Narrative Expansion and "Terrorist" Labeling: Discursive Conflict Escalation by State Media

How does state rhetoric change as conflict intensifies against intrastate enemies? We forward the concept of narrative expansion and labeling, to analyze the escalatory transformation of conflict discourse by the Philippine state media. The data set includes 4,098 articles from the state’s official news agency, covering early attempts at reconciliation and the eventual failure of peace negotiations between the Philippine Government and the National Democratic Front (NDF). Analysis involves a mixed methods approach, combining computational network analytics of word networks with a qualitative interpretation of emergent themes. Results reveal a discursive shift emanating from the state’s mouthpiece, alongside the political deterioration of peace talks with the NDF. The state narrative initially expands to include not only conciliatory but also confrontational talk. Eventually combative talks dominate, including a shift in labeling the enemy as terrorist rather than rebel . Narrative expansion likewise refers to state news discursively increasing the number of social actors involved in the conflict as either enemy or ally. Our findings contribute to understanding how discursive shifts may move from conciliatory to hostile discourse in a protracted intrastate conflict.

Politicians utilize discourse to create public norms and values needed to rationalize government action and mobilize public support against the perceived enemy (Altheide & Grimes, 2011;Berinsky, 2007;Mackay, 2015). During intrastate social clashes, politicians generate public patronage by identifying themselves as part of the masses and tapping popular collective emotions (Montiel & Uyheng, 2020;Obradović, Power, & Sheehy-Skeffington, 2020). In addition, leaders' talk can shape their constituents' view of reality, by focusing on national issues that emphasize social cleavages against a perceived public enemy (Obradović et al., 2020). These discursively constructed social fractures promote binary lenses in viewing a divided world inclined to dispute eruption. Through these rhetorical strategies, state leaders aim to activate a population's consensual support of aggressive measures against political antagonists. In new democracies especially in the Global South, such state enemies tend to be intrastate oppositions that challenge the legitimacy of governments in power (De Juan, 2015;Sivakumaran, 2006;Than, 2006).

Expanding the Narrative
We identify at least two forms of discursive conflict escalation, namely, expanding the narrative and destructive labeling. We posit that discursive agents swell narratives by employing discursive strategies that sequentially transform, destroy, construct, and perpetuate talks over time. These tactical discourses aim to achieve a specific social, political, psychological, or linguistic goal (Reisigl & Wodak, 2009).
To expand narratives, discursive agents first transform talks by reformulating a situation into another (Van Leeuwen & Wodak, 1999). This happens through recontextualization as a discourse from one context moves to another (Erjavec & Volčič, 2007;Wodak &Fairclough, 2010). As recontextualization takes place, two different discourses emerge about a phenomenon. To promote one of the discourses, discursive agents employ destructive strategies in their talks by reducing or silencing rhetorical productions of one of the discourses (Van Leeuwen & Wodak, 1999).
As discursive producers promote a specific discourse, they then utilize constructive strategies by establishing group categorization through utterances that constitute the "we" talk, which evokes solidarity and the "they" discourse, that promotes marginalization (Wodak, De Cillia, & Reisigl, 1999). Finally, discursive sources perpetuate self-thriving talks to acquire support and justify their actions (Van Leeuwen & Wodak, 1999). For instance, the US extended their "War on Terror" rhetoric to sustain and legitimize their actions against their enemies. When President George W. Bush addressed the Philippines Congress in 2003, he declared: The Philippines and the United States have seen the enemy on our own soil…Our two nations have made our choice. We will defend ourselves, our civilization, and the peace of the world. We will not be intimidated by the terrorists. (Bush, 2003, para. 9) Through his political rhetoric, the US President discursively perpetuated his so-called US terror war to a global confron tation that included antagonists in the Philippine-based Moro liberation movements. Such discursive acts intensified the US campaign against terrorists by expanding the number of issues, framing the conflict beyond the specific Twin Towers attack in New York to a generalized global war, and adding more players like the Philippines to a conflict that was originally a US-only problem.

