Exploring anti-Armenian hate speech in Azerbaijani state-aligned news outlets: A post-conflict discourse analysis

Abstract

This research paper studies anti-Armenian hate speech narratives in three Azerbaijani state-aligned news outlets during the early post-2025 TRIPP Peace Treaty period—a critical period for peace-building. Although the TRIPP Peace Treaty officially ended the Nagorno-Karabakh conflict, Azerbaijani state-aligned news outlets continue to construct anti-Armenian narratives. Existing research mainly focuses on wartime media discourse, leaving a gap in understanding post-conflict hate speech in the understudied South Caucasus region. Hence, this study attempts to fill the gap by examining three Azerbaijani state-aligned news outlets—Azertac, AZTV, and Azernews—during the early post-TRIPP period. This study incorporates a hybrid framework to assess hate speech intensity and examine linguistic, semantic, syntactic, and rhetorical mechanisms in constructing anti-Armenian hate speech. Quantitative findings reveal a hierarchy of hate speech intensity, with Azertac the highest (M = 0.57) and Azernews the lowest (M = 0.31). Qualitative findings demonstrate all three outlets apply van Dijk’s ideological square framework to construct hostile identities.

Keywords

anti-Armenian hate speech Azerbaijani state-aligned news outlets post-conflict discourse TRIPP peace treaty south Caucasus

Introduction

The Nagorno-Karabakh conflict has been one of the most protracted conflicts in the modern world history dating back to 1991. After the collapse of the Soviet Union, the conflict escalated between Armenia and Azerbaijan for more than three decades, which culminated in political and military developments in 2023, resulting in the mass displacement of the Armenian population from the region. On 8 August, 2025 Armenia and Azerbaijan signed a peace treaty (The Trump Route for International Peace and Prosperity-TRIPP; U.S. Department of State & Government of Armenia, 2026), officially marking the end of the long-lasting conflict and the transition to normalization of the South Caucasus region. This event was an important milestone for both countries. While the Peace Treaty is a formal step for conflict resolution, Azerbaijani state-aligned media outlets continue to cover narratives of victimhood, aggression, and historical grievance. Much research has been conducted on Azerbaijani media and conflict, providing insightful outcomes on the role of state-aligned news outlets in legitimizing military actions and shaping public opinion during conflict (Azerbaijan.az, n.d); however, few studies have integrated quantitative hate speech detection with qualitative discourse analysis to study how Azerbaijani state-aligned news outlets construct anti-Armenian hate narratives. Existing literature primarily focuses on anti-Armenian narratives in active conflict times (de Waal, 2021; O’Loughlin et al., 2019) leaving a significant research gap in understanding post-treaty media coverage and the persistence of ethnicized hate speech against Armenians. Filling this research gap is important for understanding how Azerbaijani media outlets shape public opinion, national ideology, and influence collective memory after formal conflict resolution. By applying a mixed-method design, the following research paper not only evaluates the intensity of hate speech across Azertac, AZTV and Azernews news outlets but also examines linguistic mechanisms through which anti-Armenian hate speech is constructed. Moreover, extensive research conducted on hate speech focuses on Western media on active-conflict periods, while the role of ethnicized media discourse in the South Caucasus region remains underexplored. This study attempts to address these interconnected research gaps by: comparing multiple state-aligned news outlets to evaluate variation in hostility intensity during the post-conflict period, and by integrating natural language processing, corpus linguistics, and ANOVA statistical testing to provide a well-structured and solid methodology. The contribution of this study to the existing literature is threefold: first, it shifts research focus from wartime media discourse to post-conflict period, second, it advances political hate speech methodology by combining automated-detection techniques with qualitative analysis, third, the research contributes to wider theoretical discussions on post-conflict nationalism and durability of adversary identities in the South Caucasus region. Hence, this research paper seeks to answer the following research questions:

RQ 1: What is the intensity of anti-Armenian hate speech in three Azerbaijani state-aligned news outlets (Azertac, AZTV and Azernews) during the early post-2025 TRIPP peace treaty period, and how does the intensity rate differ across the selected news outlets?

RQ 2: What lexical, syntactic, semantic, rhetorical, and other linguistic mechanisms are used by the selected Azerbaijani state-aligned news outlets to form anti-Armenian narratives in the early post-2025 TRIPP peace treaty period?

Theoretical background: Media, state-aligned discourse and ideology

Media discourse plays an important role in shaping public opinion and national ideologies, especially in the context of ethno-political conflicts. State-aligned news outlets employ ideologically charged linguistic mechanisms to shape certain social realities (Fairclough, 1995; van Dijk, 1998). The narratives of state-aligned news outlets are closely linked to the political agendas. Walker and Orttung (2014) define state-aligned media as authoritative tools used by states to influence domestic and international audiences (Walker and Orttung, 2014). In this study, “state-aligned news outlets” refer to news agencies that are deeply influenced by governmental authority through ownership and political affiliation.

According to the propaganda model (Chomsky and Herman, 1988), media content is created based on political interests and institutional constraints sidelining dissenting viewpoints (Chomsky and Herman, 1988).

Critical discourse analysis and ideological polarization

Critical Discourse Analysis (CDA) is as a powerful tool to examine how national ideologies and political stances are realized in language (Fairclough, 1995; Wodak, 2001). In this context, media texts are perceived as channels through which ideological meanings are produced (Machin and Mayr, 2012). Therefore, the language used by media outlets that contain ideologies like political stances can be analyzed through critical discourse analysis (Liu, 2024; Reisigl and Wodak, 2009).

van Dijk’s (1991, 1998) theory of ideological polarization (often referred as ideological square) becomes prominent in the context of analyzing conflict-related media discourse. The model highlights the positive representation of in-group (us) and negative representation of the out-group (them), which is achieved through lexical selections, syntactic structures (passive and active voice), modality, presupposition, etc. In ethno-national conflicts, ideological polarization is often realized through narratives of victimhood, historical grievance, superiority, which function to legitimize state actions and delegitimize the opposing group (van Dijk, 2006). These linguistic mechanisms exist in many post-conflict contexts, where military actions are ceased, but ideological struggles continue (Fuchs, 2018).

Hate speech, dehumanization news media narratives

Hate speech is often defined as discourse that targets certain individuals or groups due to their identity markers, such as nationality, ethnicity, or religion (Parekh, 2012; Wodak et al., 2001). In media discourse, hate speech does not always appear in an explicit form (Assimakopoulos et al., 2020). According to Wodak & Reisigl, discourse-historical approach, five discursive strategies are used to construct hate speech: (1) naming social actors, (2) assigning attributes, (3) justifying hostility, and (4) intensification, and (5) mitigation. These strategies enable news outlets to frame the opposing group as aggressive, illegitimate, and criminal (Wodak et al., 2001).

Corpus-based critical discourse analysis

To examine large volumes of media texts, scholars increasingly combine CDA with corpus linguistics, resulting in corpus-assisted critical discourse analysis (Baker et al., 2008). Corpus-assisted methods ensure transparency and validity of the research, while CDA provides interpretive background for revealing ideological meanings (Fortuna and Nunes, 2018). Corpus-assisted CDA has been widely applied to analyze representations of ethnic minorities, migrants, and conflict actors, revealing how repeated lexical patterns contribute to the normalization of ideological frames (Baker et al., 2013).

