To Trump (v): /tɹʌmp/ "to fabricate, devise"

Transcription

1 To Trump (v): /tɹʌmp/ "to fabricate, devise" ANALYSING THE (AB)USE OF LANGUAGE IN POLITICS: THE CASE OF DONALD TRUMP Ana Alonso Curbelo School of Sociology, Politics and International Studies University of Bristol Working Paper No

2 Title: A Bourdieusian Analysis of the (ab)use of Language in Politics: The Case of Donald Trump Name: Ana Alonso Curbelo Abstract: This paper analyses political language in the setting of Donald Trump's recent 2016 U.S. presidential election campaign. Viewing speech as a material, social and cultural phenomenon, rather than simply a linguistic one, the text-analysis software package IRaMuTeQ was used to analyse 15 rally speeches given by Donald Trump, focusing on how he deployed language and political discourse to appeal to different audiences. The two analyses performed, i.e- analysis of similarities and descending hierarchical analysis, reveal that Trump adapted his political stances and language depending on the political party dominating the state where the rally speech was given. To comprehend how the identified language patterns reflect (and impact) the 'space of political position-takings' in U.S. society, I draw on Bourdieu's work on Language and Symbolic Power (1991), concluding that Donald Trump's language and political discourse were instrumental in his pursuit of symbolic power and legitimisation. Keywords: Donald Trump, Bourdieu, discourse analysis, linguistic space, strategies of legitimisation 2

3 1. Introduction The election of Donald Trump, a billionaire businessman and television personality, as president of the United States, shattered expectations from most pollsters and pundits, media outlets and politicians from both main political parties. Moreover, it created a global political earthquake that swept away all preconceptions about politics and presidential elections, and undoubtedly, what it takes to win them. Donald Trump is said to have achieved one of the most improbable victories in modern U.S. history (Roberts et al, 2016), which prompted multiple reactions, from outrage to euphoria. However, despite the antagonistic beliefs driving these reactions, there was one underlying assumption; for better or worse, Donald Trump was not a politician. The following research does not intend to give an answer to 'why' Donald Trump won the presidential election, as such a response lies beyond the scope of this project, but to explore 'how', by analysing the ways he used language to legitimise his political stances. "Legitimisation deserves special attention in political discourse because it is from this speech event that political leaders justify their political agenda to maintain or alter the direction of a whole nation, and in the case of the U.S. leaders, the entire world." (Reyes, 2011: 783) Indeed, exploring the ways in which Donald Trump used political discourse in his campaign rally speeches could reveal the instrumental role of language in his pursuit of power and legitimisation. As Dallmayr (1984: 4, 2) claims, it is through language's "architectonic role" that politicians cast or grid for an entire way of life", thus creating an image of what is 'real' in public consciousness. Most modern linguistic theories have often treated language as an abstract system of linguistic forms, analysing it in a purely linguistic or logical manner. By attributing such an autonomy to language, its socialhistorical conditions of production, reproduction and use have been widely ignored. Moreover, while language has been recognised as an instrument of action and power, its political use in the setting of the 2016 U.S. presidential campaign has not been yet analysed. By combining text analysis through the software package IRaMuTeQ, and Bourdieu's (1991) conceptualisation of language, this study will examine Donald Trump's political discourse in his pursuit of power, aiming to bridge methodological and theoretical gaps present in the political discourse analysis literature. Specifically, it will (i) treat language as a multi-layered system and recognise the recurrently neglected interplay between the different levels of linguistics. Secondly, (ii), it will account for the overlooked pragmatics of language, through recognising the physical and socio-political space where the speeches were given. This will allow investigation of the ways in which Donald Trump might have shown an element of adaptability in his pursuit of power. Finally (iii), it will empirically evaluate the applicability of Pierre Bourdieu's work on Language and Symbolic Power (1991) to this particular scenario, with the goal of achieving a better interdisciplinary understanding of the relationship between Donald Trump's linguistic space and the political space, by unifying social theory with linguistic theory and political discourse. As Lemke claims, "it is through discourse formations that we construct the very objects of our reality" (2005: 24). However, as Bourdieu (1991) argues, being able to construct or shape a symbolic reality, and consequently, "reality" itself, involves creating 'discourse formations' that tap into the space of political position-takings. This, in turn, depends on possessing a specific competence to impose the social representation of the social reality capable of achieving the greatest support. Thus, the following study will analyse how language was deployed by Donald Trump as a means to win political capital, revealing how he might have possessed an intuition of the homologies between the political field, the space of political attitudes and, ultimately, the social space. 2. Literature Review "Language is never innocent. (Barthes, 1970: 16) The following review will explore the linguistic turn in political discourse analysis, which focuses on the ways in which linguistic choices in political language are manipulated for the achievement of specific goals. Indeed, defining political discourse is not straightforward as some analyses of political discourse adopt a definition that encompasses all discourse analyses. However, since the focus of my research is on how language is used in political speeches, this review will mainly explore linguistic analyses of political language used in 3

