Daniel Karell y Michael Freedman
«Rhetorics of Radicalism»
American Sociological Review, vol. 84, n.º 4 (2019)
American Sociological Review | American Sociological Association (ASA) (@ASAnews) | Washington | Estados Unidos
Se incluye a continuación un extracto seleccionado de las páginas 2, 4 a 8 y 23 a 27 de la publicación en PDF. Las referencias pueden consultarse en la ubicación original.
Disponible a través de SocArXiv (@socarxiv) | Center For Open Science (@OSFramework) | Universidad de Cornell (@Cornell).
«ABSTRACT
»What rhetorics run throughout radical discourse, and why do some gain prominence over others? The
scholarship on radicalism largely portrays radical discourse as opposition to powerful ideas and enemies, but
radicals often evince great interest in personal and local concerns. To shed light on how radicals use and
adopt rhetoric, we analyze an original corpus of more than 23,000 pages produced by Afghan radical groups
between 1979 and 2001 using a novel computational abductive approach.
»We first identify how radicalism
not only attacks dominant ideas, actors, and institutions using a rhetoric of subversion, but also how it can
use a rhetoric of reversion to urge intimate transformations in morals and behavior. Next, we find evidence
that radicals’ networks of support affect the rhetorical mixture they espouse, due to social ties drawing
radicals into encounters with backers’ social domains. Our study advances a relational understanding of
radical discourse, while also showing how a combination of computational and abductive methods can help
theorize and analyze discourses of contention.
»PATTERNS OF RADICAL RHETORIC
»Social scientists seeking to discover unknown concepts and themes in large textual corpora increasingly
use unsupervised machine learning techniques —most commonly, topic models— to identify patterns of word
usage (e.g., Quinn, Monroe, Colaresi, Crespin, and Radev 2010; Grimmer and Stewart 2013; Miller 2013;
Mohr and Bogdanov 2013; Roberts, Stewart, Tingley, Lucas, Leder-Luis, Kushner Gadarian, Albertson, and
Rand 2014; Light and Odden 2017). This inductive analysis can be combined with expert knowledge, a
deep engagement with source material, and other automated content analysis techniques to yield novel and
compelling inferences, an approach recently described as computational grounded theorizing (Nelson 2017).
»However, an inductive approach is less applicable when researchers come to the data already having some
idea of the direction in which hypotheses, arguments, and theories may develop.
»When ethnographers find
themselves in such circumstances, they often undertake abductive analysis: identifying surprising findings
in light of prevailing arguments, and reasoning through what makes them unexpected to generate a newly
refined theory with greater explanatory power (Tavory and Timmermans 2014). Similar to ethnographers
familiar with arguments relevant to their field site, we have a wealth of scholarship on radicalism to draw
from, as well as literature on and experience with our empirical case, Afghanistan. As a result, we develop an abductive framework for the analysis of a large corpus of radicals’ writing to conceptualize the rhetorical
underpinnings of radical discourse.
»Afghanistan, crucible of radicalism
»Our study relies primarily on two decades of text produced by radical groups in and around Afghanistan.
This period of time begins with an uprising against the central government in 1979, stretches through a long
and evolving conict, and ends on September 11, 2001. During this period, radical groups used printed text
as a key medium for broadcasting their rhetoric to diverse audiences at home and abroad —a necessary task
due to the country's entanglement with regional and international power-struggles.
»For example, magazines,
newspapers, speeches, sermons, and pamphlets could be printed in multiple languages to entertain local and
foreign fighters in the battlefield (Rana 2008), sent abroad to attract financial support and new recruits
(Rubin 2002), and distributed in refugee camps and on university campuses —where the early radical groups
formed (Roy 1986)— to boost support and draw recruits, a tactic still used today.
»Studying the period between 1979 and 2001 affords two analytic advantages for examining radical rhetoric
and its adoption.
»First, there is variation in rhetoric, both across and within radical groups. The numerous
groups operating in and around Afghanistan during this time espoused a range of ideas and arguments that
formed a general discourse of radicalism, and some of these groups changed their rhetoric over time (Christia
2012).