Computational Hate Speech Detection and Hybrid Methodologies: the application of machine-learning models to detect and measure the intensity of discursive hostility has become widespread in NLP (Sirulhaq et al., 2023). While machine-learning models cannot substitute qualitative interpretation, they still serve as critical quantitative markers of hostility. By integrating intensity rates of discursive hostility with micro-linguistic analyses, researchers can not only identify how much hate speech is persistent, but also understand how news outlets are linguistically constructed.

To study anti-Armenian hate speech in Azerbaijani state-aligned news outlets, our research team combined Chomsky and Herman’s propaganda model (1988), CDA (Fairclough, 1995; van Dijk, 1991), and corpus-assisted linguistic methods (Baker et al., 2008). This research paper enriches existing literature by examining hate speech in a post-conflict environment and provides a deeper understanding of how ethno-national hate speech is constructed and maintained discursively after conflict resolution.

Methodology

The following study employs mixed-method research design incorporating both quantitative (HateBERT, ANOVA) and qualitative approaches (Corpus-driven Critical Discourse Analysis-CD-CDA). According to Creswell & Inoue, mixed-methods research design contains the innovative feature of integration in data analysis by connecting the databases to derive additional information beyond the results learned from each database (Creswell and Inoue, 2025).

Data collection

The first step of this research was to collect data from Azerbaijani news outlets. Our research team collected data from three purposefully selected state-aligned news outlets:

Azertac (Azerbaijan State News Agency – national news agency, domestic/international (mainly domestic-oriented), multilingual),

AZTV (Azerbaijan Television – state television, domestic/international, multilingual),

Azernews (state newspaper, domestic/international (mainly international-oriented), multilingual).

We selected these outlets because they are the most circulated and influential state-aligned news platforms, making them crucial for studying how state-aligned news outlets form anti-Armenian hate speech (Sukiasyan, 2025). The selection also complies with the theoretical perspective of Chomsky and Herman’s (1988) “propaganda model,” which highlights the importance of elite state-aligned media in shaping ideological agendas.

Timeline and keyword sampling

Our data collection spanned from August 8, 2025 (the official date of the trilateral treaty signing) to December 31, 2025, a crucial early phase examining the immediate post-treaty discursive patterns before long-term transition to reconciliation.

During this period, our research team retrieved all relevant textual content from the official websites of Azertac, AZTV, and Azernews. Using keyword search and filtering functions, we gathered all articles explicitly referring to Armenia, Nagorno-Karabakh, and the TRIPP Treaty. A total of 174 news articles were collected (71 from Azertac, 55 from AZTV, and 48 from Azernews). We selected news articles only referring to Armenia/Armenians and the Nagorno-Karabakh conflict.

Sentiment analysis: HateBERT

Next, a sentiment analysis was conducted to evaluate the intensity of anti-Armenian discursive hostility in Azerbaijani state-aligned news outlets. The HateBERT model was employed to detect and quantify the intensity of discursive hostility. HateBERT is a re-trained BERT architecture explicitly optimized for abusive language detection in English (Davidov et al., 2020). Initially, HateBERT model was trained for social media hate speech detection, however, nowadays, it is considered to be one of the most commonly used and methodologically open models for hate speech detection in online discourse analysis (Caselli et al., 2021). Our research team does not take HateBERT results as definite and objective measures of hate speech, but as a quantitative complement of hostility in language. The determination of hate speech remains context-based, qualitative and theoretically grounded in CDA.

Although these Azerbaijani news outlets are published in Azerbaijani, English, and Russian, our analysis focused on English content. This choice was also determined by the linguistic limitations of the HateBERT model, as the model detects only English texts. While this model’s limitation may not reflect nuances of anti-Armenian hate speech in other languages, English versions are designed for an international audience, making them relevant for studying how state-aligned news narratives are reflected for international readers. Hence, the findings cannot apply to Azerbaijani and Russian content. Future studies could employ multilingual hate speech detection models to assess content in Azerbaijani and Russian.

The intensity of discursive hostility was measured on a 0–1 scale, where 0 represents no hate speech and 1 represents extreme hate speech, consistent with HateBERT’s gradient output.

To ensure data accuracy and consistency, all collected news articles underwent initial screening. Non-textual elements were removed. After preprocessing, each news article was analyzed for hostility intensity. The mean hostility score for each outlet was then calculated. Median and standard deviation (SD) were also measured for descriptive statistics.

Statistical testing

To calculate statistically significant differences in discursive hostility intensity across the selected news outlets, our research team employed one-way ANOVA and Tukey’s HSD Post-Hoc Tests. ANOVA test measures the overall effect of news outlet on hate speech intensity (dependent variable: hate speech score; independent variable: news outlet (Azertac, AZTV, Azernews). While Tukey’s HSD Post-Hoc Test: to determine pairwise significant differences between outlets (a post-hoc test was employed).

All statistical tests were conducted in R (v4.3.1) with a significance level of p < 0.001 (a stringent threshold Effect size (η²) was measured for the ANOVA test to assess the degree of variance in hostility intensity accounted for by news outlets.

Corpus-based discourse analysis: Voyant tools

While ANOVA test results illustrate the intensity of discursive hostility, it is also important to analyze how these news outlets linguistically construct their narratives. To complement the quantitative hate speech analysis, a corpus-based linguistic analysis was employed to analyze linguistic and discursive patterns of Azertac, AZTV and Azernews in framing Armenia, Armenians, and Nagorno-Karabakh after the TRIPP Peace Treaty. This corpus-assisted linguistic analysis was conducted by employing Voyant Tools, which enabled us to determine key lexical elements, frequency counts, and elicit significant discursive and framing devices within news articles (van Leeuwen, 2008).

The following methodological approach aligns with CDA (Fairclough, 1995; van Dijk, 1991), which underscores the role of language in forming national ideologies and social realities. By studying recurring lexical patterns, metaphors, and syntactic structures, the analysis identifies how state-aligned news outlets not only disseminate explicit anti-Armenian hate speech but also position certain political and ethnic narratives.

Corpus preprocessing for CDA

All 174 English-language news articles were collected in 3 mini-corpora (Azertac, AZTV, Azernews) and preprocessed in Voyant Tools.

Tokenization: Text was split into individual words/tokens.

Stop word removal: To focus on meaningful lexical items, we removed all English stop words (e.g. the, and, of).

Lemmatization: To avoid redundancy of inflected forms, words were set to their base form.

Proper noun retention: All proper nouns remained.

The high frequency words, collocations and discursive framing of each mini-corpus are presented in Tables 1 to 3.

Table 1.

Descriptive statistics for anti-Armenian hostility scores by media outlet.

Media outlet	Number of articles (n)	Mean hate speech score	Median	Standard deviation
Azertac	71	0.57	0.59	0.18
AZTV	55	0.43	0.42	0.16
Azernews	48	0.31	0.30	0.14

Table 2.