Destructive Labeling
In addition to enlarging narratives, discursive conflict escalation likewise hinges on the deployment of destructive labels against one's enemy. Red-tagging, or labeling a person or group as a communist, was a tactic frequently used during the Cold War, in both superpower countries and struggling democracies. In post-Cold War conflicts, and especially after the infamous 9/11 attack in New York, the label terrorist emerged as a new discursive instrument deployed to isolate and hit an oppositionist group. This can be seen in how carefully media distinguishes groups as rebels or terrorists. A greater sense of moral value is seen in the former, and positions the latter as vulnerable to negative treatment by the public (Baele, Sterck, Slingeneyer, & Lits, 2019;Montiel & Shah, 2008).
Labeling creates binary social identities between the protagonist versus the antagonist (Coy, Maney, & Woehrle, 2008;Hodges, 2013;Robinson et al., 2005), moves toward dehumanization and demonization (Dovi, 2001;Pruitt, 2005), and positions the state as the benevolent protector of moral values against the attackers (Anker, 2005). Such discursive social acts frame reality with political narratives that justify government aggressive measures (Hodges, 2011). Hence, discursive tagging connotes public social action (Reisigl & Wodak, 2009) against or in favor of the marked group.
After creating division and dehumanizing the out-group, leaders convince their publics to support aggressive measures like armed attacks (Hodges, 2011). Such convictions are aggravated by tapping the emotion of fear from threat (Atawneh, 2009;Hodges & Nilep, 2007). Destructive labeling thus creates a new categorization that not only puts one faction at a disadvantage, but also rationalizes militarized measures against such groups. Such a discursive move escalates conflict.
Because of large population sizes, political discourses do not arise during interpersonal conversations, but through communications carried by media outlets. In the following section, we expound on the vital role of state media in the public sphere, especially in states under relatively authoritarian governments.

Vital Role of State Media in the Public Sphere of Authoritarian Governments
With the scope of its influence, media can shape how citizens see their reality (Van Klingeren, Boomgaarden, & De Vreese, 2017) and cipher political agendas to vast audiences, rapidly and efficiently (Lunt & Livingstone, 2001;Robinson et al., 2005;Woolley & Howard, 2020). As a communication vehicle of conflictual discourse exchange, media holds the power to intensify or end contention (Kempf, 2001). The power of media to construct various standpoints likewise carries the capacity to create divisive polarities in the public sphere (Elcheroth & Reicher, 2014;Van Klingeren et al., 2017). For example, media plays a key role in the process of legitimization (Lunt & Livingstone, 2001) by altering meaning making and public emotions (Altheide & Grimes, 2011;Reyes, 2011) through scripts and images.
Governments use their own channels to set state-vested agendas (Campbell, Clapp, & Wallin, 2014;Gao, 2015;Jiang, 2014;Zhang, 2010). For example, the China Central Television (CCTV) holds three goals: 1) to communicate the nation's ideologies and beliefs, 2) to be heard and to dominate the existing discourse, and 3) to reach the whole population (Zhang, 2013). As an information arm of the government, state media can likewise aggravate conflict or promote peaceful interventions during crises (Chinyere, 2020;Gilboa, 2009;Kuusik, 2010). Using its own media arm, the government can shape the publics' perceptions and expectations of disputes to be congruent with statist inclinations (Altheide & Grimes, 2011;Atawneh, 2009;Robinson et al., 2005).
By owning their news outlets, governments keep the flow of public information coherent with the state's domi nant storylines (Altheide & Grimes, 2011;Hodges, 2015a). Through their media outlets, governments rationalize state violence by presenting the state as the benevolent and defensive party. For example, governments can identify with nations and organizations that have the obligation to protect the public and pacify so-called state enemies (Anker, 2005;Dunmire, 2007;Hodges, 2011;Reyes, 2011).
In relatively authoritarian states that muffle a free press, state media emerges as the dominant source of political storylines in the open public sphere. Such is the situation in the Philippines, with the government's strong handed moves to silence independent media by filing libel cases against free-minded editors (Gonzales, 2019) and effecting the closure of an independent TV station (Mercado, 2020).
We posit that after peace talks crumble, state media's discursive shift follows a defined conflict intensification pattern that justifies state action against its perceived enemies. Subjecting press statements of state media to text analytics computations, we seek empirical answers to the question: How does state media depict conflict escalation and justify militarized action against its perceived intrastate enemies? We view the Philippine political arena in the discursive reality where media serves as a vehicle of political narratives needed to rationalize political actions (Altheide & Grimes, 2011;Cole, 2006;Dimaggio, 2009;Montiel, Salvador, See, & De Leon, 2014).
State initiatives to escalate conflict can be gleaned through the state's mouthpiece, the Philippine News Agency (PNA), the official news organ of the Philippine Government. As of this writing, the government news agency stands at the receiving end of accusations that it merely services the political whims of the Duterte government and its loyal allies (Macaraeg, 2020;Ranada, 2019).

