Phobia: a corpus study of political diagnostics

This article is a rhetorical corpus study of the use of -phobia in online alternative media. The term phobia is used in the psychiatric domain to refer to a range of anxiety disorders, but is now also commonly used to identify social tensions. Terms such as transphobia and Islamophobia have within a few decades become central to contemporary political debate. The article examines in what way such coinages are used in a variety of online publications, and thus seeks to contribute to a better understanding of the entanglement of political and medical vocabulary. The study is based on data from the Genealogies of Knowledge Internet corpus, which are analysed in terms of collocation and other forms of linguistic patterning. The analysis reveals that online articles situated on the left of the political spectrum contain a large number of formulaic but open-ended lists of socio-political phobias, supplemented with other other undesirable attitudes such as sexism and racism. The article further reflects on the connections between the rhetoric of listing and intersectional analysis, on cultural conceptions of pathology, and on the polarised alignment of attitudes towards scientific and political issues in today’s public debate.


Introduction
I n June 2020, the world witnessed an international wave of antiracist iconoclasm. In the US, confederate memorials were important targets. On the European continent busts of colonial rulers were defaced. In the UK, statues of slave traders bore the brunt of the protester's anger, and on the website toppletheracists.org a map documented landmarks nominated for removal. Some saw the destruction of monuments as a shameful attempt at erasing history, while others were baffled there could be any genuine resistance to removing celebratory reminders to a history of oppression. The discussion was complicated by the fact that those who behaved worst abroad were often great benefactors at home. The call to topple statues was instigated by a campaign of the Black Lives Matter (BLM) movement. Major protests erupted in response to the killing of the black man George Floyd by a police officer-neither the first nor the last in a series of similar tragedies. Floyd quickly became a quasi-religious sacrificial figure, whose effigy expressed both the desperation and the hope of the protesters in the face of continued injustices and inequalities. The 2020 BLM protests had the support of numerous government officials, mass media outlets, international corporations, and educational institutions, and US polls suggested wide support among the general public (Parker et al., 2020). Voicing support for the protests, however, was not straightforward in the context of the COVID-19 pandemic. The global disease outbreak had led to quarantines, lockdowns, and social distancing measures, and in the months before the racial unrest, anti-lockdown protests were condemned for their recklessness. What made the BLM protests different?
One argument held that racial inequalities themselves are a matter of physical well-being, and that therefore "racism is a public health issue" (Murray, 2020). With a widely used metaphor deriving from that sentiment, supporters of the BLM protests argued that "racism is a pandemic", and one that should be treated with urgency (e.g., Herman and Davis, 2020). A number of lawmakers and health experts in the US and the UK did not oppose this metaphorical usage, but rather amplified its legitimacy by ascribing to the view of "discrimination as a chronic illness" (Singh, 2020). Thus, while newspapers and public health bodies were at first quite particular about the exact designation of the term pandemic and when to apply it to COVID-19, the experts left their posts as terminological gatekeepers once the metaphorical bridge into the sphere of racial tensions was crossed. This metaphorical operation illustrates the profound entanglement of science and politics in today's public sphere: BLM protests were ratified by at least a portion of the scientific community, whose members could simultaneously continue to claim that freedoms should be restricted when movement was not motivated by racial struggle. Thus, prejudiced attitudes were temporarily characterised as a greater threat to society than viral infections. Meanwhile, nothing stopped politically opposed protesters from seeing people of different ethnic backgrounds as parasites and from ascribing, as of old, "the sickness of the nation to the influx of foreign populations" (Inda, 2000, p. 47). Indeed, the arrival of foreign elements on the borders of the body politic, as well as their development within the host society, is regularly conceptualised as a parasitical source of contamination and contagion (Esposito, 2011, p. 2). Thus, the conceptualisation of perceived threats as diseases occurs across the political spectrum and, in general, political and scientific discourse share a vocabulary of illness. This article is concerned with a remarkably productive element of this shared vocabulary, namely the lexical item phobia.
The Oxford English Dictionary (OED) lists 139 entries ending in -phobia. According to the dictionary, the most frequent compounds containing phobia are, in chronological order of appearance, photophobia (1772), agoraphobia (1871), claustrophobia (1879), homophobia (1901), and xenophobia (1909). All these words etymologically derive from Greek and Latin roots, and the oldest such compound is potentially hydrophobia (1547): "A symptom of rabies … an aversion to water or other liquids, and difficulty in swallowing them" (OED, 2020c). The symptom is mental as well as physical, and the same holds for photophobia, which refers to an aversion to light caused by neurological sensitivity (OED, 2020d). Agoraphobia and claustrophobia can both be described in terms of dread of confinement, and although they may cause physical discomfort, they are said to occur without there being a bodily cause. Indeed, for a person to be diagnosed with agoraphobia, symptoms must be in excess of any corporeal medical conditions (American Psychiatric Association, 2013, p. 218). Xenophobia is defined as a "deep antipathy to foreigners" and homophobia as "hostility towards, prejudice against, or fear of homosexual people or homosexuality" (OED, 2020b;2020e). This definition of homophobia is documented only since 1969. Earlier and alternative uses, now largely obsolete, signified hatred of men or humanity, or "fear of being thought to be homosexual" (OED, 2020b). In contrast to their precursors, xenophobia and homophobia refer neither to physical symptoms of affliction, nor to mental disorders that can be diagnosed, but rather to a general idea of prejudice. The social vocabulary that equates prejudice and antipathy with phobia is growing, and terms such as transphobia and Islamophobia have within a few decades become central to contemporary political debate. This article is a rhetorical corpus study of the use of -phobia in online alternative media. It seeks to investigate who uses the term, in what way, and for what purpose, and thus to contribute to a better understanding of the entanglement of political and medical vocabulary.