4 strictly political environments. Almost all levels of linguistics have been studied in political discourse, from lexis to pragmatics. However, there has been insufficient recognition of the interplay between these, as most studies often fail to analyse language as a multi-layered system. Furthermore, many political discourse analyses, regardless of their compelling insights into political language, have repeatedly lacked a theoretical framework through which to analyse and situate their findings. Following Fairclough's (1992) claim that textual analysis needs to be based on a multifunctional theory of language, this research will draw on Pierre Bourdieu s work on Language and Symbolic Power (1991). Various scholars of political discourse have highlighted examples of how specific words have been strategically deployed by politicians in their pursuit of political goals (Wodak, 1989; Geis, 1987; Bolinger, 1982). For example, many have drawn their attention to how politicians linguistically have subverted negative associations in the political discourse of nuclear weapons (Chilton, 1985; Cohn, 1987). In his analysis, Montgomery showed this obscurantist and euphemistic language of nuclear weapons by translating manipulated words into their more realistic definition, i.e- enhanced radiation weapon into neutron bomb which destroys people and not property (quoted in Tannen et al., 2015: 401). Moreover, there have been attempts to use statistical methods to analyse how single words are used. For example, Ashby used logistic regression to analyse variation in the use of the negative particle ne in French, and found that age, social class, and gender all affect negative-particle deletion in French, with the young, the lower middle class, and women being more likely to delete 'ne' (Ashby, in Grieve, 2012). Similarly, the system POLITICS has been recurrently applied in sociolinguistics to analyse inferences and defining the context relative to certain ideological formations in which interpretation takes place (e.g. conservative or liberal) (Wilson, 2003). However, although undoubtedly revealing, analyses of single words have failed to take into account some crucial levels of linguistics such as the structural dimension of syntax (Ibid.). Indeed, politicians legitimate what is going on in the world not only through their specific use of vocabulary or lexis, but also through the way their sentences are constructed as a whole. Following Halliday s (1985) system of 'transitivity' which provides a set of choices for describing events, Goodman (1996: 56) illustrated how the use of active and passive sentences can also obscure certain aspects of the story. Similar to Goodman s analysis, Kress and Hodge (1979) explored how grammatical categories may be operationalized in different ways and illustrate how politicians use what he terms negation, this is, terms such as but, nevertheless, yet, well, etc.., to articulate contrastive points. However, although the aforementioned studies do account for the impact that syntax can have in how a reality is represented, they fail to acknowledge the social and historical contexts of language production which can be crucial for the formation of meaning (Bourdieu, 1991). Filling in this existing gap, Dunmire (2007) explored linguistic transformations in a contextualised manner, analysing how language actively assisted the U.S. in positioning itself for military intervention in Iraq through analysing key security texts and speeches in the Bush administration. In her analysis, Dunmire recognised how the political discourse displayed in the 2002 U.S. National Security Strategy operated through the use of strategies of legitimisation, such as defuturizing the future by projecting images of an imminent threat to justify action in the present. According to Dunmire, the future constitutes an ideologically significant site in which dominant political actors and institutions can exert power and control (Ibid.: 19). Following Dunmire s analysis, Reyes (2011) developed and proposed five key strategies of legitimisation accounted in previous studies (i.e. Rojo and Van Dijk, 1997; Van Dijk, 2005; Van Leeuwen, 1996, 2007, 2008; Van Leeuwen and Wodak, 1999), and further showed how these underlined justifications of military presence in the 'War on Terror', through examples of speeches given by leaders with differing ideologies, specifically by Barack Obama and George W. Bush. Indeed, in the discourse genre of political speech, the choice of words deserves special attention as the generally planned nature of speeches makes language more reflective of what the politician wants others to perceive (Hermann, 2008: 10). Although Reyes' study on strategies of legitimisation undoubtedly acknowledged the semantic level of political language, i.e- the construction of meaning, the abstract nature of his analysis makes it unclear how these strategies might have been employed differently depending on the spatial-temporal conditions in which the speeches were uttered. According to Edelman (1985: 130), meanings are always a function of the context from which it issues, of the disparate needs and interests of the audiences involved, and of their respective modes of perception. Indeed, there are many levels of linguistics involved in political language and their intrinsic relation to underlying social structures needs to be fully recognised. 4