»Second, these groups differed in ways that could inuence their generation and adoption of rhetoric.
For example, some mujahideen, the rebels who fought Afghan communist and Soviet forces during the Jihad
(1979 to 1989), received more financial and military support from the United States and its allies than other
mujahideen. Similarly, some were more active on the battlefield, and suffered more casualties, than others
(Rashid 2000; Rubin 2002). We collect data on these relevant group characteristics for use during the
second part of our study.
»In addition, this period in Afghanistan's history helps us understand contemporary radicalism. Not all
radical discourse and behavior is tied to Islam, but there has been an increase in activity by militant Muslim
groups over the past few decades (Walter 2017b). Many of these groups’ lineages can be traced back to
Afghanistan, 1979 to 2001. For example, the Islamic State arose from al-Qaeda of Iraq, an affiliate of the
original al-Qaeda, which was organized during the Jihad and has roots in Abdullah Azzams Maktab al-Khidamat (the Services Bureau), one of the groups included in our study. Thus, our interest in Afghanistan
extends beyond its analytic usefulness —we gain a window into the early formulations of one of today's most
prominent radical traditions.
»Textual data
»We built our corpus by collecting magazines, newspapers, speeches, and pamphlets in Afghanistan, Iran,
and Pakistan. These materials were originally produced by 22 different individuals or groups operating in or
around Afghanistan between 1979 and 2001. The authors include both famous and lesser-known mujahideen
organizations —for example, Jam'iyyat-i Islami and Hizb-i Islami (Gulbuddin faction), in contrast to Jami'ah
al Da'wah ila al-Quir'an wa al-Sunnah and Hizbullah Afghanistan— as well as the eventual enemy of some
mujahideen groups, the Taliban.
»We found relevant material in five languages: Arabic, Dari, English, Pashto,
and Urdu. Appendix A provides details of our data collection and text preparation.
»The corpus has three characteristics allowing us to uncover meaningful rhetorical patterns within radical
discourse, and, later, the reasons for their adoption. First, our corpus captures as wide a swath of the
period's radical discourse as possible. Of course, as with nearly all datasets of historical material, it does
not contain the entire population. However, we approximate that population by including texts from a
wide range of groups, origins, and languages. For example, instead of only including publications from large
mujahideen groups, we sought out texts produced by more obscure groups.
»Similarly, rather than focusing
solely on the famous Sunni-majority “Peshawar Seven” groups, we include publications by Shi'ite-dominated
groups. In addition, we collected publications by the same group in different languages whenever available.
Furthermore, our corpus includes full runs or near-full runs of many publications, including those produced
by influential groups like Jam'iyyat-i Islami, the Taliban, and precursors to al-Qaeda. In summary, we are
confident that our extensive data collection has minimized, and resulted in non-systematic, omissions. For
more information on the coverage of our corpus, see Appendix B.
»A second advantage is that the data do not suffer from the response errors of social desirability bias and
inaccurate recall common in survey and interview data. The documents capture the rhetoric that radicals
chose to use. Finally, the corpus offers observations over many years. We exploit this longevity in the
second part of our study, when we use variation in global oil prices over time to identify the direction of the
relationship between networks of support and rhetoric.
»Computational abductive analysis
»We developed our conceptualization of radical rhetoric via a computational abductive analysis of textual data,
presented in the following section. This kind of analysis parallels work using ethnographic and comparative-
historical data (Tavory and Timmermans 2014; Anderson 2018), except we rely on computational methodologies to “defamiliarize” the object, or concept, under study.
»We began as other abductive analyses do: we studied existing social scientific theories and arguments
regarding radicalism, as well as the details of the case. The latter involved multiple rounds of informal
interviews in different parts of Afghanistan, reading the secondary literature, and a close reading of two
independent journalistic publications that were collected alongside the radical texts but not included in the
corpus. Both of these sources, Afghanistan Forum (1983 to 1985) and Afghanistan Information Center
Bulletin (1988 to 1992), were sympathetic to the anti-Soviet cause but were not allied with any particular
militant group. During this initial stage, we surmised that radical discourse is commonly understood as
directed outward: radicals attack powerful individuals, organizations, ideas, and ways of life in their society,
but outside of their immediate community and network of supporters.