Tukey’s HSD post-hoc test results.

Comparison group	Mean difference	Standard error	95%confidence interval	p-value
Azertac vs AZTV	0.14	0.2	(0.10, 0.18)	<0.001
AZTV vs Azernews	0.12	0.2	(0.08, 0.16)	<0.001
Azertac vs Azernews	0.12	0.2	(0.22, 0.30)	<0.001

Table 3.

Key high-frequency words: initial framing based on word cloud (Azertac).

Rank	Terms	TPM	Collocation examples	Framing elevation
1.	Azerbaijan	8.90	Azerbaijan-glorious victory, martyrs, liberated territories, the great patriotic war	Glorification of National Identity: Positioned Azerbaijan as a powerful and heroic country that liberated its occupied lands by framing it as a patriotic duty.
2.	Violent	8.90	Violent + acts, Armenian violent aggressive acts, provocative	Delegitimization: Frames Armenian actions cruel and provocative to humanity, associates violence and terrorism to Armenian actors.
3.	Separatist	8.90	Separatist + separatist units, separatists units	Claim to international legitimacy: presents Armenia as separatist force, which actions contradict international law.
4.	Criminal	7.50	Criminal + acts, war criminals, Armenian war criminals, Nagorno-Karabakh, Armenian officials	Criminalization of the Enemy: entitles the governmental officials of Nagorno-Karabakh, rich and famous people of Armenian origin as war criminals (Samvel Karapetyan, Ruben Vardanyan).
5.	Aggressive	6.20	Aggressive + aggressive war, conduct of an aggressive war, Armenian military actions	Aggressor/victim frames: presents Armenia as an initiator of conflict, and positions Azerbaijan as a victimized party.
6.	War	5.70	War + aggressive war, patriotic war	US vs Them: presents Azerbaijani acts as patriotism and Armenian acts as aggression, liberator/aggressor.
7.	International	4.90	International + violation of the international law, international humanitarian law, international protection	International Normative Accordance: makes Nagorno-Karabakh conflict as a violation of international law, refugees as victims of the international law violation, and justifies Azerbaijani actions in accord with international law.
8.	State	4.30	State + state prosecutor, crossing the state border, border violation	Territorial Legitimacy: Armenian military actions were considered as violation of Azerbaijani statehood.
9.	Army	4.10	Army + illegal Armenian army groups, military aggression	Delegitimization of military actions: labels Armenian military forces as “illegal.”
10.	Terrorism	3.89	Terrorism + Armenian citizens, acts, crime against humanity, genocide, inhumanity	Dehumanization: associates all Armenian military actions as acts of terrorism and genocide, justifies devastating countermeasures.

CDA analytical steps

For each mini-corpus, the following analytical steps were conducted:

Lexical frequency analysis: We determined the 10 most frequent content words (tokens per 1000 words (TPM)).

Lexical clustering analysis: To identify thematic framings, we pinpointed recurring lexical chains (e.g. armed + terrorist + violent).

Qualitative CDA: Lexical, syntactic, semantic, and rhetorical mechanisms of hate speech construction were presented in alignment with van Dijk’s (1991) ideological square and Wodak et al.’s (2001) DHA.

Results: Quantitative findings

As shown in Table 1, Azertac has the highest intensity of anti-Armenian hostility (M = 0.57), followed by AZTV (M = 0.43), while Azernews shows the lowest intensity (M = 0.31).

Next, our research team used one-way ANOVA test to measure whether the observed differences in mean scores across the three news outlets were statistically significant. The ANOVA results indicated a significant effect of news outlet on hate speech intensity, F (2, 153) = 41.27, p < 0.001, η² = 0.351 (Cohen, 1988), demonstrating that 35.1% of the variance in hate speech intensity is explained by the news outlet. To identify pairwise differences, Tukey’s HSD post-hoc tests were conducted.

As shown in Table 2, Azertac (M = 0.57) showcases higher hate speech scores than both AZTV (M = 0.43), p < 0.001, 95%CI (0.10, 0.18), and Azernews (M = 0.31), p < 0.001, 95%CI (0.22, 0.30). AZTV (M = 0.43) also scored significantly higher than Azernews (M = 0.31), p < 0.001, 95%CI (0.08, 0.16). These results clearly reflect the hierarchical differences in anti-Armenian hate speech intensity rate across Azerbaijani state-aligned news outlets: Azertac >AZTV > Azernews. The hierarchy fulfills the theoretical argument (Chomsky and Herman’s (1988) propaganda model, van Dijk’s (1991) critical discourse framework, and Fairclough’s (1995 ideology-in-text approach) that news outlets aligned with state policy excessively vary in disseminating and shaping ethnicized hate speech.

Qualitative findings-Azertac mini-corpus (research question 2)

The first mini-corpus analysis consists of news articles from Azertac with a total of 8693 tokens, and 1589 unique word forms. The size of each term indicates its normalized frequency (tokens per 1,000words, TPM). The high frequency terms are: Azerbaijan, criminal, violation, terrorism, international law, aggressive, etc., (See Table 3). This terminology uncovers core thematic focus of news articles: framing Azerbaijan as a victim and Armenia an aggressor, which is consistent with van Dijk’s ideological polarization theory (van Dijk, 1991).

The examples presented in the Micro-Linguistic Analysis section should not be seen as direct or isolated instances of hate speech. Rather, they illustrate how recurring patterns in Azerbaijani news media gradually normalize hostility—through moral delegitimization, repetition, negative other-representation, irony, and threat construction. In this study, hate speech is understood not as direct, individual insults, but as something that evolves continuously over time.

Azertac: Micro-linguistic analysis

Lexical Analysis: Azertac news outlet frequently labels Armenia and Armenians with the pejorative adjectives criminal, aggressive and violent in all late 2025 coverage. These negative linguistic labels are used to describe not only actions taken place in Nagorno-Karabakh conflict time, but also implicitly apply to Armenian identity, and no mitigating lexemes (e.g. “accused,” “alleged”) are used to neutralize the judgment (Fairclough, 1995).

The criminal Armenian sabotage units have once again carried out violent raids on Azerbaijani border settlements, confirming their ongoing commitment to terrorizing. . . (2025, October 12)

Violent Armenian militants have breached border lines for the third time this month, their criminal acts endangering the lives of Azerbaijani women and children . . . (2025, November 28)

This turns the lexical environment of Azertac news outlet asymmetric: Armenia and Armenians are positioned only through negative, and accusatory language elements, while Azerbaijan is framed only with positive implications.

News articles referring to Armenia and Armenians are full of legal terminology: termination of OSCE Mink group, international law, Armenia’s military aggression against Azerbaijan, war crimes, crimes against humanity, genocide, aggression, international murderer, criminal entrepreneurship, terrorism, violations of the laws and customs of war, financing terrorism, and the violent seizure and retention of power.

The dominance of legal terminology is especially obvious in news articles covering the trial and court hearings of Nagorno-Karabakh’s previous governmental officials and leaders, who are subjected to trial in Azerbaijan after 2023 events.