Data and methods
The data for this study consist of a large set of online texts collected in the context of the Genealogies of Knowledge project (GoK). GoK is a multidisciplinary corpus research project, funded by the AHRC from 2016 to 2020. The project examines how constellations of concepts central to political and scientific, expert discourse are transformed through translation and other processes of mediation. For example, Jones (2019) illustrates how different Victorian translators of classical texts mould the concept of statesmanship to suit their own worldview, as well as the sensitivities of their cultural environment. The project has built separate corpora of ancient Greek, Latin, medieval Arabic, and modern English, languages that each at a particular historical moment served as a lingua franca for the transmission of knowledge. The English corpus consists of two main subcorpora, which differ in terms of the medium that is covered: one corpus mainly contains material from printed books, whereas the other corpus consists of born-digital texts, taken from a variety of virtual outlets such as blogposts or online magazines. A further distinction is that the subcorpus consisting of printed material largely represents canonical academic knowledge, whereas the Internet corpus represents opinions, narratives, and positions that are situated outside of the mainstream.
Researchers can work with either of these two English subcorpora separately, with both English subcorpora combined, or with a smaller subcorpus containing any selection of texts from these resources. The data for this study consists of all texts included in the English Internet corpus. The outlets populating the Internet corpus reach a large audience but are not considered mass media in the sense that traditional newspapers are. At the time of writing, the Internet corpus is still growing considerably, and contains over 4 million tokens, sourced from >2900 texts ARTICLE HUMANITIES AND SOCIAL SCIENCES COMMUNICATIONS | https://doi.org/10.1057/s41599-020-00593-w from 35 different outlets 1 . Among other publications, one finds openDemocracy, Climate Depot and Activist Post, websites that shift between scientific and political viewpoints across broadly imagined spectra of dissent and contestation. Most texts in the Internet corpus were published between 2010 and 2020. Earlier studies drawing on this corpus have examined, in specific contexts, the online representation of migrants (Baker, 2020), political authority (Buts, 2020), and women's agency (Karimullah, 2020). Specificity of context must be stressed here, as the corpora were compiled thematically to capture common vocabulary at the interface of science and politics, within the limited sphere of Anglophone alternative media. At present, voices from the left are most broadly represented. Thus, the GoK Internet corpus is not a general corpus of online language use, but a window on particular instances of discursive contestation in the digital era. It is not, and could not be, free from bias.
All corpora built by the GoK project are made available through a freely accessible concordance browser (GoK, 2020). The browser's main interface is a classic keyword-in-context (KWIC) view; a search for a given keyword returns all of its occurrences in the corpus, accompanied by a stretch of surrounding text on either side. Longer stretches of text can be viewed through an "extract" function, but full texts are not made available due to copyright restrictions. In addition to purely textual representations, the software offers functional prototypes of a number of visualisation tools (Luz and Sheehan, 2020). In the analysis below, I make use of two such tools. The first one is called Mosaic, and represents the most salient lexical items in the environment of any given keyword by means of coloured tiles. The tool is partly inspired by principles of visualisation theory, and partly by the work of the corpus linguist John Sinclair, who presented similar diagrams in his didactic analyses Sheehan, 2014, 2020;Sinclair, 2003). The Mosaic tool can visualise the contents of a concordance based on raw frequency counts, or based on a number of different collocation measures such as mutual information, z-score, and log-log (Luz and Sheehan, 2020, p. 11). The second tool is called Metafacet and serves to provide a graphic overview of the material represented in a concordance on the basis of any attribute recorded in the metadata, such as author, translator, or outlet.
Next to concordance-based representations, the GoK software suite also offers information about one's working corpus as a whole, by means of tools such as the "corpus description browser" and the "frequency list". The corpus description browser lists metadata attributes such as author and year of publication for each text in the corpus, along with a summary of its content. This information can also be requested for individual concordance lines. The frequency list ranks all word types in the corpus from most to least common. In this study, the GoK Internet corpus is queried in full, and the frequency list records >80,000 unique word types for this corpus. The most frequent word types are mostly grammatical elements such as this and that, but some content words are included in the top 100, such as power, science, rights, climate, and government. The output of this basic frequency count, thus, reflects the goal of the GoK Internet corpus, which is to document the online evolution and contestation of political and scientific vocabulary. As previously stated, the Internet corpus is not representative of the Internet as a whole, but gives a good indication of some of the central debates around science and politics in English-language online alternative media from the early twenty-first century.