»Next, we identified latent categories of meaning, or “topics”, running throughout the corpus using a
structural topic model (Mohr and Bogdanov 2013). Topics are defined using frequency distributions of
terms probabilistically estimated to cluster together; they are themselves distributed across documents in
the corpus (Blei, Ng, and Jordan 2003; Grimmer and Stewart 2013).
»Put in the language of the sociological
study of discourse, topic models first show us collections of words based on how they appear together. Then,
for each topic, the words can be interpreted in relation to one another to gain insight into the meaning
represented by the topic (Somers 1995; Mohr 1998; Somers 2008).
»Identifying topics in the corpus facilitated a key methodological step: defamiliarizing radicalism. That
is, we used the computational measurement of the corpus to identify puzzling aspects of radical expression.
In our case, we found that although several topics corresponded with the scholarly literature's prevailing
understanding of radical rhetoric —what we call “rhetoric of subversion”— some did not fit. This compelled
us to ask questions about radicalism that we otherwise would not have (Tavory and Timmermans 2014:
57).
»The pursuit of these questions led to a recursive movement between the computational results, a
close reading of selected corpus material, further literature on social movements, and additional theories
that engendered a novel conceptualization of radical rhetorics. The computational abductive approach also
involved systematically iterating through a range of topic model specifications (see Wilkerson and Casas
2017).
»The first step in our computational examination was to prepare the text data in line with recently
developed strategies for comparative computational textual analysis. Non-English texts were translated
into English (Lucas, Nielsen, Roberts, Stewart, and Tingley 2015), all the text material was preprocessed,
and publications were parsed into documents, or pages of the publications, that act as our observational
units (Grimmer and Stewart 2013). This preparation yielded 23,343 documents (see Table 1).
»Next, we
estimated a structural topic model (STM). An STM works by incorporating metadata, or information about
documents, as covariates (Roberts, Stewart, and Airoldi 2016). We included the covariates of authorship
(i.e., the individual or group that published the text), year of publication, and the original language of the
text. This way, we isolated patterns in the semantic content of the corpus rather than differences due
to idiosyncrasies related to any values of the covariates (Lucas et al. 2015). After an iterative process
evaluating model quality based on qualitative validation (Grimmer and Stewart 2013), we selected a 10-
topic solution.
»This solution was further validated using a data-driven assessment. A detailed discussion of
our computational procedures, including the selection of the number of topics, validation assessments, and
robustness checks can be found in Appendix C.
»We then drew on the model's output of topics, a close reading of documents highly associated with
specific topics, and our knowledge of the case to understand each topic. Following convention, we labeled
the topics based on our understanding; this helps make them more accessible to readers unfamiliar with
the case (Mohr and Bogdanov 2013). When examining topics' words, we paid special attention to their
FREX scores, a measure of terms' probability of appearing under a topic and their exclusivity to that topic
(Roberts et al. 2014). Next, we interpreted them individually and in groups in light of the prevailing
ideas of radical discourse.
»When we encountered topics that challenged these ideas, we turned to a broader
literature —specifically, research on religious social movements— to construct a new conceptual framework of
radical rhetoric. Our resulting conceptualization identifies and characterizes two rhetorics of radicalism.
Of course, textual material —whether entire publications or individual pages— ontain a mixture of rhetorical
types. Therefore, in the final step of the analysis, we calculated the prevalence of each topic per document
and used the topics' associations with one of the two rhetorical types to determine a rhetoric ratio score for
each document. This score indicates the relative prominence of rhetorical types at the document level.
»The
more a document is made up of (multiple) topics forming one rhetorical type, the more its score approaches
one or zero. The rhetoric ratio score can be used to index the rhetorical variation throughout the corpus,
as well as further check our interpretation. Namely, previously un-read documents selected solely by an
extreme rhetoric ratio score should contain text closely aligning with one rhetorical type or the other.