The overuse of evaluative adjectives (glorious victory, monumental data, etc.) is another rhetorical and lexical data to encode subjective stance. Unlike neutral descriptive adjectives, evaluative adjectives easily attach positive and negative connotations and impacts reader’s emotional interpretation of the events (van Dijk, 1998).

8 November 2025 marks the fifth anniversary of Victory Day. This monumental date, etched into the hearts of Azerbaijanis, commemorates the glorious victory achieved during the 44-day Patriotic War and serves as an enduring symbol of national strength. . .(November 7, 2025)

Supreme Commander-in-Chief H.E. Ilham Aliyev, and with the unprecedented bravery of our armed forces. . .(November 7, 2025)

Syntactic Framing: Active Accusation & Passive Victimhood: Azertac’s mini-corpus is marked with a fixed syntactic patterns: passive voice is employed to refer Azerbaijan and Azerbaijani as victimized survivors, while active voice constructions are used to label Armenian actions intentional, destructive and provocative.

Azerbaijani border checkpoints and civilian homes were targeted by Armenian armed groups, who launched unprovoked violent attacks on unarmed Azerbaijani residents in the Kalbajar region. (2025, August 18)

Critical Azerbaijani agricultural infrastructure was destroyed by Armenian raiders, who intentionally set fire to crop fields and water pumps serving rural Azerbaijani communities. (2025, September 15)

Modality : To balance journalistic objectivity, and guide readers’ interpretations, Azertac news outlet utilizes many modal verbs: must, should, etc. Modal verbs are often paired with legal terms to eliminate the subjectivity and make news articles objective and factual.

Armenia must respect human rights and international law and ensure the right of Azerbaijanis . . . (December 1, 2025)

First of all, it should be noted that compensation means recognition of the offense committed and the losses incurred. . . (October 11, 2025)

Semantic-Pragmatic Unpacking: Connotative Overload (Wodak et al., 2001): the usage of terminology in Azertac news outlet is marked with heavy connotation. Many terms override the literal definitions. Majority of terms are positioned as unsubstantiated, and presupposed facts without providing evidence-based context.

The Armenian regime’s long history of genocide against Azerbaijani civilians has re-emerged in its latest cross-border attacks, a criminal pattern of violence that the international community cannot ignore. (2025, November 2)

Armenia’s systemic campaign of terror against Azerbaijani minorities in Karabakh has resurfaced, its genocidal tactics unchanged and its disregard for human life absolute. (2025, December 7)

As shown in the above-mentioned examples, phrases like “long history” and “systemic campaign” label this harm as a regular, inherent characteristic of the Armenian identity, rather than situational event.

Thematic Lexical Clustering: the lexical chain armed + terrorist + violent + aggressor + war criminal is a repetitive and dominant framing of Armenian and Armenians in late 2025 Azertac coverage. Moreover, this chain is paired with another frequently-used cluster (violating + international law/ceasefire), associating moral delegitimization (terrorism/violence/aggression) with legal delegitimization (international law/human rights breach) to build a solid linguistic loop that depicts anti-Armenian hate in a bilateral normative frame.

Armed Armenian terrorist groups have launched a violent assault on Azerbaijani border outposts, violating international humanitarian law and breaking the UN-brokered peace accord. (2025, October 29)

Rhetoric: Us/them Polarization: an absolute us-them binary is pervasive in the majority of Azertac’s articles. Armenia and Armenians are consistently labeled as a permanent “criminal state” “violators of human rights”, “terrorists”, and “aggressors”. Meanwhile, Azerbaijan is framed as a “defender,” “upholder” of international humanitarian norms, liberator, and peace-seeker. Verbs like persists and endures attach Armenian actions as an enduring, unchanging implications, and position anti-Armenian claims as objective fact.

Azerbaijan stands as an unwavering defender of international law and territorial sovereignty, while Armenia persists as a criminal state that continues to terrorize innocent Azerbaijani civilians and breach global peace agreements. (2025, September 29)

Azerbaijan remains a steadfast upholder of global peace and human rights, while Armenia endures as a violent criminal state that terrorizes Azerbaijani civilians. (2025, November 15)

Second mini-corpus: AZTV news outlet

The second mini-corpus analyzes articles of AZTV news outlet. It has 22,894 tokens, out of which 2237 are unique word forms. This corpus consists of 22,894 total words and 2237 unique word forms. The readability index is 19.08, and average words per sentence is 30.6, indicating a high frequency of syntactic complexity peculiar to formal political discourse. The most frequent words are Azerbaijan, Armenian, armed, district, and forces (See Table 4).

Lexical Peculiarities: AZTV is marked with the abundance of legal (territorial integrity, sovereign territory, international law) and military terminology (armed, forces, fortifications, military equipment). This lexical concentration come in line with what van Dijk (1998) identifies as the strategic use of institutional authority: a strategy to legitimize state actions through media discourse.

The territorial integrity of the Republic of Azerbaijan is the cornerstone of international peace and security. Any attempt to question this inviolable right contradicts the UN Charter and the 44-day peace framework, and Azerbaijan will take all necessary measures to defend its sovereign lands, including the Nagorno-Karabakh district, which has been an integral part of our state since 1918. (Dec 8, 2025)

Table 4.

Key high-frequency words: initial framing based on word cloud: AZTV.

Rank	Terms	TPM	Collocation examples	Framing elevation
1	Azerbaijan	8.71	Azerbaijan + triumphant conquest, liberation campaign	National Heroization and Glorification: AZTV frames Azerbaijani attacks as historic mission to retake its occupied territories, national duty.
2	Armenia (Armenian)	8.55	Armenian + Armenian occupiers, separatist units, Armenian occupation forces	Dehumanizing Enemy: Armenia and Armenians are labeled as occupying, destructive and separatist forces, violation of Azerbaijan’s territorial integrity.
3	Republic	8.17	Republic + Nagorno-Karabakh illegal regime, puppet republican structure, unrecognized republic	Illegitimacy of the Republic of Nagorno-Karabakh: AZTV consistently names the Republic of Nagorno-Karabakh as illegal, a separatist republic that caused instability in the region.
4	Armed	7.64	Armed + Armenian armed separatist groups, armed provocations, armed militant infrastructure	Violence Legitimization: AZTV grounds Azerbaijani actions as legal defense, while Armenian armed presence in Nagorno-Karabakh illegal and provocative.
5	District	7.47	District + liberated district, occupied district, district reintegration	Territorial Reunification: Nagorno-Karabakh district liberation is marked as a major keystone for Azerbaijani national reunification after 30 many years.
6	Forces	7.28	Forces + Azerbaijan’s liberation forces, Armenian occupation forces, counteroffensive forces	Sovereignty Enforcement: AZTv frames the operation of Azerbaijani armed forces as the legitimate enforcement of territorial integrity.
7	State	7.09	State + state sovereignty restoration, state constitutional order, state territorial integrity	Sovereignty restoration: AZTV news outlet scales Nagorno-Karabakh events as a legal and moral mission of liberating Azerbaijani state territory.
8	Occupied	6.82	Occupied + occupied territories, occupied district settlements	Occupation Criticism: AZTV frames Armenian actions as occupation of Azerbaijani territory, illegal.
9	War	6.64	War + patriotic war, war causalities, war effort	Historicized Conflict: AZTV describes the second Armenia-Azerbaijan war as liberation war.
10	October	6.27	October + October liberation, October ceasefire agreement, October war anniversary	Day of Great Victory, commemoration Day: AZTV positions October as one of the best time in Azerbaijani history, marking it as the start of new, reintegrated and liberated Azerbaijan.