Overview of Method
We carried out two related studies to trace discursive patterns of narrative expansion and labeling during conflict escalation. Using text-based social network computations, Study One examined how state discourse changed across time, from conciliatory to hostile narratives. Because the first study showed how the word terrorism emerged as confrontations intensified, we decided to implement a subsequent analysis. By calculating word association metrics on state media publications, Study Two checked if indeed the state media's discourse employed the terrorist label more frequently than the rebel label, as conflict peaked.

Data Source
Both Studies One and Two analyzed the same body of text. To build our data corpus, we collected 4,098 articles from the government's PNA. Our search for relevant news reports relied on keywords like communist, communism, CPP, its armed wing the New People's Army or NPA ("The New People's Army and the Armed Struggle, " 2019), and the CPP's political representative, the National Democratic Front or NDF (National Democratic Front of the Philippines, n.d.). Because we wanted to see discursive trends that occurred as conflict escalated, we time-bounded our archival search to cover articles published from 01 July 2016 to 31 December 2019. This period spanned the early negotiation attempts at the start of the Duterte administration, until the escalation of conflict when peace talks broke down.

Data Cleaning
We used the tidytext package of the data analysis software R, to organize our data. The tidytext package holds modules for basic text cleaning procedures which precede data analysis. Data processing of raw text involved organizing articles into a table containing one line per row; transforming the variant names of organizations into a single unifying tag (e.g., Communist Party of the Philippines to CPP); tokenizing or separating lines into single words; removing unmeaningful words such as articles (a, the) and linking verbs (is, was); and reducing words into their base forms through stemming (e.g., attacked to attack). These steps ensured proper identification and quantification of meaningful words needed for further analysis.

Study One, Narrative Expansion-Discursive Transformation, Destruction, Construction, and Perpetuation
Method Study One examined how discourse intensified during conflict escalation. We analyzed narrative expansion by first grouping the preprocessed articles into annual quarters or periods of 3 months each. We disaggregated the articles along a temporal dimension so we could track changes in textual data patterns across time. Hence it was possible to describe the evolving context of the discourse at different points of conflict escalation from start to end. After grouping, we identified and retained the frequently used words for each quarter. We did so in order to minimize noise from infrequent utterances and extract salient patterns from our corpora. Upon filtering, we determined words in the discourses that frequently appeared together in the same state media articles using the co-occurrence function of the widyr package. This produced a matrix that indicated the number of times each pair of words collocated or co-occurred for each selected time quarter. To further minimize noise in our semantic networks, we filtered our co-occurrence matrix to exclusively account for salient connections between words.
Using the cleaned matrix of co-occurrence, we proceeded to generate the semantic networks with a visualization software called Gephi. To detect discursive patterns at each time point of the conflict, we utilized Gephi to transform the numeric data generated in R into interpretively analyzable semantic networks. We applied the program to calculate for network characteristics such as centrality (identifying central words in the discourse) and modularity (identifying communities of words). We then visualized the central nodes (denoted by size) along with its related word communities, which were represented by specific colors. For example, in a network of words deployed by state media during peace negotiation, we found a cluster consisting of the word peace along with related terms like talk, president, and panel. To facilitate cleaning, we removed stray words that did not connect to the main word network.
Upon generating the networks, we interpretively analyzed how the discourse evolved by identifying word clusters that became more salient, and clusters that disappeared across time among the networks. For certain periods of time that span a few quarters, discursive patterns were relatively similar; the same trends in word clusters and content appeared in the network. For parsimony, we excluded these repetitive networks, choosing only periods marked by significant changes in network patterns and content.
We ended up with networks for the three sequenced time-quarters, the first was during peace talks between the government and the NDF (October to December 2016), the second was after the peace talks initially stalled (January to March 2017), and the third was during the height of the intrastate conflict (April to June 2019). These semantic networks are presented in the following results section.