Finally, we used our conceptualization of radical rhetoric to specify two general dimensions structuring
radical discourse, connectivity and reach, or how radicals are urged to achieve their goals and on what
timescale (Mische 2009).
»Where radical rhetoric falls on these dimensions gives it its character. In the
Discussion section, we explore how considering rhetorical connectivity and reach helps shed light on other
forms of contentious discourse.
[...]
»DISCUSSION
»To advance our understanding of radical discourse, we identified and analyzed radicals' rhetoric across two
decades in Afghanistan. We found that radical discourse emerges from an interplay between rhetorics, such
as arguments for changing society through attacks against perceived outsiders (i.e., a radical rhetoric of subversion)
or through an inwardly focused remaking of individuals and communities (i.e., a radical rhetoric of
reversion). In addition, the balance between rhetorics reflects radicals' social relationships. Radicals' avenues
of support will generate encounters between their own social domain and the domain of their backers (White
2008).
»The resulting inter-domain space —what Mische and White (1998) term a “public"— engenders an
interaction that constrains and guides the construction of ideas and arguments, or rhetoric. Radical discourse
thus develops through relational processes comprising interactions between radicals and supporters
over time, mirroring the relational mechanisms driving radical behavior and tactics (della Porta 2013; Alimi
et al. 2015; della Porta 2018).
»Contributions and future research
»Our conceptual work represents an initial step toward theorizing radical discourse, but we are encouraged by
the empirical evidence of rhetorical varieties and the dynamics behind their adoption. Future research could
draw on our insights to develop discursive explanations for radicalization and radical action. That is, which
varieties of rhetoric, or mixtures of rhetorics, appeal to which kinds of individuals? And, how do different
types of rhetoric affect these individuals' behavior?
»The research on radicalization increasingly suggests that the search for belonging motivates many individuals
to join radical groups (Roy 2004; Bosi and della Porta 2012; della Porta 2018; Mitts 2019).
Often, a sense of belonging is achieved through social relationships (Sageman 2004; Wiktorowicz 2005) and
ideological resonance (Hegghammer 2010; Borum 2011; Gutiérrez Sanín and Wood 2014; Walter 2017b;
Mitts 2019). Yet, it is rhetoric that often helps add meaning to relationships and shapes an individuals
understanding of a radical group's ideology which itself frequently morphs in response to strategic concerns
(Goldstone et al. 1991; Christia 2012; Walter 2017a).
»Thus, a focus on rhetoric could shed light on how radicals use language to construct particular experiences
of belonging, helping to explain why certain communities, and not others, provide the home that radicalizing
individuals are searching for. Moreover, it opens the door for exchanges across disciplinary subfields. For
example, recent findings in cultural sociology (e.g., Lizardo 2006; Lewis, Gonzalez, and Kaufman 2012;
Lewis and Kaufman 2018) can inform research on how rhetorics of subversion and reversion shape the
formation of social ties. Such work would help lay the foundation for a new research agenda on the role of
culture in violent conflict and contentious politics.
»Future research can also examine the unintended consequences of supporting radicals. As radicals craft
their rhetoric in the public shared with patrons, we expect they will typically converge with their backers'
values and perspectives (Snow et al. 1986; Snow and Benford 1988). However, this might not always
be the case. Radicals, like all individuals, enter publics with their own social domains in tow, and these
domains might inadvertently spur tensions between radicals and their supporters. Some radicals may in fact
repudiate supporters, whom they might see as beneficiaries of the social systems they seek to change (Ferree
2003).
»Our empirical analysis suggests how this potential paradox can unfold. The groups we studied that
received external support were also those that advocated for a history-spanning, civilizational fight potentially
threatening to their patrons (i.e., a radicalism of subversion). Indeed, some of the mujahideen groups that the
United States supported in the 1980s —groups that expectedly adopted a rhetoric championing an upending of
the world order— challenged their benefactors during the first Persian Gulf War (Desert Storm) by expressing
support for Saddam Hussein.