5 Bernstein (2003) conceptually resolved this gap by linking the social and symbolic systems through his key concept code. He argued that, regardless of register difference or the social activity taking place, habits of language use are highly dependent on social class position. He distinguished two major semantic coding orientations, the restricted and elaborated, used by the working class and middle class respectively. Although Bernstein accounted for crucial social aspects of language in society, his collapse of the British class system into a basic twofold division has been contested as being overly rigid, simplistic and stereotypical (Trudgill, 2000; Montgomery, 1992). Indeed, in the contemporary neo-liberal and post-industrial society, class differences can seem invisible. Contrary to Bernstein s class system, Bourdieu (1984) proposed an understanding of the social world as a multi-dimensional space where individuals are positioned in relatively autonomous fields, determined by the relative quantities of symbolic, economic and cultural capital they possess. By recognising the relative nature and underlying impact of the social space in lifestyles and practices, Bourdieu's work highlights how the system of linguistic oppositions can also be understood as the retranslation of a system of social differences (Bourdieu, 1991: 54). According to Bourdieu (1984), there is a homology between opposed modalities of political language and political judgement or ideology. Following this, in his book Language and Symbolic Power (1991), Bourdieu recognised how the political field is the site par excellence of symbolic power, claiming that political agents might use language strategically to maximise their symbolic efficacy: "The well informed politician is one who manages to master practically the objective meaning and social effect of his stances by virtue of having mastered the space of actual and especially potential stances or, better, of the principle underlying these stances, namely, the space of objective positions in the field and the dispositions of those who occupy them." (Bourdieu, 1991: 177) In other words, politicians with a better feel for the political game would possess a specific competence to impose and legitimise the social representation of reality capable of achieving the greatest support. Through examining the speech given in Béarnaise by the Mayor of his hometown, Bourdieu (1991: 68) uncovered what he termed "strategies of condescension". According to Bourdieu, these are linguistic strategies that occur when "agents occupying a higher position in one of the hierarchies of objective space symbolically" adopts the speech, style or dialect of those occupying a lower position, "thus ensuring they gain profits of recognition accorded to a purely symbolic negotiation of distance". Indeed, phonological variables correlate with class, gender and ethnicity (Wilson, 2003), understood as "fields" in Bourdieusian terms, and therefore, Donald Trump might have adjusted his phonological system for political effect. Indeed, as outlined by Heritage (1993: 214), "Bourdieu represents an important effort to rethink issues of language, symbolism, and legitimacy in the context of a flexible conceptualization of social class and its potentials". Therefore, in this study, text analysis of rally speeches during Donald Trump's U.S. presidential election campaign will be linked within Bourdieu s (1991) theory of the social space and symbolic space, as his conceptualisation of the social world offers the best theoretical framework through which to uncover potential socially positioned meaning-making linguistic habits in Donald Trump's political speeches. To account for the space of political 'position-takings' homologous to this multi-dimensional social space, the speeches will also be analysed in terms of the geographic location (State) in which they were given, thereby adding another linguistic dimension to this study. Bourdieu's work on Language and Symbolic Power (1991) has been criticised as being too 'macro-oriented' due to it lacking connections with micro-sociological concepts, which has posed issues for more concrete and focused empirical research (Thompson, 1991; Heritage, 1993). Thus, this analysis of political discourse aims to fill in this research gap present in Bourdieu's work on political language, by offering a systematic and empirical analysis of language in the specific social setting of Donald Trump's 2016 presidential campaign. Rather than approaching this abstractly, the following study will consider it in connection to the recent strategies of legitimisation developed by Reyes (2011), extending his analysis on language use to a different socio-political and cultural context. 5

6 3. Methodology The corpus used in this analysis is ad-hoc constructed. It contains 15 rally speeches given by Donald Trump during his 2016 presidential election campaign 1, formally launched on June 16th until November 8 th. Each speech was defined as a text unit, therefore the corpus of analysis comprises 15 text units (speeches). These text units were further divided into text segment (TS) or words context, which were automatically generated by the software and were on average three lines long. Given that each speech is a text, these were separated in the corpus of analysis through command lines (****) which include the identification number of the speech (E.g: speech_01), followed by important features (variables) of the speech for the research design. The 82,518 tokens of the complete corpus, these are - every individual occurrence of a linguistic unit in speech, were classified according to the State where the speech was given and its political party affiliation (variables: *state_, *poli_). The *poli_ variable was divided into three categories: 'Republican', 'Democratic' and 'swing state'. The two former refer to states that have been dominated by those political parties for over 15 years ('safe states'). Contrastingly, the 'swing state' category refers to those states that could have been reasonably won by either the Democratic or Republican presidential candidate. Table 1. Corpus composition; 15 Donald Trump speeches organised by the name of the state, political party dominating it and length of the speech. N o Political Party Speech Location Word Length 1 Republican Arizona 6,928 2 (*poli_rep) Texas 3,410 3 Mississippi 3,612 4 Montana 8,606 5 Republican National Convention 5,143 6 Democratic New York 4,253 7 (*poli_dem) Michigan 6,802 8 Washington 4,374 9 Wisconsin 3, Pennsylvania 5, Swing State Ohio 4, (*poli_swi) North Carolina 7, Florida 4, Nevada 4, Indiana 7,994 82, Analysis Software: IRaMuTeQ To support the data analysis of this research, the software IRaMuTeQ 2 was used to analyse Donald Trump s speeches. Being a free statistical software, IRaMuTeQ is gaining space in social and psychological research as it offers a range of useful tools to describe and analyse textual corpora and people/characters matrixes, including five types of analyses: classic text statistics; word clouds; descending hierarchical classification; analysis of similarities and analysis of specificities or correspondence. Furthermore, it can perform different types of visualisation on large text corpora (over hundreds of millions of occurrences), providing quality graphics output, effectively fitting with theoretical perspectives interested in communication and social construction of knowledge 3. Indeed, the bird s eye view of the corpus offered by IRaMuTeQ is crucial when accounting for significant words in a clear way. As Kami et al (2016: 2) highlight, the use of software is not a method of data analyses, but a tool to process them. Thus, the interpretation of the findings is essential, with the software being most effective when the text is contextualised by applying substantive knowledge of the subject matter (Lahlou, 2012). 3.3 Analysis Strategy In the following study, both simple and multivariate analyses were performed. The classic lexicographical analysis identified and formatted text units, turned texts into text segments (TS) and identified word frequencies and lexical forms. As in other forms of automatized analysis of large textual corpora, its use 1 Due to the novelty of the research area, the vast majority of the data had to be collected manually, transcribing the rally speeches via YouTube videos. All the speeches analysed were given publicly and subsequently retransmitted through public means of communication, hence the following research does not expose any ethical considerations. 2 IRaMuTeQ stands for R Interface for Multidimensional analysis of Texts and Questionnaire ( and is a visual interface based on the R statistical software and on Python language. 3 IRaMuTeQ has been frequently used in association with other methods to examine various typologies of textual corpora, from twitter exchanges during mass emergencies to political settlement speeches (Sarrica et al, 2014; Pola et al, 2015). 6