While the current study makes use of several basic quantitative operations, the analysis presented is mainly qualitative. This is a rhetorical corpus study, meaning that lexical patterns are taken to represent linguistic strategies of positioning and persuasion. The purpose of the analysis is to gain insight into how and by whom the element -phobia is used, for what purpose, and to what effect. In order to gather all relevant textual material, the GoK Internet corpus was queried for all lexical items corresponding to the regular expression. *phobi[ac] 2 . Regular expressions represent search patterns that can be instantiated by a variety of concrete forms. The dot and the asterisk or wildcard in the expression together stand for any number of occurrences of any character, as long as the string is continued by the sequence -phobi, and ends in either an -a or a -c. The expression will thus not only find words ending in -phobia, but also related words ending in -phobic. For instance, both transphobia and xenophobic will be found, if they are present in the corpus. In the following analysis, the concordance lines containing the cluster of keywords generated by the search are examined in search of recurrent linguistic, as well as cultural patterns. Cultural patterns can be derived from the metadata of the concordance lines: who uses words relating to phobia in the corpus, and who does not? Does everyone use the same compounds or is there variation across authors and publications?
In order to discuss linguistic patterning, the following analysis takes as its starting point the work of Sinclair (e.g., 2004) and Stubbs (e.g., 2009), more specifically in terms of the search for extended units of meaning. The empirical observation that words usually occur within a relatively fixed pattern of use has been interpreted as evidence for the idiom principle, which is the hypothesis that "a language user has available to him or her a large number of semi-preconstructed phrases that constitute single choices, even though they might appear to be analysable into segments" (Sinclair, 1991, p. 110). Single words thus tend to form part of extended units, and evidence for this hypothesis is provided by collocational phenomena. Collocation is here understood as the frequent co-occurrence of lexical items, and as noted in the discussion of the GoK software, the research community handles a variety of measures for calculating the statistical significance of the association between two "collocates". Collocation is an observable phenomenon that may point towards various more abstract linguistic inferences, such as colligation, semantic preference, and semantic prosody.
Colligation can be defined as "the co-occurrence of words with grammatical choices" (Sinclair, 2004, p. 174). For instance, early work by Tognini-Bonelli (1993, p. 196) shows that the adjectives actual and real, while quite similar at first sight, show very different grammatical behaviour, as actual is much more likely to be preceded by a definite article. Semantic preference is "the association of formal patterning with a semantic field" (Stewart, 2010, p. 10). For example, Partington (1998, pp. 34-39;2004, pp. 145-146) illustrates that the intensifying adjective sheer tends to occur together with words expressing "magnitude" or "force", but also "persistence" and "emotion". Adjectives sometimes considered synonymous with sheer, such as complete and absolute, do not share these preferences. Finally, semantic prosody is an attitudinal concept that indicates the "communicative purpose" of an extended unit of meaning, and therefore its boundary (Sinclair, 2004, p. 34;Stubbs, 2009, p. 125). Units of meaning containing build up, for instance, seem to invariably come with a positive attitude when the verb is used transitively, as in build up organisations (Louw, 1993, p. 171). However, when used intransitively, as in toxins build up, the evaluation tends to be negative (Louw, 1993, p. 171). Thus, meaning is encoded in the structure of the extended expression. Features such as semantic prosody tend not to be intuitively accessible, which makes uncovering the relation between words and their direct environment, as well as the attitude expressed by such extended units, invaluable for improving the work of translators and other text producers and mediators (Stewart, 2009). Furthermore, it has been argued that breaches in established patterns of use may be indicative of irony or insincerity (Louw, 1993). If someone tells you that joy builds up when you enter the room, you may feel that something is off. Nevertheless, language changes, norms vary across genres and registers, and semiotic patterning is never absolute: excitement builds up without a hint of irony. The concept of semantic prosody is widely researched, but also has met with considerable criticism for its lack of a rigid definition (e.g., Whitsitt, 2005). Nevertheless, the idea of semantic prosody at the very least provides a useful reminder that recurrent linguistic patterns are not accidental or anomalous, but arise in accordance with a communicative purpose that extends beyond the provision of information.

Analysis
At the time of writing, a search for the regular expression. *phobi [ac] in the GoK Internet corpus returns 345 concordance lines. 197 of those lines end in -phobia, and 148 end in -phobic. Table 1 lists the five most common compounds among the nouns and the adjectives, together with their frequency.