»We find further suggestive evidence of this paradox when we analyze how
groups' adoption of rhetoric affects future external support: receiving external support leads to adoption of
the subversion rhetoric, but an increase in subversion rhetoric is not significantly associated (p > .05) with
receiving external support during the following year (Appendix G).
»Examining how networks of support can shape radical discourse in unexpected ways also has the potential
to help explain the puzzling proliferation of transnational violent jihadism in recent decades (Walter 2017b).
Namely, as petrostates in the Arab world grew wealthier, their citizens and governments increased support for
groups espousing doctrines of personal and familial morality (i.e., rhetoric of reversion) they found agreeable.
But, as our findings suggest, this support may have inadvertently helped transform some of these groups
into transnational Islamic militants.
»Finally, future research could build on our conceptualization of radical rhetoric to develop a new theory
of radicalism, as well as a broader framework for understanding the similarities and differences between
various discourses of contention. To reprise, the rhetorics of subversion and reversion contrast by how they
fall along dimensions of connectivity and reach (Mische 2009). That is, they are characterized by how they
urge radicals to achieve their goals (connectivity), as well as by the timescale offered for achieving the desired
societal transformation (reach).
»The dimensions draw our attention toward a theoretical understanding of radical discourse. Rhetoric
that answers questions of connectivity and reach constructs a projection of the future (Mische 2009; 2014).
In the case of radical discourse, the projected future excludes specific elements of present-day society: subversion
rhetoric erases disdained powerful individuals, organizations, or institutions, and reversion rhetoric
erases individuals and groups that do not undertake the advocated inward transformation.
»So, while many
discourses offer visions of the future, such as the utopian thinkers of industrializing Europe (Calhoun 2012),
we posit that radical discourse is typified by its clear articulation of a boundary of exclusion from the future.
Radical discourse makes explicit that some segment of society —or its ideas or behavior— is unwelcome in an
imagined future.
»Moving beyond radical discourse, considering its rhetorical dimensions also helps us distinguish it from
similar discourses, such as revolutionism. Whereas revolutionism primarily focuses on mass contention and
transformation of sociopolitical institutions (Beck 2015), radicalism can interweave an intimate personalized
and localized struggle into fundamental social change. Furthermore, the notion of rhetorical boundaries
suggests a boundary perspective for interpreting discourses of contention as a whole. After all, a large body
of research on these discourses note their exclusionary nature (Rodgers 1987; McAdam, Tarrow, and Tilly
2001; Tarrow 2013). For example, nationalist discourse describes the political boundaries people believe
they deserve by identifying who is permitted to participate within the polity (Hechter 2000; Bonikowski
and DiMaggio 2016).
»Populism defines the bounds of the “common people" by excluding a corrupt elite,
detestable lower classes, and other outsiders (Mudde 2004; Jansen 2011; Bonikowski and Gidron 2016;
Müller 2016; Brubaker 2017). Fascist speech excludes the unvirtuous from the imagined nation (Stanley
2018), which, intriguingly, often uses a rhetoric of the past to depict the nation as having been always
virtuous.
»Developing a theory of contentious discourse based on the metaphor of boundaries could help illuminate
how these discourses are distinct, as well as what they share in common. In addition, it can help us analyze
how discourses merge. For example, it might become apparent that “radical nationalist" discourse entails
the invocation of both a political boundary and a boundary of futurity. Radical nationalist discourse would
thus be understood as a discourse excluding people from the contemporary politics of a nation, as well as
from the nation's future.
»Limitations
»Our study is not without limitations. We address one of these, potential biases due to selective inclusion of
historical data, by building the most expansive corpus possible and assessing its coverage. Moreover, this
limitation —common when using historical data— is a trade-off with the opportunity to mitigate common
response errors. Historical data like ours are not biased by social desirability or faulty memory; we can
accurately capture the rhetoric that radical groups chose to espouse.
»A second limitation is also related to the data type: achieving causal identification when using observational
data. With this in mind, we examine the direction of the association between support and rhetoric.