7 involves pre-treatment and modification of the original text in order to reduce complexity. IRaMuTeQ does lemmatization (grouping together the inflected forms of a word so they can be analysed as a single item) and therefore the analysis is subject to a degree of language dependence. For example, the term believing was dealt with as believe. The terms with a space in their name (Eg Democratic Party or Hillary Clinton ) were replaced in the body by democratic_party and hillary_clinton respectively, for the software not to understand two distinct words. Furthermore, the software cannot recognise synonyms, phraseology or alternation between an individual's name and the common noun. To be able to account for this to an extent, the personal pronoun 'she' was treated as a noun, as it was only used throughout the speeches solely to refer to Hillary Clinton. Other pronouns such as 'me',' them', 'their' and 'your' were also analysed to try to uncover how Donald Trump's language might contribute to shaping the audiences' identities through establishing and maintaining 'us versus them' boundaries (Gregory, 2004: 251). Analysis of similarities: An analysis of similarities was performed to identify the co-occurrence or connectedness present between the words (Ratinaud & Marchand, 2012). Based on the theory of graphs, such an analysis helps to identify the social representations structure of the text corpus content. Hence, an analysis of similarities (index: co-occurrence; layout: fuchterman reingol; maximum tree; size of text: 10) was conducted to identify the structure, central core and peripheral system of Donald Trump's discourse present throughout his speeches during his campaign. Three separate analyses of similarities were performed to the corpus grouped according to the political party variable ('poli_'). This enabled not only to account for the various recurring strategies of legitimisation 4 employed by Trump through the visual connectivity between the words, but also to compare whether the presence or nature of these strategies differed depending on whether the state where the speech was given was Democratic, Republican or a swing state. Method of Descending Hierarchical Analysis (DHA): In addition to the analysis of similarities, a DHA was performed. The DHA method aims at identifying clusters with similar vocabulary within text segments, but different from other segments. Similar to the analysis of similarities, the classification obtained is based on lexical proximity and therefore it is not based on counting frequencies, but on highlighting relations among words. The implementation of the algorithm of the DHA is probably the feature that has received the most interest. It was initially developed by Max Reinert (1987), a student of the founder of the 'French Data Analysis' school, Jean-Paul Benzécri, and first implemented in the software Alceste (Ratinaud & Marchand, 2012). The purpose of this analysis is to find and measure what Reinert (1987) terms lexical worlds in the speaker s discourse. we assume that the speaker, during his speech, is investing successive different worlds and these worlds, by imposing their properties, thereby impose a specific vocabulary. Therefore, the statistical study of the distribution of this vocabulary should be able to trace these mental rooms that the speaker has successively inhabited; traces perceptible in terms of lexical worlds " (Reinert, quoted in Schonhardt-Bailey, 2013: 1) In other words, a lexical world is a specific vocabulary, which inherits its properties from the speaker's discourse. Following this, Schonhardt-Bailey (2013: 2) recognises how, as politicians also run through semantic fields when producing discourse, their statements provide us with a lexical distribution that reflects the content of these fields (their lexical worlds). Indeed, classifying together the statements that contained similar words present in Donald Trump s speeches allowed for the visualisation of the semantic territories that were behind the construction of his political discourse. Hence, this method allowed the quantification of Donald Trump speeches by extracting the most significant, underlying semantic structures (Rosa et al, 2013: 32), so that the differences in vocabulary refer to the different social representations he used to legitimise his political discourse. Specifically, the DHA decomposes the corpus into text segments which are then clustered (in repeated tests of chi-square type), according to their vocabulary usage, resulting in a dendrogram which displays the hierarchy of classes/clusters (or worlds ) present in the corpus. The way these clusters are interpreted depends on the theoretical scope of the research 5. In my own analysis strongly 4 As the focus of the analysis is on whether Donald Trump's language showed an element of adaptability, the strategies that did not present variability between states were not included in the analysis section, such as legitimisation through 'rationality' and 'altruism'. 5 As aforementioned, Reinert considered each cluster or lexical universe as a world. Research in linguistic rather considers these clusters as semantic contexts, whereas for research in social psychology, especially when interested in studying the common sense knowledge, these clusters have been interpreted as social representations or images about a certain object (Veloz et al, 1999). 7