The correspondence between the order found for nouns and adjectives is striking, as are the similarities in frequency. Only in relation to Islam is there a clear preference for the nominal form. Correspondences extend beyond the keywords or node words themselves into the sphere of collocation. Within the standard span of 4 words to the left and the right of the keyword, xenophobia collocates strongly with racism, which occurs at all positions in the vicinity of xenophobia except for N + 3, the third position to the right of the keyword. Racism is the most frequent collocate at the N − 2 position, and the most frequent overall collocate except for the function words and, the, and of. For xenophobic, racist similarly occurs in every position from N − 3 to N + 2, and is the most frequent collocate at the N − 2 position. Figure 1 is a Mosaic representation of the collocates of xenophobic ranging from the N − 2 to the N + 2 positions, based on the log-log measure. Some words, such as racist and nationalist, occur on both sides of the keyword, but overall the majority of items on the left of the keyword are other adjectives, while on the right side mostly nouns are found. Some of those nouns, such as prejudice and attitudes, refer to mental states. Others, such as impulses and reaction, suggest that those mental states may prompt a behavioural response. Finally, the mosaic representation implies that those responses may be harmful, as in violence and attacks.
The association with other forms of prejudice such as racism and sexism is strong for adjectives as well as nouns, but is most pronounced for the latter. Homophobia, for instance, is unlikely to appear in isolation in the corpus. Most occurrences of homophobia form part of a list that is relatively formulaic, as the examples in Fig. 2 illustrate. Figure 2 shows multiple occurrences of the pattern "racism, sexism, homophobia". The list is in some instances expanded to the right of the keyword with a variety of further prejudicial phenomena, such as transphobia, xenophobia, disablism and ableism. The latter two words are usually taken to express the same attitude: discrimination of or prejudice towards the disabled. The order of the list is set to some extent: in the corpus, within a span of four words, 22 cases of sexism occur after racism, while only two occurrences of sexism are recorded before racism. Similarly, there are nine occurrences of homophobia after sexism, while there are none that come before. Despite the tendency of the list to conform to a set of implicit rules, there is considerable variation among the concordance lines, and the selected examples below illustrate this mixture of patterning and variation.
Example 1: All the reactionary phenomena-misogyny, homophobia, racism, contempt for the poor, bureaucratic statism, deference to the mighty, repressive conformism, hatred of the intelligentsia, feudal nostalgia and, especially, the ethnic-racial closing of the cultural horizon-are rampant. (Tamás, 2016, "Into the Dustbin of History", Salvage).
Example 2: The debates illustrate a toxic combination of cowardice, misogyny, homo-and transphobia, religious fundamentalism, and a closing of space for realistic discussion of how to solve pressing global challenges by governments and civil society together. (Fried, 2016, "Ending HIV: ideology vs. evidence at the UN", openDemocracy).
The first example comes from an article written by a wellknown Hungarian philosopher, and speaks of Eastern Europe and its current rejection of egalitarian and internationalist values. The second example is from an article written by a health justice specialist, who argues that conservative values hinder education and engagement policies that would benefit those likely to contract HIV. Although they vary considerably in scope and topic, both examples diagnose a certain cultural climate as repressive and regressive by means of a list of symptoms consisting of prejudiced principles and practices. These reactionary phenomena (example 1) form a toxic combination (example 2). The most frequent collocate of toxic in the Internet corpus is chemicals, meaning that the literal usage of toxic persists, but the phrase toxic combination is always used metaphorically, and in two out of three instances it occurs in the context of a list of prejudiced attitudes. Thus, the term toxic simultaneously functions in the vocabulary of the natural sciences, and that of condemned sociopolitical behaviour.
Both examples speak of a closing, opting for the spatial metaphors of either the cultural horizon (example 1), or the space for realistic discussion (example 2). Both examples also make mention of human characteristics, such as cowardice or conformism, that do not, like the various phobias, revolve around an identifiable object of aversion. Finally, they both foreground misogyny, Collocates two positions to the left and right of the keyword, using the log-log measure. the hatred of women. The element miso-, which does not occur in the corpus except as part of misogyny, is a lot less productive than -phobia, which has also come to encompass "hatred". Phobia can indeed designate a wide variety of emotions, and it can also refer to a broad behavioural spectrum, as people can not only be subject to a xenophobic environment, they can also be responsible for creating it, and the impact of emotions such as fear, hatred, or aversion may vary depending upon whether or not they have tangible consequences. In short, uses of phobia and its derivatives can also be differentiated across spectra of agency and potency, as the set of examples presented below will further illustrate. Example 3: In the place of the imagined community of nationalism and its bedfellows of racism, sexism, homophobia, xenophobia and other forms of oppression, we must construct a real community based on the material interdependence of all life on earth. (Forman, 2015, "Theses on a unionism beyond capitalism", ROAR Magazine) Example 4: "The members of the Security Council", according to a UN press release, have "strongly condemned all forms of racism, racial discrimination, xenophobia and related intolerance". (Heinze, 2014, "Are all forms of prejudice really equal?", Left Foot Forward) Example 5: Sure, she said, but it's much more than that: "He's been racist, he's been sexist, he's been Islamaphobic, he has been anti-LGBTQ." 3 (Drum, 2018, "Hillary Speaks", Mother Jones) Example 3 is quite specific about the category that contains the phobias and isms under scrutiny: they are forms of oppression. Furthermore, the central node connecting the rejected attitudes is identified as nationalism. The author of the relevant article is a labour educator and organiser, and the article consists of 49 short fragments about the current state of society, and about a proposed way out of the current capitalist predicament: a "global worker's economy". Example 4 contains a statement from the United Nations (UN), which was sent out in response to a shooting at the Brussels Jewish Museum. The article rejects the UN's phrasing, arguing that "inequalities are not all equal", and that violence against Jews is not just bigotry, but symptomatic of the pervasive anti-Semitism evident in conspiracy theories about global Jewish control and subversion. Example 5, which contains an alternative spelling of Islamophobic, quotes Hillary Clinton on Donald Trump. The statement, backed up by a video on the web page, is an accusation. In Clinton's wording, racism and Islamophobia do not seem to be conditions or beliefs, but actions: you can actively engage in Islamophobia, or in dismissing the concerns of LGBTQ people. Note that the abbreviation LGBTQ is itself an abbreviation containing several socio-cultural groups, and that in this case homophobia and transphobia on the part of Trump are therefore implied.