Alike Azertac, many AZTV news articles also frame Armenia and Armenians through negative labels: aggressor, separatist units, occupants, etc., reinforcing us/them dichotomy.

In AZTV news articles there is an explicit reference to international laws or acts (Helsinki Final Act, trilateral peace treaty, UN, Vienna Convention on the law of Treaties) and districts (Khalbajar, Lachin, Fizuli). This linguistic strategy reflects Chomsky and Herman’s (1988) propaganda model, where application of legal terminology naturalizes state narratives and minimize alternative interpretations.

Armenian forces continue to occupy Azerbaijani territories in the Jabrayil district, violating the territorial integrity of our republic. The international community must condemn this occupation and uphold the principles of the Helsinki Final Act. . . (November 18, 2025)

Emotionally charged lexicon: Suffering, blood, martyrs (evaluative nouns) evokes collective grief and solidarity, while the abundance use of personal pronouns construct a shared national narrative unifying audiences around territorial restoration (Lemmens et al., 2021).

Armenian claims to Nagorno-Karabakh are rooted in historical revisionism, not fact. Our republic was founded on the principle of territorial integrity, and every generation of Azerbaijanis has fought to defend this principle—from the battles of 1918 to the liberation of 2020. This is our legacy, and it is non-negotiable. (Sep 5, 2025 )

Emotionally charged lexicon is also filled with the abundance of evaluative adjectives and metaphors: bright future, golden letters, iron fist, beautiful houses:

The bright victory achieved under leadership of the Commander-in-Chief Ilham Aliyev has been written in the history of Azerbaijan in golden letters. (October 25, 2025)

The borders of Aghali are becoming even larger. Beautiful houses and buildings have been constructed for 180 families in the village of Mammadbayli. (October 25, 2025)

Modality Deployment: modal verbs play a central role in AZTV news articles. To justify Azerbaijani actions and claims in align with international law, human rights, deontic modality elements (shall, must, will, inviolable) are used in AZTV news articles.

In this example must carries a normative obligation on the international law.

The international community must not only condemn Armenia’s occupation of Azerbaijani territories . . .(September 18, 2025)

The peace framework shall serve as the foundation for resolving the conflict, and Armenians shall fulfill its obligations. . .(August 30, 2025).

Syntactic Patterns: Passive-Active Voice/long and complex sentences: the syntactic choices of the news outlet, passive constructions and long sentences, neutralize Azerbaijani state claims and actions, turning them into objective facts, and enhance the validity of anti-Armenian narratives, a technique widely tackled in CDA (Fairclough, 1995).

Since 1918, Nagorno-Karabakh has been an integral part of Azerbaijan’s sovereign territory until the 1990s, when separatist forces-backed up by Armenian support-occupied the region. (September 22, 2025)

Long and complex sentences underscore historical continuity with multiple prepositional phrases (since 1918, from 1994, etc. Until 1990).

Rhetorics: US/Them aggressor, occupant, separatist -victim dichotomy: in many news articles Armenia is presented as an aggressor, occupant, separatist unit while Azerbaijan a victim of Armenian actions whose counteractions are justified on the basis of international law. This heavy polarization aligns with van Dijk’s ideological square, reinforcing positive self-representation and negative other-representation.

Armenian occupants have exploited natural resources in the occupied territories of Azerbaijan for decades, depriving the Azerbaijani people of their rightful economic inheritance. (September 30, 2025)

The international community must hold Armenia accountable for its role as an aggressor. (October 12, 2025)

Binary opposition is also reflected in past events marking Armenian history revisionist (negative evaluation) while Azerbaijani historical legacy (positive evaluation). This is especially obvious in all news articles referring to Azerbaijani people who went to live in Nagorno-Karabakh. It is always mentioned that Armenian destructed many historical mosques and constructions.

Separatist units backed up by Armenia have destroyed historical Azerbaijani cultural sites in Nagorno-Karabakh region, erasing evidence of Azerbaijan’s long-standing historical presence in the area. (August 17, 2025)

Semantic Overloading: Historical Grievance & Collective Identity (Bar-Tal, 2013): like Azertac, AZTV news articles also contain many heavy-connoted terms: genocide, historical injustice, etc., Majority of terms are used without providing evidentiary contextualization. To form emotional impact and solidarity, many over-connoted terms (occupation, genocide, historical injustice and collected pronouns (our people, our nation) were used.

For over three decades, Armenian forces have inflicted suffering on our people in the occupied districts. The blood of Azerbaijani civilians, soldiers, and children killed during the occupation is etched into our national memory. We will never forget these crimes, and we will restore our lands to honor the memory of our martyrs. (Dec 20, 2025)

Azernews mini corpus

The third mini-corpus of our research paper consists of news articles from Azernews media outlet. The corpus has 31,495 tokens and 4891 unique word forms. High frequency words are Armenia, Azerbaijan, war Nagorno-Karabakh and international (See Table 5).

Table 5.

Key high-frequency words: initial framing based on word cloud (Azernews).

Rank	Terms	TPM	Collocation examples	Framing elevation
1.	Armenia	8.85	Armenian military aggression, Armenian occupation of Nagorno-Karabakh region	Dehumanizing Enemy: Aggressor Armenia-Armenia is always framed as aggressor, occupant, separatist, violator of human rights. Armenian actions are characterized as the violation of international law.
2.	Azerbaijan	8.78	Liberation of Azerbaijani occupied lands, Azerbaijani’s right of self defense	Glorification of Azerbaijan as Liberator-Azerbaijani actions and hatred are explained in align with the international law. It is their right to defend and take actions.
3.	War	8.15	War criminal, patriotic war, glorious war, post-war reconstruction	Criminalization/Legitimization of the War: Patriotic War-Azerbaijan is positioned as a liberator of occupied territories. Azerbaijani actions in the war were glorious, and it takes all the proper measures to restore Armenian occupied territories.
4.	International	7.72	International law, international community	Sovereignty Enforcement: International law-Azerbaijani actions against Armenia, court hearings against Armenian officials are always described in accord with the international law, a right that Azerbaijan needs to restore.
5.	State	7.55	Sovereign state, Armenian state-sponsored terrorism	Legitimization of Azerbaijani actions: The territorial integrity of Azerbaijan-reinforced the sovereignty and legitimacy of the Azerbaijani state and positions Armenian actions as violation to the state sovereignty.
6.	Crime	7.48	War crime, crimes against humanity in Khojaly, military crimes	Criminalization Armenian Actions, War Crimes-highlights Armenian military actions not just military acts but long-lasting crimes against humanity.
7.	Peace	7.37	Post-war peace, peace negotiations, Armenian sabotage of peace negotiations	Sovereignty Enforcement: Peace-Azerbaijan’s actions were characterizes as measures for providing peace in the region, while Armenian actions aggressive and destructive to peace negotiations.
8.	Criminal	7.18	Armenian criminal groups, criminal acts of Armenia, Armenian criminal elements	Delegitimization of the existence of Nagorno-Karabakh and it previous officials, War Criminal-the officials of Nagorno-Karabakh Armenian by ethnicity that are sentenced in Azerbaijan are labeled as war criminals, criminals for collecting, guiding and sponsoring military actions against the integrity of Azerbaijan.
9.	Armed	6.85	Armenian armed forces, illegally armed Armenian separatist forces, armed attack on Azerbaijan	Violence Legitimization: Azernews labels Azerbaijani actions as legal defense, while Armenian armed actions illegal and provocative to the region.
10.	Terrorism	6.67	Terrorist attacks, Armenian terrorist infrastructure, Armenian state terrorism	Armenia state-sponsored terrorism-this labeling delegitimizes military actions carried out by Armenians, and prepare a solid background for Azerbaijani actions.