8 driven by Bourdieu s theory of the social space and symbolic space, each cluster identified by the DHA was interpreted as a lexical territory in Donald Trump's linguistic space where differing strategies of legitimisation were employed by the candidate as a means to win political capital. According to Bourdieu (1991: 192), "political capital is a form of symbolic capital". The DHA provides another way of presenting data arising from the correspondence factor analysis, a technique Bourdieu also discovered from the 'French Data Analysis' led by Jean-Paul Benzécri. Bourdieu (1984) used this technique to build the space of social positions and the space of cultural tastes/practices, ultimately tracing a homology between these two spaces. Indeed, the DHA offers the best methodological tool through which to analyse Bourdieu's conceptualization of 'symbolic power', as the social representations arising from the clusters reflect Donald Trump's power to elaborate and diffuse a "legitimate principle of division of the social world" (Bourdieu, 1991: 181). Furthermore, through accounting for the political party dominating the states where the speeches where given in both the analyses of similarities and the DHA, this approach enables the visualisation of whether the 'principles of di-vision' arising in Trump's linguistic space were homologous to the space of political 'position-takings' constituting the political field, mainly composed by the Democratic and Republican political parties. Moreover, using the text analysis software IRaMuTeQ will allow for most levels of linguistics to be recognised, from the lexical level (through words frequencies and modalities) to the syntactic level (through an analysis of similarities visualising connectivity between words) and the pragmatic level (through performing a DHA accounting for the contextual variable of physical location). 4. Results and Discussion 4.1 Analysis of similarities With the analysis of similarities we are able to visualise whether Donald Trump used words strategically to define and describe key elements of his linguistic space, by constructing and portraying a specific reality to legitimise his political stances. The output of the analysis of similarities applied to all 15 Donald Trump's speeches is shown in Figure 1. Furthermore, three separate analyses of similarities were performed to the speech corpus grouped according to the political party variable ('poli_'). The three figures below were analysed and compared to identify whether Trump's use of strategies of legitimisation differed depending on whether the state where the rally speeches were given were swing, Republican or Democratic states. The outputs of the analyses are shown in Figures 2, 3 and 4 respectively. 8

9 Figure 1: Analysis of Similarities of the Complete Corpus The analysis of similarities output (Figure 1) shows a central core composed by three main organisational axes in Trump's discourse, which are 'people', 'country' and 'Hillary Clinton'. The thickness of the edges is proportional to the similarity index, while the size of the words reflect their frequency. In the periphery outstands 'job' directly linked to 'country' and in the top-right area of the periphery, we can find words strongly associated to 'Hillary Clinton' such as 'she' and 'has'. Around these main components, we can find word cloud groups, which allow for the identification of the linguistic strategies Trump employs to legitimise his political stance regarding the main organisational elements or axes of his discourse, through connecting them to certain verbs, adjectives and nouns 9

10 Figures 2, 3 and 4: Separate Analyses of Similarities ('poli_') At a first glance, the structure of Trump's discourse appears visually different throughout the three outputs. Figure 4 (swing states) shows a central core composed by only one organisational axis: people, Figure 2 (Republican states) is organised by two axes: country and people, and finally, Figure 3 (Democratic states) shows a central core composed by three main organisational axes: 'people', 'country' and 'Hillary Clinton', hence paralleling Trump's overall discourse structure shown in Figure 1. Furthermore, it is also noticeable that the edges connecting words in the Republican and swing states outputs are thicker compared to the ones in Democratic states, showing a higher similarities index present between words in Republican and 10

11 swing states. This strong connectivity between words suggests a use of repetition in Trump's discourse throughout the aforementioned states. As Van Dijk notes, "repetition operations are one of the major strategies to draw attention to preferred meanings and to enhance construction of such meanings in mental models" (1997: 35). For example, the strong connectivity present between the words 'make', 'America' and 'great' highlights the ongoing repetition of his campaign slogan "make America great again". Arguably, through its repetition, Trump naturalized his political stance that America was not great at that moment and would get worse in the future if he was not elected. A) Legitimisation through Emotions Through comparing Trump's main organisational axes and peripheral system between the analyses of similarities outputs, it is noticeable how the way he appeals to emotions varies throughout states. The appeal to emotions is a widely identified strategy of legitimisation in political discourse analysis. According to Reyes (2011: 788), "emotions have the potential of distorting [ ] reality". By employing words with negative connotations, a series of emotions are evoked in the listener s mind, which are linked emotionally with previous experiences. Thus, by attributing negative representations of social actors or political matters, politicians aim to skew the opinion of their audience. In Trump's Democratic speeches, we can clearly identify how this strategy is mainly employed through the high frequency of one of its main axis 'Hillary Clinton' (freq.= 342, compared to Republican freq.=182 and swing freq.= 228). The word cloud surrounding Hillary Clinton in Democratic states shows how she is strongly associated to negative adjectives such as 'fail', 'corrupt' and 'horrible', and nouns such as 'death', 'war' and 'ISIS'. Interestingly, in Republican states, Hillary Clinton appears less frequently and is associated to 'Barack Obama', a name that does not appear in Democratic states, indicating Trump's awareness of the "space of objective positions in the field and the dispositions of those who occupy them" (Bourdieu, 1991: 177). Furthermore, in Hillary Clinton's word cloud we can identify various verbs associated to another common strategy displayed in political discourse generally exploiting the emotion of fear: the hypothetical future. Politicians often display the present time as a period that requires imminent action. As Reyes (2011: 793) effectively summarises, "these actions are related to a cause (which occurred in the past) and a consequence (which may occur in the future)." This strategy is reflected in Trump's use of past tense verbs such as 'was' and 'has' associated to Hillary Clinton's past 'avoidable' mistakes ('crime', 'lie' and ' ') and his use of the modal verbs 'could' and 'would', which often appear in conditional sentences, to refer to the 'unavoidable' future if Hillary Clinton is elected president. Speech_06, poli_dem [New York]: "The Hillary Clinton foreign policy has cost America thousands of lives and trillions and trillions of dollars and unleashed ISIS across the world." Speech_07, poli_dem [Michigan]: "She is unfit and unqualified to be the President of the United States, and her election would mire our government and our country in a constitutional crisis that we cannot afford." According to Van Leeuwen and Wodak (1991: 92), politicians also appeal to emotions by using the pronouns we and they, which allow for the construction of 'us versus them' boundaries. When analysing the Republican analysis of similarities output, we can identify how Trump constructed a "we" group and a "they" group, as the pronouns "your" and "me" are linked to the main axis 'people' (generally referring to the audience/american citizens), whereas the pronoun "them" is strongly connected to "criminal" and "alien". Indeed, Trump's appeal to emotions in Republican states is visible through the prominence of a peripheral axis about immigration (with words 'immigration' and 'immigrant' having a freq. =95, compared to Democratic freq. = 21 and swing freq. =15). Speech_02, poli_rep [Texas]: "Hillary Clinton also wants a 550 percent increase, over Barack Obama who is letting them come in by the thousands, in Syrian refugees. We have no idea who they are, where they come from, no paperwork - the great Trojan horse, it's going to be trouble folks. There have been hundreds of immigrants and their children charged with terrorism inside the United States since 09/11 and a number of terrible attacks like the Boston bombing, San Bernardino and Orlando, yet Hillary Clinton wants to fling open the flood gates to our borders." 11

