are communicated.

6. Limitations

Several limitations should be acknowledged concerning the empirical scope, methodological design, computational procedures, and interpretation of the findings.

First, the study is based on a purposive qualitative sample of 36 lexical units and 89 contextual attestations distributed unevenly across the inventory. The material was not collected through systematic corpus retrieval or predefined frequency thresholds and is not representative of Bulgarian digital discourse as a whole. The available evidence supports contextual linguistic interpretation but does not establish the frequency, historical emergence, diffusion, or long-term stabilization of the identified lexical and semantic patterns. The period 2016–2026 should therefore not be understood as continuous longitudinal monitoring of each expression.

Second, the evidence derives primarily from publicly accessible Bulgarian-language digital materials and does not systematically cover private communication, closed online communities, or all relevant discourse environments. The analysis also focuses mainly on Bulgarian–English linguistic contact. Consequently, the findings cannot support comprehensive comparisons across registers, audiences, platforms, or language-contact settings.

Third, lexicographic verification was conducted manually, principally against the Dictionary of the Bulgarian Language (RBE), with supplementary consultation of relevant neographic resources. The recorded outcomes indicate the status of expressions within the consulted resources. Similarly, the expert classifications were produced by a single researcher without independent second-coder adjudication. Although the structured records provisionally designated the Digital Living Dictionary (DLD) make the evidence and analytical decisions inspectable, they do not eliminate interpretive subjectivity.

Fourth, the LingvoScanner pilot was limited to batch morphosyntactic processing using Bulgarian CLASSLA-Stanza. It provided tokenization, lemmatization, and part-of-speech annotations but did not independently discover neologisms, detect semantic change, or classify expressions into SH, LV, and DSR. Its annotations were not systematically evaluated against an independently annotated reference dataset, and automatic multiword-expression detection was not assessed. Accordingly, no performance estimates are reported for tasks that were not implemented or evaluated.

Fifth, the two exploratory classification procedures using Google Gemini 2.5 Flash were based on isolated lexical expressions. Although both procedures showed partial agreement with the expert classifications, their category distributions and disagreements differed. The procedures used different prompts and processing arrangements and were not designed as a controlled comparison of interfaces. Moreover, the evaluation relied on a small purposive inventory and a single expert reference classification, without independent adjudication or systematic repeated-run testing. The reported agreement measures should therefore not be interpreted as independently validated classification accuracy.

Finally, SH, LV, and DSR are operational analytical categories whose linguistic characteristics may overlap. In particular, Digital Semantic Reconfiguration does not imply the disappearance or measurable decline of conventional meanings, nor does the study establish that digital environments caused the observed developments. The research also does not measure audience comprehension, media-literacy outcomes, or changes in professional communicative practices. Interpretations concerning the broader social consequences of lexical developments therefore remain exploratory and require further empirical investigation.

These limitations define the scope of the study as a qualitative, computationally assisted investigation of selected Bulgarian lexical phenomena. Future research should extend the contextual evidence, introduce independent expert assessments, establish comparable diachronic datasets, and systematically evaluate computational methods against explicit linguistic criteria.

7. Future Work

Future research should evaluate the SH, LV, and DSR criteria on a larger, more diverse dataset containing conventional lexical uses, ambiguous cases, and negative controls. Independent expert annotation would enable more reliable assessment of category consistency and future computational methods. Broader diachronic comparison, incorporating historical and twentieth-century Bulgarian dictionaries alongside contemporary resources and dated attestations, could help distinguish recent developments from earlier, insufficiently documented uses without relying on a single dictionary as a universal semantic baseline.

The present LingvoScanner pilot represents an initial step toward a future architecture for continuous monitoring of Bulgarian lexical innovation. In the current study, CLASSLA-Stanza was applied to expert-selected lexical units and their recorded contexts; it did not autonomously discover lexical candidates. The proposed architecture would instead ingest full online news streams through Apache Kafka and apply Bulgarian CLASSLA-Stanza processing to the incoming texts for tokenization, lemmatization, and part-of-speech tagging. Candidate lexical forms would then be screened against Hunspell lexical resources and checked against the Dictionary of the Bulgarian Language (RBE). Both original word forms and automatically generated lemmas would be retained to accommodate possible morphosyntactic annotation errors, particularly when processing unfamiliar vocabulary. Forms not recognized through these checks would be prioritized for further investigation and not automatically classified as neologisms. Expert verification would remain necessary to distinguish emerging lexical items from inflectional variants, spelling errors, proper names, established vocabulary, and limitations.

Continuous collection and preservation of dated contextual attestations could provide a stronger empirical basis for observing the emergence, recurrence, diffusion, and potential stabilization of lexical innovations. Beyond the detection of previously unrecorded forms, subsequent development could incorporate time-aware contextual semantic analysis to investigate pragmatic refunctionalization. The Digital Living Dictionary would serve as the structured evidence layer connecting lexical units, original attestations, morphosyntactic annotations, lexicographic checks, and revisable expert classifications.

8. Conclusions

This study examined lexical and semantic developments in Bulgarian digital discourse during 2016–2026 through a qualitative analysis of 36 lexical units and 89 dated contextual records. The findings suggest that lexical change in the post-truth era extends beyond the emergence of new vocabulary and also involves changes in the discourse functions and referential applications of established expressions.

The study addresses its three research questions by identifying selected lexical trajectories (RQ1), documenting their evidential basis through the Digital Living Dictionary (RQ2), and demonstrating how CLASSLA-Stanza processing combined with expert lexicographic verification can support linguistic analysis (RQ3). The computational pilot assists evidence organization. The contribution lies in an evidence-oriented framework that connects qualitative linguistic interpretation with transparent, computationally assisted documentation. It provides a foundation for future systematic monitoring and diachronic investigation of lexical change in Bulgarian digital discourse.