Results
Our findings demonstrate how state discourse follows a shifting pattern during conflict escalation. For purposes of a longitudinal lens, we present three semantic networks at three points in time, depicting expansion of political discourse deployed by the Philippine New Agency in their articles reports. Figure 1 pictures state media's semantic network during peace talks between the Philippine Government and the NDF. Figure 2 visualizes the state's discursive shifts, as peace negotiations stumble, and confrontational talks begin to rise in state media. The third semantic network illustrates narrative expansion at its peak, as social conflict intensifies between the government and groups represented by the NDF. Figure 3 shows an expanded narrative with more discursive elements in the larger networks of words. Further, this third word network depicts the appearance of the label terrorist in the emergent, more aggressive talk.
We maintain the original words as they appear in the media reports, to reflect the naturalness of local talk in our paper. Understandably, the Figures contain acronyms and names that only Filipinos understand. We list these local terms and their meanings in the Appendix. Figure 1 shows that during peace talks with the NDF, Philippine state media deployed conciliatory utterances. For easier visual interpretation, we encircle the dominant semantic network and present the community of related words in green colored text. Notice the presence of peace-related phrases and words such as negotiation, peace, bilateral ceasefire, and release of political prisoners. These appear in conjunction with key political leaders of the state, such as President Duterte, and government peace representatives Dureza and Bello.

Semantic Network During Peace Talks: Conciliatory Talk
In the context of state media claims, these words pertain to the Philippine Government's willingness to release political prisoners if the NDF signed the bilateral ceasefire agreement. For illustrative purposes, we quote state media articles and underline words included in the semantic network. One news article announced that: President Rodrigo Duterte has said he will only order the release of 130 political prisoners if the government peace panel and the National Democratic Front (NDF) sign a bilateral ceasefire deal during talks in Oslo, Norway this week.
State media also reported the willingness of the NDF to comply with ceasefire: Government peace negotiator Labor Secretary Silvestre Bello III, in a statement, said Luis Jalandoni's statement on the readiness of the NDF to sign a ceasefire agreement even before the release of the political prisoners, is indeed a welcome development as this bodes well to a positive atmosphere when we meet again for the third round of talks in January.

Semantic Network as Peace Talks Break Down: Transforming Conciliatory Into Confrontational Talk
As relations sour between the government and the NDF, a second set of aggressive words emerge, creating two types of talk in a single text network. Figure 2 shows the original peace cluster joined by a second network of more confrontative words. The semantically distinct peace and war talks are found in the same public sphere, as peace negotiations stumble. This suggests that state media employed a transformative strategy in their discourse to shift talks about peace into conflict.

Figure 2
State Media Reports' Word Network Analysis as Peace Talks Stumble (January-March 2017) In Figure 2, we visually identify the emergent and more antagonistic talk as the encircled red network. While still reporting about peace processes, the PNA likewise highlights militarized encounters between state troops and anti-gov ernment forces. Note that the new red cluster of aggressive words includes terms such as battalion, operation, military, attack, wound and kill. Central in this word network is the phrase NPA or New People's Army, the military arm of the Communist Party, while the militarized words pertain to the NPA's harassment activity and killing of military forces. For example, as conflict escalates, state media writes that: A trooper from the 30th Infantry Battalion was killed while another three were wounded in what the military describes as a New People's Army (NPA) "harassment attack" in Barangay Mati, Surigao City Monday early morning. State media also claims military victories over the NPA by reporting that: Two NPA members were killed in the operation launched by elements of the 61st Infantry Battalion in Maayon, Capiz. The troops also recovered two shotguns and two hand grenades from the NPAs.
Interestingly, Figure 2 introduces the terrorist tag. During peace talks (Figure 1), only the rebel label appears, but rebel conflates with terrorist in Figure 2. We further investigate the linguistic evolution of rebel to terrorist in Study Two. Figure 3 depicts the state discourse at the peak of intrastate clashes with groups represented by the NDF, more specifically the CPP. At this conflict juncture, the state stops talking about peace as state media deploys destructive rhetoric. The peace-related semantic network salient in Figure 1 disappears in Figure 3. However, the combative words that emerged when peace talks stalled (Figure 2), remain in the picture as clashes intensify.