12 Indeed, the most emotionally exploited event has been the attacks of 11 September 2001, which Trump also references as shown in the quote above. Furthermore, he mentions recent terrorist attacks such as the "Boston bombing" and the shootings occurred in "San Bernardino and Orlando." Apart from exploiting emotive reactions through his reference to these terrorist attacks, employing such connotative terms also works towards the spatial proximization effect. The concept of spatial proximization was developed by Chilton (2004), who defines it as a rhetorical device whereby the speaker describes events as physically close to and thus consequential for the addressee. Through analysing the American interventionist rhetoric, Cap (2008: 29) argues that proximization was enacted "by maintaining the stance of continual reference to the 2001 attacks on the WTC and the Pentagon, as a means for building up a future oriented cause-and-effect analogy". Furthermore, as seen in the quote above, Trump appeals to emotions to legitimise his political stance against refugees entering the U.S., by constructing them as future 'immigrants charged with terrorism'. Trump's metaphorical use of the 'Trojan Horse' inference to describe the situation of refugees getting asylum also acts as means through which proximization is achieved, as metaphorical language "bears a close resemblance to the assertion-implicature framework" (Cap, 2008: 30). Moreover, in the quote above, it is also noticeable how Trump also constructs a "we" and "they" group in the construction of refugees through the pronouns 'them' and they', with refugees (constructed as future terrorists) being classified as the unknowable 'other'. The lesser emphasis given to the discourse around immigration in Democratic and swing states compared to Republican ones is exposed by the absence of the word 'Immigration' in the swing states analysis of similarities output, and the words 'build' and 'wall' in the Democratic output. Moreover, when analysing the word cloud surrounding the consistent organisational axis 'people', we also find that the word 'media' does not appear in the Democratic states analysis of similarities output. Frequencies of the word 'media' were requested showing that it appears more than three times more in Republican states (freq.= 14) and swing states (freq.=16), compared to Democratic states (freq.=4). Trump's anti-media and anti-press discourse reflects the rise of what has been widely termed 'post-truth' politics. As defined by the Oxford dictionary, post-truth is an adjective relating to or denoting circumstances in which objective facts are less influential in shaping public opinion than appeals to emotion and personal belief (Flood, 2016). Indeed, the idea that journalism is meant to safeguard truth along with the notion of a nonpartisan, balanced and fair media (McIntyre, 2015) is repeatedly contested in Trump's political discourse. Speech_01, poli_rep [Arizona]: "We are going to declare our independence from politicians who do not listen to us and media executives that do not care about us and by the way who never report our massive and very, very friendly crowds, that is for sure." B) Legitimisation through Condescension vs. Voices of Expertise Furthermore, Bourdieu (1991) recognised how political agents might use language strategically to maximise their symbolic efficacy, by adapting their language to the demands of the social field of their audience. He terms this strategies of condescension, defining them as those symbolic transgressions of limits [of language] which provide, at one and the same time, the benefits that result from conformity to a social definition and the benefits that result from transgression (Bourdieu, 1991: 124). In other words, a strategy of condescension occurs when someone at the top of a social hierarchy adopts the speech, style or dialect of those at the bottom (Haney, 2008). In the specific field of politics, Bourdieu argues that strategies of condescension are deployed as an exercise of political capital to pursue legitimisation as "the capacity to manipulate is greater the more capital one possesses" (Bourdieu, 1991: 71). Arguably, Trump's use and abuse of the interjection 'you know', treated as one word in the analysis: 'you_know' (freq. = 220), which might appear totally meaningless in our day to day lives, could have a legitimising effect in Trump's discourse. By not speaking in the politically correct language that dominates the political field, he can strengthen his outsider position and portray himself as parallel to the social field of the audience. Furthermore, his use of 'you know' might also reflect what Althusser (1971) terms interpellation, this is, a process in which individuals are 'hailed' into subject positions. Arguably, when Donald Trump addresses the audience and claims they already know the political stance he is presenting, he encourages the naturalisation, internalisation and inevitable acceptance of his ideas. 12