Azernews: Micro-linguistic analysis

Lexical Analysis: Azernews lexicon contains less militarized and legalized terminology in general reporting. Legal and lexical terminology becomes prominent in news covering court hearings against Nagorno-Karabakh previous officials.

Ruben Vardanyan is accused of committing crimes against peace and humanity, war crimes, as well as terrorism, financing of terrorism, and a range of other particularly serious offenses under the Criminal Code of the Republic of Azerbaijan. (December 30, 2025)

Fifteen defendants of Armenian origin are accused in the criminal case concerning numerous crimes committed during the aggressive war waged by the Armenian state on the territory of Azerbaijan, in violation of domestic and international legal norms. (December 25, 2025)

Excessive usage of evaluative-emotive adjectives is marked in all news articles: false champion of democracy and European values, deeply compromised politician, decisive and shocking patriotic war in 2020, etc.

Spanish MP Jon Iñárritu has once again emerged as an ardent propagandist for Armenia, this time advocating for the EU to support Yerevan’s bid for membership and falsely portraying Azerbaijan as an aggressor. He is a deeply compromised politician serving Armenian interests. His history of receiving honors from Garabagh separatists. . .(August 29, 2025)

Time revealed the truth even though Armenian society insistently relied on the pseudo-power of international organisations and opposition groups for many years trying to create momentum within the country after the decisive and shocking 2020 Patriotic War. (December 12, 2025)

The heavy reliance on juridical language constructs Armenian actions as criminalized facts rather than contested claims, reinforcing legal legitimacy for Azerbaijani narratives.

Terminological Choice: “Artsakh” vs. “Garabagh” Azernews is the only Azerbaijani news outlet that frequently uses the Armenian term “Artsakh,” (enclosed in quotation marks), expressing explicit irony and delegitimization. At the same time, the Azerbaijani term “Garabagh” is used without quotation marks, emphasizing territorial legitimacy. This linguistic choice of naming serves as a discursive marker of ideological stance and denial of Armenian territorial claims.

A new chorus of political theatrics has emerged in France, where forty Armenian organisations have released a joint statement demanding the return of the people of the so-called Artsakh. (13 November, 2025)

Once a separatist, he had kept his distance from the Artsakh authorities before Azerbaijan’s 2023 anti-terrorist operation, even advocating coexistence with Garabagh Azerbaijanis . (November 11, 2025)

Semantic Analysis: Unlike Azertac and AZTV, Azernews uses milder connotative terms: words like genocide, systemic terror are used randomly, and mainly in referring to certain historical events (e.g. Khojaly).

Azernews is marked with positive/negative semantic prosody: Armenia is framed with negative prosody (violence, illegitimacy), while Azerbaijan with positive prosody (justice, peace, liberation). However, it should be mentioned that this imbalance is milder than in Azertac and AZTV.

Rhetorical US/Them polarization: Armenia -like other news outlets, Azernews also uses negative linguistic elements to frame Armenia. Terms like “aggression,” “occupation,” “terrorism,” “crime,” “sabotage,” and “separatist” characterize Armenian actions as inherently unlawful and malicious to humanity. Meanwhile, positive lexicon is applied to refer Azerbaijan. Words like “liberation,” “self-defense,” “peacemaker,” “legitimate,” and “sovereign” form a narrative of legal righteousness, labeling Azerbaijan as a defender of international rights.

Armenian armed separatist groups continued their acts of sabotage against Azerbaijani civilians in the liberated Karabakh region. (October 12, 2025)

This polarization accords with van Dijk’s ideological framework, reinforcing positive self-presentation and negative other-presentation.

Repetition and Anaphora: to maintain narrative memorability and consistency, Azernews uses repetitions and anaphora (e.g. “Armenian aggression,” “Azerbaijani peace”). These stylistic devices link current actions to the past events, evoking historical and moral outrage and paving a solid ground for Azerbaijani’s stance. “Armenia’s aggression in 1992, Armenia’s aggression in 2020, Armenia’s aggression in 2025. . .”

Armenia’s aggression in 1992, Armenia’s aggression in 2020, Armenia’s aggression in 2025 – the pattern of unlawful behavior persists. (December 27, 2025)

As we honor the victims of the Khojaly massacre, we cannot ignore Armenia’s ongoing cruelty toward Azerbaijani civilians today. (December 1, 2025)

Syntactic Framing: Active/passive Voice, Nominalization: like other news outlets, Azernews also employs passive voice constructions to describe Armenian actions. Passive voice constructions position Azerbaijan in the role of victim, highlighting the harm Armenia has done to Azerbaijan. While, active voice constructions were used to justify Azerbaijani actions. (e.g. “Azerbaijan restored . . . ”)

Azerbaijan restored its territorial integrity through a swift, legally sanctioned military operation. (September 3, 2025)

Azerbaijani border posts were attacked by Armenian forces, resulting in multiple casualties. (August 15, 2025)

Paratextual strategy, lowercased tags: at the end of each article, Azernews uses some tagging practices that encode anti-Armenian stance. News articles, framing Armenia negatively, the tags such as Armenia, Armenians, the names of the Armenian leaders are written in lowercase, highlighting symbolic marginalization (Table 6).

Table 6.

Comprehensive cross-outlet comparison.