Semantic Network at the Height of Conflict Escalation: No Peace-Related Words, More Numerous Conflict Actors, Reified Terrorist Tag
We point out that the discursive shift in Figure 2, as peace talks stumble, takes on a loud militaristic tone against the state enemy, targeting the military arm of the NPA. But as conflict escalates as shown in Figure 3, the discursive attacks turn more subtle, more political, targeting alleged civilian political fronts of the CPP rather than its armed troops. We quote from state media and underline words that appear in the semantic network of Figure 3: Parlade said other NDF allied organizations include, MAKABAYAN, Karapatan, NUPL.

State Media Reports' Word Network Analysis as Conflict Escalation Peaks (April-June 2019)
Notably, a new type of discourse reveals itself in Figure 3, a kind of talk that enlarges the number of social actors in the combative narrative. The yellow cluster marked in Figure 3 suggests that state media constructed the "they" discourse by implicating civic groups as part of CPP. The conflict narrative expands to the civilian population, as state media throws a wider accusatory net and associates other political organizations with the CPP. Figure 3 also shows that the discursive expansion from few to many social actors does not only include linking civilian groups with the Communist Party. Another rhetorical expansion takes place as the PNA brings international countries into the discourse. State media includes foreign countries as backers of the claim that the CPP is a terrorist organization, thus legitimizing the terrorist label hurled at the CPP as produced in news articles. For example, one news report claims: The NPA, which has been waging a five-decade armed struggle against the government, is considered as a terrorist organization by the United States, European Union, the United Kingdom, Australia, Canada, New Zealand, and the Philippines.
The underlined words in the above quote, and the new semantic network encircled in Figure 3 show the emergence of Western democratic entities in the discourse. This demonstrates how state media discursively acquired support from other countries to legitimize the government's militarized response against the CPP.
We also found an emerging discursive strategy, which transformed talks about rebellion into discourse about terrorism. A deeper analysis of this lexical phenomenon reveals a legitimization strategy. Study Two further examines the emergence of the terrorist label as conflict intensifies.

Study Two, Discursively Legitimizing Militarized Confrontations-From Rebel to Terrorist
Study One showed that the word terrorist appears as conflict escalates. Note that the identity of the CPP on state media changed across time, from rebel to terrorist. We posit that this discursive shift transformed the rebel discourse, and recontextualized the terrorist label by borrowing the term from the more global discourse of the war on terror.
The underlying script legitimizing civilian-supported state violence against terrorists is subtle yet powerful. From the political viewpoint of the state's mouthpiece, what would the discursive shift from rebel to terrorist imply? First, rebellion targets military forces of a state, whereas terrorism intentionally carries out acts of violence toward civilians (Ganor, 2002). Hence, the latent argument here is that defensive state violence against terrorists should be supported by the public at large. Second, to address rebellion, states enter peace negotiations (Baquiano, 2019), but governments respond with militarized actions toward terrorists (Aguirre, 2009). We suspect that this aggressive terrorist label emerged as a discursive attempt to rationalize military confrontations and permit harsher treatments of anti-state forces.
In Study Two, we investigated in more detail whether indeed state media indeed shifted the identity of the CPP from rebel to terrorist. Operationally, we asked the question: as conflict intensifies, does the term CPP appear more frequently with the label terrorist than with the word rebel in state media?