13 The separate analyses of similarities highlight how the interjection 'you know' is connected to different axes depending on the state's political party. In Republican and swing states, the terms appear to be most present and connected to the main organizational axis 'people' (generally used to refer to the audience/american citizens), suggesting that Trump might be deploying the interjection in a 'condescending' way. However, in the Democratic analysis of similarities, 'you know' appears connected to the main organisational axis 'Hillary Clinton', which highlights how, in this case, Trump uses the terms in an 'interpellative' way, i.e.- as a way to reinforce and naturalise his stances about Hillary Clinton. Speech_04, poli_rep [Montana]: "I mean China and I do a lot of business with China, they are great, you know." Speech_07, poli_dem [Michigan]: "As you know, the FBI has reopened it is investigation into Hillary Clinton and discovered, and discovered, you heard this yesterday, it is hard to believe, and discovered another 650,000 s." As Bourdieu (1991: 167) claims, relations of communication are always power relations which depend on the symbolic power accumulated by the agents. Indeed, symbolic power is exerted through words when there is a belief in the legitimacy of the person uttering them. Reyes (2011) contends this type of authorization is generally evoked by the institutional position represented by the speaker. In Trump's linguistic space, the use of the words 'know' and 'believe me', which are linked to the main axis 'people', highlights how Trump often grounds his opinions and stances in his institutional position, by exploiting his status as a 'successful' businessman continuously evoked by the use of his own name (also linked to the 'people' axis), and by portraying himself as an insider-turned-outsider. Through analysing the separate analyses of similarities, it can be noted that the edge connecting 'believe' and 'me' is noticeably thicker in Republican states, suggesting a stronger connectivity between the words and therefore a higher presence of this strategy. The frequency of the terms 'Donald Trump' and 'I' were also requested, finding that these terms appear more often in Swing States (freq.= 729), followed by Republican states (freq.= 558) compared to Democratic states (freq.= 412). The legitimisation strategy through voices of expertise refers to the voices or data that the political actor brings into the speech to support or strengthen his stances. This is noted in Trump's discourse through his use of percentages, highlighted by the word 'percent'. The way he uses this strategy also varies depending on the state's political party affiliation, with it appearing connected to different discourses. In Democratic states, 'percent' is connected to the discourse about 'Hillary Clinton', showing how he generally employed data to strengthen his stances about his opponent. However, in Republican states the term is connected to the main axis 'people'. The Republican speech corpus was reviewed to further understand how this strategy was employed, finding that in the majority of the cases, the term 'percent' was uttered to back up his stances on increasing homicides and crime rates. Speech_02, poli_rep [Texas]: "Homicide is up nearly 50 percent in Washington DC, and up more than 60 percent in Baltimore, and up throughout all of our inner-cities. Violent crime grew 16 percent in the 50 largest United States cities." This specific use of statistics in Republican states concords with the presence of the peripheral axis 'law' connected to the main axis 'country'. In swing states, the term 'percent' is associated to the words 'tax' and 'business', suggesting that in these states, Trump mostly used percentages to back up his stances towards the economy, which is also suggested by the presence of the peripheral axis 'job', connected to the main axis 'people'. Bourdieu notes that a dominant actor who employs strategies of condescension must do so without appearing to be ignorant or incapable of satisfying [the audiences] demands (Bourdieu, 1991: 69). Arguably, Trump's consistent use of data throughout his speeches not only strengthened his political stances, but also allows him to maintain a balance between the outsider position he portrays through his use of condescension strategies, and the powerful insider position he constructs through claiming he is informed about what the audiences' needs are, and is capable of satisfying their demands. 13

14 4.2 Descending Hierarchical Analysis As already detected through the analysis of similarities of the whole corpus, the most repeated nouns in Trump's discourse were "people" and "country", employed 615 and 489 times respectively. The verb most used appears to be "was", uttered 252 times, followed by "make" (310) and "do" (307). The adjectives with most frequent occurrences were "great" (321) and "bad" (159). The most common nouns, verbs and adjectives were classified through a DHA and four clusters or classes of words were obtained. The Descending Hierarchical Classification dendrogram shown in Figure 5 illustrates a partial composition of each of these clusters, elucidating the themes that Trump most appealed to in his political discourse. As shown through the dendrogram, the corpus was divided into two subgroups. These subgroups were further divided into two clusters, showing how cluster 1 has a greater connection or proximity with cluster 2, with the same being true for the other branch, which includes clusters 3 and 4. Figure 5. Dendrogram of Top-Down Hierarchical Classification of the Speech Corpus As shown in the figure above, cluster 3 has the greatest representation, constituting 36.1% of Trump's political discourse, followed by cluster 1 (24.7%), cluster 4 (21.8%) and finally cluster 2 (17.4%). The DHA assesses the degree of statistical significance of the words that occur most often in each cluster. The significance threshold of the chi-square test was set at Table 2 presents the results of the chi-square test for the 17 most significant words present in each cluster (all having obtained a p-value below 0.001), meaning that there is a very low probability that the correspondence between the words and their clusters would have appeared at random. 14