In general, the examples confirm that nouns of prejudice collocate with other nouns of prejudice, and adjectives of prejudice collocate with other adjectives of prejudice. The relation is not exclusive, but it is remarkably strong, to the extent that we can speak of parallel patterns of collocation and colligation: the adjectives that textually precede transphobic, such as racist, will correspond to the nouns preceding transphobia, such as racism. The extended unit of meaning here is not a phrase that can be subjected to a grammatical analysis of hierarchy and dependency, but a list, a horizontal pattern without clear boundaries. Overall, words ending in phobia and phobic have a semantic preference for other social attitudes deemed undesirable. Indeed, the evaluation of those attitudes seems to be invariably negative. The communicative purpose of the expressions is to accuse, to condemn, and thus, in terms of cultural norms, to prohibit.
This semantic prosody of judgement stands in close relation to the enunciative position of those who produce the lists of undesirable attitudes, as resemblances between the various quoted sources indicate. The authors tend to have an academic profile, but an activist motive. The websites are all Anglophone, and are mostly based in the UK and the US, but they engage with global debates and international affairs. All publications quoted so far situate themselves on the left progressive side of the political spectrum, with hints, particularly in Salvage and ROAR Magazine, towards a Marxist revolutionary outlook. One may question whether the vocabulary of phobia is indeed restricted to this cultural environment, or whether these characteristics primarily reflect a bias of the GoK Internet corpus. This question can partly be answered by comparing which outlets in the corpus use the items -phobia and -phobic the most, which publications use them the least, and which do not use them at all. The Metafacet representation in Fig. 3 provides an overview of all occurrences of -phobia or -phobic in the corpus, categorised by Internet outlet and sorted by frequency. Although the order of the frequency ranking changes slightly when the nominal and adjectival forms are queried separately, the outline of the graph remains mostly the same. Figure 3 shows that the website Salvage.zone uses -phobia and -phobic the most. There are 93 occurrences spread across >30 articles, out of the 73 articles from this website in the Internet corpus. At 28 instances, Islamophobia is most frequently addressed. Salvage is an explicitly communist publication, and the emblem of the hammer and the sickle is embedded in its logo. Their pessimism about the prospect of radical change for the better leads them to write "by and for the desolated Left" (Salvage, 2020). At the other, lower end of the Metafacet graph, we encounter a publication situated on the opposite side of the political spectrum: Amerika situates itself "furthest right" and calls for "zero tolerance for Leftism and its associated ideas (feminism, egalitarianism, pluralism, diversity)" (Amerika, 2020). Amerika.org only has a single hit for xenophobia, from an article that argues that conflict is the logical consequence of migration, and that "diversity destroys civilisation" (Stevens, 2017). The use of the term xenophobia in the article is through quotation from an article in The Atlantic about the difficulty of ensuring both diversity and equality, especially in the face of significant demographic change (Thompson, 2017). There are thus no instances of original, unattributed writing containing -phobia and -phobic in the 99 articles from Amerika that are included in the corpus. Despite being at the opposite ends of the political spectrum, Salvage and Amerika share a number of characteristics. They both present the status quo as most dire, and hope to inspire society to rise from the ashes of its imminent demise, but whereas Salvage argues for a "communism of the ruins," Amerika hopes to "restore Western civilisation" (Amerika, 2020; Salvage, 2020). They share radicalism and a predilection for the rhetoric of tragic hyperboles, but they condemn each other's political position, and seem to do so with a vocabulary that is distinctly their own. In Salvage, phobia is ever present, in Amerika it has no role to play.