Method Computing Identity Labeling: How Often Does CPP Appear With Rebel or Terrorist?
Our data source consisted of the same 4,098 articles collected in the first study. Similar to Study One, we sorted state news reports according to time quarters, covering 3 months per set of articles. Unlike Study One, however, Study Two analyzed all the quarter sets. Since the time span covered 42 months or 14 quarters-of-a-year, this second study consisted of 14 sets of news reports, juxtaposed to each other along a temporal dimension.
We proceeded to compute for confidence, a technical term intuitively similar to a correlation score, but applicable to calculating how often one word appears together with another word in the same article. The mathematical goal was to determine the percentage of text co-location between the identity term CPP, and the two political labels rebel and terrorist. A higher confidence score indicated that one political label appeared more frequently than the other label, in relation to the identity word CPP. Our text computations produced a matrix of rebel-CPP and terrorist-CPP confidence scores for each of the 14 time points, making it possible to detect any changes across time based on the changing amount of association between political labels (rebel, terrorist) and political identity (CPP).

Unpacking the Contextual Meaning of Rebel and Terrorist
After determining initial quantitative patterns, we further unpacked the meanings of rebel and terrorist, as contextually deployed by state media. To explore the meanings of these labels, we carried out correlational analysis to identify the words most likely to occur with the labels. We assumed that words in the discourse that appeared most frequently with rebel and terrorist would lend themselves to revealing the contextualized meaning of these two political labels. To do this, we computed for phi coefficient, a measure equivalent to Pearson correlation when applied to binary data. We selected the top 15 words most correlated with rebel, and another set of top words most associated with terrorist.
At this point, we shifted to qualitative analysis. Using these two sets of words as our guideposts, we returned to the original articles to detect rebel and terrorist storylines based on their respective sets of correlated words. Through this analytical procedure, we derived state storylines showing contextualized meanings of the labels rebel and terrorist during conflict escalation. Note that in Study Two, the storylines presented are not direct quotes from news reports, but rather narratives constructed by the researchers, based on a reading of all state media reports containing the words rebel or terrorist.

Discursive Conflict Escalation and Political Labeling: From Rebel to Terrorist
We operationalize political labeling as the amount of computed association between a discursive identity (CPP) and a political tag (rebel or terrorist). Figure 4 plots our confidence results, showing how the association of rebel-CPP and terrorist-CPP change as conflict intensifies across time. The left side of Figure 4 shows the state's discursive deployment at the start of the peace talks. CPP's identity links most frequently with rebel. At this juncture, the label terrorist stands at an almost zero point. Note that across time, the slope of the CPP-rebel line is relatively flat, indicating that regardless of conflict intensity, the identity of CPP as a political group remains constant. The striking finding comes with the markedly positive slope of the CPP-terrorist associative track. As peace talks stumble and intrastate conflict escalates, the terrorist word occurs more frequently with CPP. In other words, terrorist labeling intensifies. At the peak of the conflict, shown on the right part of Figure 4, terrorist stands as the more salient label of the CPP.

A Contextualized Meaning of Rebel and Terrorist
In this section, we present narratives that reveal how state media frames the meanings of rebel and terrorist, during conflict escalation. We construct these storylines by cohering the 15 words with the highest phi coefficient into one meaningful sentence, based on how original news reports used these selected words vis-a-vis the political labels.
The term rebel connotes a group that clashes with government troops, while the state extends a peaceful resolve. To elaborate on this theme, we present the longer rebel storyline, with the 15 associated words underlined in the expanded narrative: Residents flee as communist NPA rebels clash in encounter with troops of the Army's infantry battalion. Rebel returnees surrender, granted assistance under the Enhanced Comprehensive Livelihood Integration Program, (E-CLIP).
Interestingly, the word terrorist does not associate with actions or descriptives that expound on its military or political features. The storyline extracted from the top 15 words linked with terrorist reads: The communist organizations CPP and its armed wing, the NPA, are listed and tagged as terrorist organizations by the USA, the EU, the UK, Australia, Canada, NZ, and the Philippines.
The state discursively produces legitimization, by contextualizing terrorist in a large array of two antagonistic political identities. On one side stands the state's leading enemies like the CPP and the NPA. The state then claims that on the other side of the fence, global democratic entities like the EU and the USA join the Philippines in terrorist accusations hurled against intra-state enemies. The domestic intrastate conflict discursively turns into a global war on terror.
We see another latent message in the subtext of this storyline. The state may be discursively preparing the stage for more militarized hits in the name of a just war against terrorists, blame-free of human rights violations, because the democratic nations are on the side of the Philippine state. Through this linguistic move from rebel to terrorist, the discourse shifts from political compromises to legitimized militarized confrontations, increases from few to many social actors, and expands from an intrastate conflict to a global war on terror.