15 Table 2. Descendent Hierarchical Classification of Donald Trump's Political Discourse CLUSTER 1: Immigration CLUSTER 2: Social Welfare CLUSTER 3: Donald Trump CLUSTER 4: Economy Forms Chi 2 % Forms Chi 2 % Forms Chi 2 % Forms Chi 2 % Illegal ** 91% Poverty ** 94% Say 54.92** 65% Deal ** 83% Immigrant ** 89% Education ** 87% Guy 48.43** 86% Dollar ** 71% Law 98.64** 81% Child ** 65% Do 46.88** 60% Trade ** 75% Criminal 75.60** 86% Future ** 73% Was 43.65** 59% Tax ** 76% Immigration 75.44** 72% School 94.42** 78% People 38.70** 50% Billion 86.53** 86% Order 65.55** 92% Failure 55.97** 64% Know 33.34** 62% Company 83.24** 75% Terrorism 57.60** 86% Fight 54.63** 65% Me 33.30** 57% China 75.28** 81% Administration 51.06** 95% Replace 46.41** 78% Right 30.87** 55% Business 70.70** 71% Open (border) 45.06** 74% Democrat 46.04** 86% Smart 29.25** 91% Move 68.54** 76% Death 44.80** 94% Change 44.19** 61% Story 27.19** 85% Money 56.37** 63% Visa 39.59** 85% Live 43.77** 53% Tell 26.08** 60% Deficit 41.95** 93% Radical 39.12** 77% Job 41.32** 33% Could 24.47** 68% Sell 40.47** 77% Police 36.33** 65% Reject 38.25** 100% Great 23.57** 53% Small 40.12** 67% Release 36.30** 78% Produce 36.69** 83% Thing 22.58** 58% Factory 37.49** 87% Enforce 35.47** 93% Hurt 36.15** 67% Happen 22.39** 54% Trillion 36.37** 67% Alien 33.92** 100% Safety 34.22** 54% What 20.47** 68% Negotiate 33.93** 86% Violence 33.55** 80% Parent 33.44** 100% Look 19.52** 58% Job 33.34** 37% Note. This descendent classification provides the most representative vocabulary (list of forms) associated to each cluster. The chi-squared value assesses the significance of the association between a word and the vocabulary used in a given cluster (all chi-squared values above correspond to a < chance of random association). Usually, significance is linked to the presence of the vocabulary in a given cluster. However, rarely a form often associated with a given vocabulary may be associated to the cluster it is less frequently used in, despite appearing more often in the rest of the abstracts. To account for this phenomenon, the percentage of a form's presence in a given cluster is provided. For instance, the most representative form in the first cluster is 'illegal' as 91% of the occurrences of this form appear in the first cluster. 15

16 As outlined in Table 2, the first linguistic cluster includes essentially words associated to immigration. The most central word in this cluster is "illegal", followed by "immigrant", "law" and "criminal". As suggested by Inglehart & Norris (2016: 2), rising economic insecurity and social deprivation among the left-behinds is believed to have made immigrants prone to the "anti-establishment, nativist, and xenophobic scaremongering exploited of populist movements [ ], blaming Them for stripping prosperity, job opportunities, and public services from Us ". Moreover, as stated by Roth (2016), this rising insecurity and social deprivation has allowed certain politicians to gain power by "portraying rights as protecting only the terrorist suspect or the asylum seeker at the expense of the safety, economic welfare, and cultural preferences of the presumed majority." Again, following the analyses of similarities, we can surmise that Trump's discourse on immigration reflects this construction of an 'us' versus 'them' divide. Additionally, Roth's aforementioned statement on populists using anti-immigration discourses to defend the majority's safety and welfare might explain the apparent connection of the Immigration cluster with the second lexical cluster (Figure 5), as this second cluster includes words associated to social welfare (with the most central word being "poverty", followed by "education", "child" and "future"). The fourth lexical cluster refers to words relating to the economy and the market, with the most central word being "deal", followed by "dollar", "trade" and "tax". According to Zakaria (2016), supporters of Trump often point to economics as the key to their success. Indeed, the most widely-held view of mass support for populism emphasizes the consequences for electoral behaviour as arising from profound changes in the workforce in post-industrial economies, such as the rise of the cultural economy, automation and the collapse of the manufacturing industry (Inglehart & Norris, 2016: 2). The resulting cauldron of discontent has fuelled a popular resentment of the political classes. Therefore, Trump's discourse on business, linked to the third cluster where we can find words such as "say", "do", "know", "me", "smart", "great" and "happen", might reflect Trump's self-portrayal as a successful businessman, instead of a politician, who will restore the U.S. economy if elected president, hence echoing the strategy of 'authorization' arising in the analysis of similarities. 16