Feature	Azertac	AZTV	AzerNews	Similarities/differences
Core words (common across all)	Azerbaijan, Armenia, war, international	Azerbaijan, Armenia, war, international	Azerbaijan, Armenia, war, Nagorno-Karabakh, international	All three Azerbaijani news outlets underscore conflict and national identity.
Unique words	criminal, violent, aggressive, violation, terrorism, army	armed, district, forces, state, occupied, republic, October	peace, crime, criminal, liberation	Azertac highlights moral/legal condemnation; AZTV emphasizes military/state actions; while AzerNews stresses international framing.
HateBERT score (mean)	0.57	0.43	0.31	Azertac demonstrates highest anti-Armenian intensity; AzerNews is the lowest.
Vocabulary density	0.183	0.098	0.155	Azertac employs diverse terminology; AZTV is simpler; AzerNews is moderate
Average sentence length (words)	46.7	30.6	27.2	Azertac uses complex sentences; AZTV moderate; AzerNews short and simple.
Sentence structure/style	Long, formal, complex	Medium-length, formal, legalistic	Short, clear, concise	Azertac is ideologically dense; AZTV bureaucratic; AzerNews more readable.
Tone/framing	Aggressive, accusatory	Formal, state-centered	Neutral, international	All three news outlets cover war/conflict, but tone differs: Azertac hostile, AZTV official, AzerNews balanced
Frequent themes	Conflict, punishment, army, terrorism	Military operations, territorial control, legal/official	Peace, liberation, crime, diplomacy	Overlaps on conflict/war, morality, state action, or diplomacy.
Use of adjectives/modifiers	Strong negative adjectives (violent, criminal, aggressive)	Neutral/official adjectives (occupied, armed, state)	Mild adjectives (peaceful, liberated)	Azertac is highly emotive, AZTV is neutral/formal, AzerNews moderate
Reporting focus	Internal politics, moral framing, war crimes	State operations, military, legal framing	International audience, peace and crime reports	Focus differs ideologically vs operational vs diplomatic

Discussion

The following research paper aims to study the intensity and linguistic construction of anti-Armenian hate speech in Azerbaijani state-aligned news outlets after the TRIPP Peace treaty. By incorporating HateBERT computational discursive hostility detection model with corpus-based CD, the outcomes provide significant insights into how ethno-nationalized hate speech persists in Azerbaijani media discourse even after the TRIPP Peace Treaty.

Post-treaty persistence of anti-Armenian hate SPEECH

RQ1: The findings of quantitative analysis show that all three Azerbaijani state-aligned news outlets contain anti-Armenian hate speech during the early post-TRIPP Peace Treaty period. The observed outlets do not reproduce neutral and peace-oriented media discourse. Statistically, the hierarchy of hate speech intensity across all three outlets is significant (p < 0.001) with a large effect size (η² = 0.351), which means 35.1% of the variability depends on the news outlet. This is an important finding to understand how state-aligned media roles differ in post-conflict Azerbaijan. A novel finding of this study is that hate speech intensity is audience-centered, and the audience form hate speech intensity and the selection of linguistic elements. Azertac mainly targets domestic audience, therefore it exhibits the highest hate speech intensity (M = 0.57), whereas AZTV is a mix of domestic and international-centered audiences, hence, it shows relatively milder hate speech intensity compared to Azertac (M = 0.43). Meanwhile, Azernews is primarily international-focused, so, the hate speech intensity is the lowest (M = 0.31). This aligns with Chomsky and Herman’s (1988) propaganda model, according to which state-aligned media designs its discourse based on audience interests. To maintain collective memory and state legitimacy, aggressive and nationalistic discourse is used for domestic audiences, while for international audiences a milder, and “international law-compliant” discourse is employed (Azadian, 2025). The reproduction of hate speech in Azerbaijani state-aligned news outlets in the absence of military hostilities confirms Bar-Tal’s (2013) perspectives that ethno-national conflicts perpetuates discursively and psychologically long after official conflict resolution. The TRIPP Peace Treaty has officially ended the Nagorno-Karabakh conflict, however Azerbaijani state-aligned media has not shifted its discourse to reconciliation; instead, it continues to reproduce the adversarial us/them ideologies. The construction of hate speech in a post-treaty context shows that discursive conflict outperforms military confrontations. This finding neglects the assumption that formal peace treaties lead to instant discursive changes.

Linguistic construction of ideological polarization

RQ 2: The results of the qualitative analysis show that all three Azerbaijani news outlets build anti-Armenian narratives through a consistent integration of lexical, rhetorical, semantic, and syntactic structures. These findings are consistent with existing literature (van Dijk’s (1991) ideological square and Wodak et al.’s (2001 DHA). All three Azerbaijani news outlets share one common feature: ideological polarization. Azerbaijan is constantly positioned as liberator/victim (positive self-representation), while Armenia is framed as terrorist, aggressor, criminal, etc, (negative other-representation). This aligns with van Dijk’s (1991) ideological square.

An important theoretical finding is that van Dijk’s (1991) ideological square is realized through different linguistic mechanisms across all three news outlets.

Azertac employs the most extreme level of the ideological square, representing Armenia as a historical harm for the South Caucasus region.

AZTV employs a state-centered approach, labeling Armenia as an illegal occupier of Azerbaijani land.

Azernews uses a moderate version of the square, characterizing Armenia with negative labels in the context of international law.

Generally, van Dijk’s ideological square and Chomsky and Herman’s propaganda models are used to analyze wartime conflicts in Western media contexts (KhosraviNik, 2010; Krzyżanowski, 2010), while the findings of this study show that ideological square and propaganda model are flexible, audience-dependent and adaptable for certain roles. Hence, they can be applied in post-conflict, and non-Western (post-Soviet) contexts where state-aligned news outlets apply it to construct ethno-national hate speech even after official conflict resolution.

Another methodological novelty is that this study extends hate speech research practices by integrating corpus-assisted CDA to HateBERT and ANOVA, thus providing a hybrid and rigorous framework for examining post-conflict hate speech. HateBERT/ANOVA findings serve a statistical foundation for qualitative CDA results (e.g. Azertac is more hostile than Azernews), while the qualitative CDA findings show why hate speech intensity differs (e.g. audience). This integration narrows the limitations of single-method hate-speech detection—an important methodological contribution to hate speech research.

Implications for post-conflict communication

The results of this research have broader implications in understanding post-conflict narratives in state-aligned media settings. Even more, the systematic generation of anti-Armenian narratives may lag discursive reconciliation, potentially hindering long-term peacebuilding efforts and normalization process (Wodak, 2015).

Limitations and future research

This study is also marked with several limitations. First, the hate speech intensity analysis is confined to English-language content. Future research can use multi-lingual hate-speech detection tools. Second, the scope of the research is limited with three state-aligned news outlets, which may not completely reflect the diversity of Azerbaijani news media discourse. Third, this study focuses mainly on linguistic mechanisms, while extra-linguistic factors (geopolitics) in shaping post-media content is under-studied (Waseem and Hovy, 2016). Further research could broaden research dataset by including Azerbaijani-language texts, social media platforms, as well as, comparative post-conflict cases. Moreover, longitudinal analysis could demonstrate how hate speech elements develop over extended post-treaty periods.

Concluding remarks

In conclusion, the outcomes of this study show how anti-Armenian hate speech is reproduced in Azerbaijani state-aligned news outlets despite the official conflict resolution. Through the integration of hybrid-methodology, the findings show how linguistic mechanisms help to construct ideological polarization and maintain adversarial narratives in a post-conflict layout. By combining computational analysis and critical discourse theory, the findings of this research paper contribute to a deeper understanding of the role of media in shaping ethno-national hostility. This study advances existing CD literature in three ways: shifting the focus from active wartime domestic discourse to post-conflict international English-language discourse; (2) enriching hate speech methodology through integrating hybrid research framework, and lastly this study extends theoretical discussions on post-conflict nationalism and adversarial identity formation in non-Western media.