The rest of the Metafacet graph in Fig. 3 shows that it might be possible to extrapolate the pattern. Many left-wing outlets, radical or otherwise, employ the vocabulary of phobia, while others, such as CapX and Newsmax, are much less inclined to do so. Given the semantic prosody of judgement established earlier it can thus be argued that the social uses of phobia constitute a pattern of accusation that does not only identify the accused, but also the accuser, whose ideological preferences are reflected in the very words used. Such observations must remain tentative, however, as the corpus is not balanced enough to investigate ideological divisions more reliably. At present, the bulk of the corpus material is situated on the left of the political spectrum. Additions of more diverse material are planned, but if the hypothesis of restricted use is correct, they will not alter the graph in Fig. 3 much. The corpus currently contains material from 35 outlets. For 13 of them, no occurrences of -phobia or -phobic are documented. Examples include Bad Science, Desmog, and Answers in Genesis. The latter is a creationist publication, and the former are blogs, respectively about the misuse of scientific information in public debate, and about climate change, energy and the environment. Thus, the main concerns of these publications are not with issues of race, gender, and interpersonal aversion. Nevertheless, some publications that publish in similar realms of scientific contestation, such as Creation Ministries International and The Risk-Monger, do feature instances -phobia or -phobic. However, these instances are restricted to specific contexts of argumentation, as exemplified in example 6 and 7.
Example 6: The anti-GMO, chemophobic tribal gurus have made their followers so afraid and enraged that while they turn to them for trust and affirmation (and donations, subscriptions and online purchases!), they are getting more and more stressed and vulnerable. (Zaruk, 2016, "Living in the Age of Stupid: How to comprehend Brexit, Trump and the Anti's", The Risk-Monger) Example 7: Indeed, one official tweet from the organisers declared that "Colonisation, racism, immigration, native rights, sexism, ableism, queer-, trans-, intersex-phobia, & econ justice are scientific issues." Thankfully, it was subsequently deleted. (Price, 2017, "On the streets with the "March for Science" protesters", Creation Ministries International) Examples 6 and 7 are contrarian, polemical statements. The first statement is made in a discussion of a March against Monsanto, and the second, as the title indicates, is about a March for Science. In both cases, the authors do not identify with the general sentiment of the participants in the march.  presents himself as an "EU risk and science communications specialist", and he preaches against the label of harm that genetically manipulated products and pesticides receive from the organic food industry. All mentions of chemophobia in the corpus occur in articles by this author, although the term circulates widely and has been in use for decades to indicate "a fear of chemicals, especially as used in artificial food products or industrial processes" (OED, 2020a). Price (2020), the author of example 7, has "studied creation science for many years", and joined the March for Science to question participants about their views. He finds that the march was "little more than a blatant attempt to hijack the word "science" for left-wing political purposes", and supports this statement with the example referenced. Considering the productive use of -phobia, this example thus presents an explicit argument against the merger of scientific and political vocabulary.
Example 6 mentions chemophobia, a coinage that functions in a discourse that runs across agrarian, medical, governmental, and corporate debates, but which does not identify a human characteristic as its object of aversion, and thus constitutes a different usage from the previously discussed instances of -phobia and -phobic. Example 7 does use the same vocabulary as the leftwing sources discussed earlier, but does so only to institute critical distance. Indeed, quotation can be used to reinforce authority, as in the Clinton case discussed above, but also to question it. In sum, the vocabulary of phobia as socio-cultural prejudice is partly confirmed to be intimately entwined with left-wing online discourse. The study of corpora other than the GoK Internet corpus may confirm or disprove this pattern. A final, very tentative observation from the data is that the social vocabulary of phobia and its pop psychiatric counterpart can co-exist: in the corpus, a mention of claustrophobic relating to a railway station does not preclude the use of xenophobic in the same article (Horn, 2014). Nonetheless, claustrophobic is mentioned only three times in the corpus, and no other mental condition occurs.

Discussion
At the 1970 annual meeting of the American Psychiatric Association (APA), the proceedings were disturbed by gay activists who demanded that homosexuality should no longer be classified as a mental disorder (Drescher, 2015, p. 387). A process of deliberation stretching over many years ensued. Eventually, in 1973 the decision was made to remove homosexuality from the Diagnostic and Statistical Manual of Mental Disorders (DSM). Nonetheless, a number of psychoanalysts objected, and asked for a referendum. Ten thousand votes were cast, and the majority decided to uphold the decision to no longer classify homosexuality as a mental disorder (Drescher, 2015, p. 388). The scientific view of homosexuality as a pathology had lasted less than 100 years, and was revoked by vote-a procedure usually reserved for political, and not scientific purposes. Currently, activists are calling for the removal of gender-related diagnoses from the DSM. Progress is slow, as a balance must be sought between reducing the stigma suffered by the transgender community, and "maintaining access to care" (Drescher, 2015, p. 392). If there is no medical diagnosis, no custom care can be provided. In the long run, however, gender dysphoria is expected to leave the realm of disorders, as questioning the sanity of transgender people is increasingly looked down upon, and labelled as transphobia. The spread of the term homophobia and the declassification of homosexuality as a mental disorder took place at the same point in history. A similar reversal of perspective is taking place in relation to transgender matters, and it seems clear that phobia is the preferential term for coining new words for condemned attitudes. There is little or no evidence that negative responses to homosexuality constitute a phobia in the medical sense of the word, and for that reason alternative terms such as homoprejudice have been proposed (Logan, 1996). They have not been successful, and one can at least surmise that the choice for phobia prevails because it frames discriminatory behaviour as resulting from mental illness (Wickberg, 2000, p. 45). Thus, the accuser, by not translating the word phobia into a more transparent term, may claim the support of scientific judgement. Yet does the characterisation of prejudice as mental illness mean that care will be provided for the biased?