Discussion
We are mindful that the discursive nature of our research disallows any causal conclusions, nor any inferences about motivations fueling utterances. In this paper, we do not claim that state discourse caused escalation. We likewise veer away from alluding to political motivations. Our study focused on a state's discursive conflict escalation. We merely point out what state discourses mark the intensification of social conflict.
An important contribution of our research lies in its decision to investigate the political discourse of state news media in a relatively authoritarian state. In stable democracies, both state owned and private media are at least formally free of state influence, although media funding and advertising constitute avenues to pressurize media toward state ideologies and interests. But a different picture emerges in relatively authoritarian countries, where regimes muffle free media and control government news agencies. Under such undemocratic political conditions, state media exerts a marked influence in shaping the public mind toward state-vested interests.
Whereas past studies focused on how conflict intensifies at the level of international disputes (Anker, 2005;Atawneh, 2009;Berinsky, 2007;Hodges, 2013;Dunmire, 2007), our study presents how conflict escalates in an intrastate clash. In particularly strong states, domestic conflicts unfurl asymmetrically, partitioning social actors into the dominant government and low power groups. Our study demonstrates how a powerful regime can expand political discourse to legitimize a militarized response toward the low power group. For example, our research showed how state media recontextualizes discourse from the war on terror, as state reports shift from rebel to terrorist, and introduce global democracies as allies of the Philippine state.

Discursive Performance of State Media During Conflict Escalation
We present new ways of conceptualizing and empirically investigating conflict escalation. The intensification of a social clash is understood as a narrative expansion, as word clusters about peace appear in the public sphere, then conflate with terms from another more belligerent context. Eventually, the aggressive words take central stage, and the peaceful terms disappear from the discourse. At the height of narrative expansion, the destructive discourse spreads, increasing the number of word clusters, bringing new social actors into the combative talk, and fortifying the associative link between militarized words such as terrorist and political identities such as the CPP.
Our lens swerves away from past discursive approaches to social disputes that analyze discursive production of conflict actors by examining their words as sources of information rather than focusing on what the actors accomplish through their discourse (Elcheroth, Penic, Usoof, & Reicher, 2019). Our approach treats state media's words as social acts or discursive performances of the government. We ask what the words do when they are deployed by state media. Words can set agendas of peace and violence, expand the narrative to a larger story that blames more anti-state groups and aligns global democracies with the state, and shift the rhetorical label of the enemy from a political rebel fighting government troops to a militarized terrorist bent on harming civilians.
The latter performance involving a shift in labeling bears much weight in an intrastate war. Labeling groups as terrorist evokes contextual features suggesting that such groups carry out atrocities towards civilians (Ganor, 2002). By calling opponents terrorists, the state generates public support from a citizenry threatened by random terrorist violence. Such labeling legitimizes state violence against a group discursively described as harmful to the entire population.