Footnotes

ORCID iDs

Jingjing Shi

Huichun Ning

Gevorg Grigoryan

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Discipline

Critical Discourse Analysis; Nationalism Studies; Media and Communication Studies

Author biographies

Jingjing Shi is an Associate Professor at Taizhou Vocational and Technical College and holds a Ph.D. in Education. Her research has been published in leading SSCI-indexed journals, with a core focus on the practical impact and translational value of language education research. She also serves as a peer reviewer for several top-tier international journals, where she evaluates the empirical rigor and real-world applicability of cutting-edge language education scholarship.

Huichun Ning holds a Ph.D. in Education and has accumulated extensive academic experience across higher educational institutions in China and Malaysia. Her scholarly output includes more than 20 peer-reviewed articles and research manuscripts. Her research interests center on interdisciplinary language studies, cross-cultural discourse analysis, with extended focus on contextual interpretation and linguistic representation of hate speech within Armenian cross-border cultural contexts.

Gevorg Grigoryan holds a Ph.D. in Applied Linguistics from Yerevan State University of Languages and Social Sciences and serves as a Guest Lecturer at Nanchang Polytechnic University. As a linguist and educator with a focused research agenda in critical discourse analysis, cross-cultural communication, and AI-mediated language use, he has published over 30 scientific articles in highly indexed international journals, with work centering on the intersections of language, power, and digital communication across cultural contexts.

References

Assimakopoulos

Vella Muskat

van der Plas

, et al. (2020) Annotating for hate speech: The MaNeCo corpus and some input from critical discourse analysis. arXiv. Available at: https://arxiv.org/abs/2008.06222

Azadian

(2025) The language of displacement: Rethinking legal approaches to hate speech through the 2023 Armenian exodus from Nagorno-Karabakh. Master’s Thesis, Columbia University. https://doi.org/10.7916/f0wb-0015

Azerbaijan.az (n.d) Mass media in Azerbaijan. Republic of Azerbaijan. Available at: https://azerbaijan.az/en/information/604

Baker

Gabrielatos

KhosraviNik

, et al. (2008) A useful methodological synergy? Combining critical discourse analysis and corpus linguistics to examine discourses of refugees and asylum seekers in the UK press. Discourse & Society 19(3): 273–306. https://doi.org/10.1177/0957926508088962

Baker

McEnery

Hardie

(2013) Corpus Linguistics and Critical Discourse Analysis. Palgrave Macmillan. https://doi.org/10.1057/9781137028319

Bar-Tal

(2013) Intractable Conflicts. Cambridge University Press. https://doi.org/10.1017/CBO9781139033939

Caselli

Basile

Jelena

, et al. (2021) HateBERT: Retraining BERT for abusive language detection in English. In: Proceedings of the 5th workshop on online abuse and harms, pp. 17–25. https://doi.org/10.18653/v1/2021.woah-1.3

Chomsky

Herman

(1988) Manufacturing Consent. Pantheon Books.

Cohen

(ed.) (1988) Statistical Power Analysis for the Behavioral Sciences, 2nd edn. Lawrence Erlbaum Associates.

10.

Creswell

Inoue

(2025) A process for conducting mixed methods data analysis. Journal of General and Family Medicine 26(1): 4–11. https://doi.org/10.1002/jgf2.736

11.

Davidov

Idan

Reichart

(2020) Semi-supervised recognition of sarcastic sentences in Twitter and Amazon. Computational Linguistics 46(1): 65–102. https://doi.org/10.1162/coli_a_00363

12.

de Waal

(ed.) (2021) Black Garden: Armenia and Azerbaijan Through Peace and War. New York University Press.

13.

Fairclough

(1995) Media Discourse. Edward Arnold.

14.

Fortuna

Nunes

(2018) A survey on automatic detection of hate speech. ACM Computing Surveys 51(4): 1–30.

15.

Fuchs

(2018) Nationalism on the Internet. Routledge.

16.

KhosraviNik

(2010) The representation of refugees, asylum seekers and immigrants in British newspapers. Journal of Language and Politics 9(1): 1–28.

17.

Krzyżanowski

(2010) The Discursive Construction of European Identities. Peter Lang.

18.

Lemmens

Markov

Daelemans

(2021) Improving hate speech type and target detection with hateful metaphor features. In: Proceedings of the fourth workshop on NLP for internet freedom, pp. 7–16. Association for Computational Linguistics. Available at: https://aclanthology.org/2021.nlp4if-1.2.pdf

19.

Liu

(2024) Critical discourse analysis in media studies: Power, ideology, and representation. Journal of Language and Politics 23(1): 45–63. https://doi.org/10.1075/jlp.23001.liu

20.

Machin

Mayr

(2012) How to Do Critical Discourse Analysis. Sage.

21.

O’Loughlin

Hoskins

Gillespie

(2019) Conflict, Media and Displacement. Polity Press.

22.

Parekh

(2012) Is There a Case for Banning Hate Speech. Public Policy Research.

23.

Reisigl

Wodak

(2009) The discourse-historical approach. In: Wodak

Meyer

(eds) Methods of Critical Discourse Analysis. Sage, pp.87–121.

24.

Sirulhaq

Yuwono

Muta’ali

(2023) Why do we need a sociocognitive-CDA in hate speech studies? A corpus-based systematic review. Discourse & Society 34(4): 462–484. https://doi.org/10.1177/09579265221126599

25.

Sukiasyan

(2025) Azerbaijan and accountability for crimes against Armenians (Policy brief series no. 190). Torkel Opsahl Academic EPublisher. Available at: https://www.toaep.org/pbs-pdf/190-sukiasyan/

26.

U.S. Department of State & Government of Armenia (2026) TRIPP implementation framework. Available at: https://www.state.gov/wp-content/uploads/2026/01/TRIPP-Implementation-Framework.pdf

27.

van Dijk

(1991) Racism and the Press. Routledge.

28.

van Dijk

(1998) Ideology. Sage.

29.

van Dijk

(2006) Politics, ideology, and discourse. In: Wodak

(ed.) Encyclopedia of Language and Linguistics. Elsevier, pp.728–740.

30.

van Leeuwen

(2008) Discourse and Practice: New Tools for Critical Discourse Analysis. Oxford University Press.

31.

Walker

Orttung

(2014) Breaking the news: The role of state-run media. Journal of Democracy 25(1): 71–85. https://doi.org/10.1353/jod.2014.0015

32.

Waseem

Hovy

(2016) Hateful symbols or hateful people? Predictive features for hate speech detection on Twitter. In: Proceedings of the NAACL student research workshop, pp. 88–93. https://doi.org/10.18653/v1/N16-2013

33.

Wodak

(2001) Methods of Critical Discourse Analysis. Sage.

34.

Wodak

(2015) The Politics of Fear. Sage.

35.

Wodak

Reisigl

, et al. (2001) Discourse and racism. In: Schiffrin

(ed.) Handbook of Discourse Analysis. Blackwell, pp.372–397.