The question is not meant to be facetious. As the voting procedure in relation to homosexuality indicates, cultural and political pressure can ultimately influence scientific decision-making. Indeed, it has been argued that the receiving public cannot be conceptualised as a passive recipient of medical knowledge, but actively contributes to its formation (Engebretsen et al., 2017). Psychiatric classifications, the symptoms of which are not restricted to the body, are especially open to cultural influence, and indeed a scale for measuring Islamophobia was already being devised a decade ago (Lee et al., 2009). Treatment procedures for attitudes such as racism, sexism, and homophobia are already being implemented, as exemplified by the case of unconscious bias training. In the UK, such training is meant to "increase awareness of one's unconscious bias and its impact on people who belong to groups denoted as having "protected characteristics" under the Equality Act 2010" (Atewologun et al., 2018, p. 6). Protected characteristics may relate to issues such as race, sex, disability, and sexual orientation, and unconscious bias training thus largely seeks to combat the phobias and isms discussed in the previous sections. Training is meant to improve the conditions for diversity, inclusion, and equality in a variety of environments, and is widely applied in workplace settings. However, there is little or no evidence that awareness of bias, conscious or unconscious, results in the envisaged behavioural changes; on the contrary, bias training may backfire, because people who feel they are being influenced may actively start to develop thoughts in the opposite direction (Atewologun, 2018, pp. 20, 31). As illustrated in the concordance lines above, contrarians will resist re-education, and may derive pleasure from flaunting an objectionable stance.
In addition to personal temperament, larger cultural clashes explain why terms such as transphobia and Islamophobia are intensely used in some environments, but rejected in others. The political left was historically preoccupied with economic inequality, but over the last decades has paid increasing attention to the various categories that make up a person's identity, resulting in the fact that today, "class struggle is just one species of identity politics" (Laclau, 2000, p. 203). As several minorities and disadvantaged groups staked out a space of recognition for their own experience, efforts against discrimination extended into the field of speech. Thus, first on university campuses, and later in ever larger swathes of public society, one saw the rise of "political correctness, a social norm that prohibits people from publicly expressing their beliefs or opinions without fearing moral opprobrium" (Fukuyama, 2018, p. 101). A common concern in this regard is the perceived push towards "diversity in everything but opinion" (Salzman, 2016). Thus, a proportion of those deemed privileged by other groups in society now view themselves as increasingly stigmatised and disadvantaged by the cultural norm. Polarisation in this regard has led to ample critique of the main discourse represented in the analysis above, a discourse that aims to raise awareness of various spaces of prejudice and discrimination. On far-right internet forums, for instance, users frequently point out that feminism and Islam are irreconcilable, and that the "politically correct establishment" displays its own hypocrisy by simultaneously condemning sexism and Islamophobia (Törnberg and Törnberg, 2016, p. 414). Such contradictions are then frequently interpreted as revealing that the left is mainly engaged in virtue signalling, or showing off one's awareness of social issues to acquire social status without actually engaging in behaviour that results in meaningful change. While political correctness is usually seen as prohibitive, placing restrictions on what one is allowed to say, listing various categories of discrimination and oppression would constitute an example of mandatory political correctness. To partake in certain sections of the public sphere, one has to display awareness of several layers of injustice, without necessarily further dealing with the various classifications involved.
However, such a superficial interpretation does not do justice to the difficulty of negotiating which patterns of discrimination, and which instances of injustice, should be articulated when and how. The difficulty is best illustrated with reference to the discourse on intersectionality as popularised by Crenshaw (1991Crenshaw ( , pp. 1242Crenshaw ( -1243, who influentially noted that "feminist and antiracist discourses have failed to consider intersectional identities such as women of colour". A notion of "triple oppression" regarding the maltreatment of Black working class women developed, which later led to the question of "how many social divisions are involved and/or which ones should be incorporated into the analysis of the intersectionality process" (Yuval-Davis, 2006, pp. 195, 201). A further complication of intersectional analysis is that, while it aims to uncover the reciprocal influence between distinct forms of oppression, it cannot do so from the vantage point of a stable hierarchical view, as this would lead to what has been called the oppression Olympics, a "victimhood tournament" in which competing factions claim to be more deserving of recognition than others, based on their greater share in the suffering of injustice (Martínez, 1993, p. 23). Thus, while intentions may be noble, the discourse of diversity that aims to identify and rectify various intertwined manifestations of social phobia is always at risk of succumbing to "the etcetera of the list" (Eco, 2009, p. 81). This does not mean, however, that the listing practice itself cannot provide a useful overview of sites of struggle to be further unpacked, articulated, and integrated. As indicated in the analysis, a list of political phobias, supplemented with other undesirable attitudes such as sexism and racism, tend not only to identify the accused, but also the accuser, who aligns with a particular discursively constituted group of people-those who intend to combat bias. In this context, the enumeration of undesirable attitudes is an assertion of their equivalence in the context of a specific field of struggle (Laclau and Mouffe, 2000, p. 63). Within this field of struggle, broad political alliances and divisions take shape in relation to the central antagonistic metaphor of prejudice as disease. The term phobia is highly productive in shaping this antagonism-it balances the logics of equivalence and difference that permeate any discursive linking process (Jacobs, 2018, p. 304): all political phobias have their own specific characteristics, but they can be equated for the purpose of enacting social change. In this process, the exact meaning of phobia becomes diluted, and the term increasingly serves as an empty signifier, a performative means of "political identification", and ultimately, mobilisation (Laclau, 2006, p. 656).