Using an exogenous variable, global oil price, in a moderation analysis, we offer evidence that external support
affects which variety of radical discourse is adopted, and not the other way around. These findings,
however, do not rule out unobserved confounders that could be responsible for our main result. But, such
unobserved variables would have to be correlated with external support and the rhetoric ratio, and have
a similar relationship with external support as a function of oil price. Because this condition is relatively
difficult to satisfy, we see a moderation analysis as a satisfactory way to understand the connection between
support and rhetorics without the strong assumptions necessary in alternative approaches.
»Finally, a third limitation, the scope of our insights into rhetoric, needs to be addressed with further
research. We do not examine the narrower scope of how case-specific cultural symbols and references are
used in the rhetoric. This would require an immersion into Deobandi and Salafist doctrine, as well as the
writings of thinkers and ideologues like Khomeini, Mawdudi, Qutb, and Faraj. Instead, we purposefully
maintain a relatively idealized perspective to be able to identify and conceptualize general forms of radical
rhetoric.
»We also do not examine the broader scope of organizational rhetoric. We aimed to analyze and interpret
variation within radical discourse, not how radical groups use language in different ways than, say, civil society
groups working within established norms and institutions. A comparison of rhetoric across different kinds
of social organizations would require expanding our corpus to include texts from non-radical organizations.
»CONCLUSION
»What kinds of rhetoric pattern radical discourse, and why do radicals adopt one kind over another? To
answer these questions, we conducted a computational abductive analysis of an original corpus of more than
23,000 pages published by Afghan radical groups between 1979 and 2001.
»We find two kinds of rhetoric within radical discourse. One, a rhetoric of subversion, targets enemies
very distinct from radicals and situates these enemies in an ongoing historically-scaled and inter-communal
struggle. The second, a rhetoric of reversion, is intimate. It turns inward, uprooting prevailing ideas and
ways of life through the judging and policing of personal and local morality and behavior. These findings
draw attention to a commonly overlooked aspect of radicalism: its everyday, ahistorical, and personal and
local gaze. In other words, radicalism is often conceived as an outward-facing attack against powerful,
well-established external ideas and actors (e.g., Ferree 2003; Calhoun 2012; Beck 2015), but we identify
rhetoric that urges radicals to profoundly change society by turning toward their immediate social and
temporal context —including themselves— in an effort to exclude unwanted beliefs, behavior, people, and
organizations.
»Our findings also illuminate how radical rhetoric falls along dimensions of connectivity and reach (Mische
2009; 2014). Through the former, rhetoric depicts a way for radicals to reach their goals —by offensive attacks
or socially intimate transformation— and, through the latter, it defines a timescale —ranging from the epochal
to a focus on the immediate past, present-day, and near future. Moreover, by articulating connectivity and
reach, rhetorics of radicalism construct a projection of the future that excludes specific ideas, people, groups,
and institutions present in a current society.
»Building on this insight, we suggest that radical discourse might
be usefully theorized as a discourse that rhetorically crafts an exclusionary boundary of futurity.
Next, we arrive at an answer for our second question. A radical rhetoric of subversion emerges when
radicals receive support from extra-local sources and, as a result, have weaker connections with their surrounding
local communities.
»In contrast, radicals are more likely to espouse a rhetoric of reversion when
groups have less external patronage and, consequently, a stronger relationship to local communities. Our
analysis shows how social relationships and social domains can impose constraints on the development and
adoption of rhetoric, but also allow for unexpected tensions between radicals and their supporters.
Alongside contributions to the computational analysis of discourse, our study offers a complement to the
relational perspective on radicalization (della Porta 2013; Alimi et al. 2015; della Porta 2018).
»Namely,
our findings point toward an avenue of research on how the dynamics of social relationships and discursive
boundaries shape the language of radicalism. Extending this relational approach to related discourses —populism, nationalism, and revolutionism, for example— can help us better understand how discourses of
contention construct different kinds of boundaries but, at the same time, share potential to engender new
forms of exclusion».