Methodological Contributions
One methodological contribution of our research lies in its use of a discursive-analytics approach to conflict escalation, studying widespread natural talk as it arises in real conflict situations. We employ a mix of text analytics and qualitative storylines, to examine our state press releases and articles. Past studies on conflict and its escalation have depended on work in case studies (Acar & Uluğ, 2016), experimental laboratories (Lee, Gelfand, & Kashima, 2014), surveys (Tagar, Federico, & Halperin, 2011), or large-scale data bases (Gleditsch, Wallensteen, Eriksson, Sollenberg, & Strand, 2002). Borrowing from the field of societal computing and text analytics, our methodological approach optimizes the strengths of other conventional strategies in one package, offering the naturalness and depth of case studies, quantitative precision of laboratory measures, and the breadth of surveys and data bases.
We appropriate knowledge from the field of computer science and apply text analytics to match the political psychological questions we pose. Semantic network analysis provides a new and more precise way to extract discursive themes, by first quantifying associations of word communities that appear together in the discourse then providing an intuitive way to examine its overall structure through visualization.
The methodological strength of semantic network analysis lies in its ability to retain natural domestic talk. This is particularly useful in studying phenomenon based on non-English languages, especially in the Global South. The local flavor of talk survives data processing. However, because it mainly provides us with broad patterns, semantic networks can be hard to interpret without contextual explanations on how words are used in discursive practice. Hence, this method can be extended by serving as an empirically counted basis for naming themes as our study demonstrated. Quantitative text analysis complements qualitative content analysis by enabling researchers to withhold subjective extractions of themes until after word computations. In this manner, results can therefore stand relatively protected from researchers' bias and interpretive error.
Aside from semantic network analysis, this study likewise applies the metric called confidence as a mathematical tool to detect associative patterns in text data. Confidence summarizes the intensity of correlation between two words. It can be used to answer any conceptual question based on relationships between words or utterances in a discourse. For example, Figure 4 illustrates how confidence measures assess the association between the CPP's political identity, and political labels rebel and terrorist.

Study Limitations and Suggestions for Future Research
The generalizability constraints that bound our findings can likewise signal directions for future research on discursive conflict escalation. In spite of its deep nuancing of discursive patterns, this study bears limitations on three features that contribute to the study's lack of breadth.
First, we are mindful that our results emanate from a single-N case, specifically state media text arising from one conflict in a particular country. Subsequent researches may want to investigate conflict escalation discourses in various political spheres, to find out if escalatory discursive patterns change according to the type of conflict or geopolitical histories. Other studies suggest that indeed escalatory talk may emerge during conflict intensification. For example, we trace similar strategies in the rhetoric of the Israeli Government as they focus their discourses more on the atrocities committed by an opposing insurgent group to evade their responsibilities over civilian deaths (Finlay, 2018). We likewise find comparable patterns in the discourses of Serbians as they construct themselves to be part of the Western and European Christians, and classify non-Christians as terrorist to justify violent actions against the conflicting group (Erjavec & Volcic, 2007).
A second limitation revolves around the one-sidedness of discursive voice, in a two-sided conflict. Our study unpacked escalatory talk by state media, but did not analyze what the communists or their affiliated fronts were talking about during the intensification of conflict. Past studies show that minority groups intensify dispute by expressing grievances (Unruh, 2015). However, we were unable to validate these discursive trends, because most of the substantial political talk in underground movements do not appear readily on the internet. Further, our sense is that when peace talks fail, at least one or both groups stop engaging in the conversational sphere, and shift to other nondiscursive combative actions.
A third generalizability constraint relates to the domain coverage of our research. We investigated discursive conflict escalation. In our global world, however, conflict de-escalation likewise arises, as internal peace building efforts succeed. Future studies may want to explore and compare talk patterns, as belligerent parties in an intrastate war cool down and negotiate toward a more lasting peaceful relationship.

Practical Applications
Because our results likewise uncover similar discursive configurations associated with conflict escalation, we forward that such talk patterns can also be used as linguistic markers that signal when an intrastate clash begins to intensify. At such inflection points, practical countermoves toward cooler talk may be enacted either linguistically or behaviorally, in the political arena.
For example, a shift to aggressive talk by the state can be noted but not inflated, as other social actors steadfastly feed the public discourse with more peaceful and conciliatory utterances. The discursive enlargement of conflict actors, enacted to either widen the blame game against oppositionists or increase the size of state allies, can be discursively noted in public, and called out for its escalatory nature. Moreover, one may remain mindful of labeling strategies when the terrorist tag emerges, as a signal to legitimize state violence and recontextualize a domestic clash into a global war on terror. Other local public actors may consciously avoid the term terrorist and instead deploy terms like rebels or other political nomenclature like freedom-fighters or liberation movements. Such discursive practices may contribute to keeping peace talks on a constructive trajectory, as state and anti-state forces talk things out at the negotiating table.
Funding: The authors have no funding to report.