Conclusion
This article has presented a rhetorical corpus study of the use of -phobia and -phobic in online alternative media. The collocational profile of terms such as xenophobia and homophobic indicates that there is a strong tendency towards the formation of condemnatory lists, particularly in outlets situated on the radical left of the political spectrum. It was argued that the phenomena of intersectionality and identity politics are central to this discursive formation, and finally it was suggested that the work of Laclau and Mouffe (e.g., 2000) may prove helpful in further contextualising the development of antagonism around the issue of prejudice. Furthermore, the article has argued that science and politics today constitute deeply intertwined spheres of discourse and practice, as is evident not only from the representation of prejudice as illness, but also from rising suspicion towards vaccination, or from the fact that climate change has become a major political point of contention. In various countries, policy debates have become heavily polarised around issues such as greenhouse gas emissions and renewable energy. Governments form and fall over the topic of migration, which today is more likely to be caused by environmental change than by human aggression (Gemenne, 2015, p. 169). The confrontations between climate change denialists and alarmists, as opposed factions have come to identify each other, are often less informed by the melting of the ice caps than by a larger set of beliefs concerning governmental and corporate control, the predictive power of scientific evidence, and the negotiation of new meta-narratives about human destiny.
Thus, positions in seemingly unrelated political and scientific debates tend to cluster together, forming broad axes of opposition.
Climate activists, for instance, tend to align their discourse with political preoccupations that extend beyond the conservation of nature, and the vocabulary of phobia and ism plays an important part in this alignment. The global Extinction Rebellion movement firmly embeds in its core values a statement against "racial domination, sexism, anti-semitism, islamophobia, homophobia, ableism, class discrimination, prejudice around age and all other forms of oppression including abusive language towards others" (Extinction Rebellion, 2020). This progressive stance is not selfevident in an environmentalist context. Conservatives could quite reasonably argue for the conservation of nature, but the distribution of causes and positions across today's political landscape has not evolved in this direction. A more extreme counterexample involves the Christchurch mass shooter, who explicitly identified as racist in his eco-fascist manifesto The Great Replacement (Tarrant, 2019). His concern was first and foremost with the preservation of an environment in which white people would be able to prosper without outside interference, an environment, which he presented as under threat of mass migration, and ultimately, "white genocide" (Tarrant, 2019). White genocide is considered a conspiracy theory, in as far as it postulates a deliberate attempt of powerful actors to first erase the culture and history of white people, and later to effectuate their physical extinction. In response to a wave of iconoclasm witnessed in the spring of 2020, such concerns once again reached mainstream political debate. To the satisfaction of some, and the indignation of others, Churchill's statue in London received the caption "was a racist." Graffiti on confederate monuments in Charlottesville, VA, spelled the words "the pandemic". For some, the statues represented noble values, for others they represented the planetary evil of contagious disease.
In the background of the racial struggle for historical rectification, a massive row on Twitter broke out: JK Rowling, famous for writing the Harry Potter book series, had implied that an article using the phrase "people who menstruate" would have been better off using the term "women" (Rowling, 2020). On 20 June 2020, 2 weeks after being posted, the Tweet had >32,000 replies, mostly discussing whether or not the Tweet was transphobic. Rowling clarified and defended her position in a blog post and in a series of follow-up Tweets, one of which, about the reality of the concept of sex, received even more replies. In response to the uproar, "politicians, novelists and actors have been called upon to make their positions known", and many "have stepped politely away from Rowling's side" (Thorpe, 2020). Thus, JK Rowling's Twitter page and the statue of Winston Churchill were similarly and almost simultaneously besieged, illustrating that revered cultural icons can quickly fall from grace once accusations of racism or transphobia are made. The semantic ersion of the signifier phobia could in principle mean that the link to the domain of mental illness is gradually losing ground. However, the link seems rather to be streghtening, as indicated by the ongoing efforts to consciously rewire bias, and by the ostracism that awaits the persistently prejudiced. The diagnosis of a form of social phobia is followed by harsh treatment, and as the use of compounds containing phobia can increasingly be interpreted as a call to action, it is of great importance to study who uses it, in what context, and for what purpose. The present article has attempted to make a start in that direction.

Data availability
The Genealogies of Knowledge Internet Corpus is available via: http://genealogiesofknowledge.net/software/.