GOD, NATION AND FAMILY IN THE IMPEACHMENT VOTES OF BRAZIL'S FORMER PRESIDENT DILMA ROUSSEFF: A CORPUS-BASED APPROACH TO DISCOURSE

Brazil’s Lower House of Congress voted the impeachment of President Dilma Rousseff on 17 April 2016. In addition to voting, 511 out of 513 Deputies decided to use their time at the microphone to justify their choices. Internet users and online newspapers soon began commenting on the vocabulary used by legislators, in most cases associating words related to God, family and nation to the pro-impeachment speeches. In order to verify whether the lexical choice of the proand counter impeachment Deputies really differed, I combined corpus linguistics and discourse analysis to examine the transcripts of those talks by investigating the keywords used by the representatives of the two largest modes of vote — yes and no. A combination of effect size and statistical significance measures were applied for the identification of keyness. The recurring keywords showed that, contrary to reports, the lexical choices of the largest two groups of voters statistically coincide in that they both try to legitimate their choices through appeal to religion, altruism, and shared responsibility, although the manual analysis showed that those words were sometimes used with different meanings by each group. A brief account of the keywords exclusive to each mode of voting is also provided. They reveal the Deputies’ obvious opposite sides towards the result of the process: pro-impeachment Deputies appealed to patriotism and promised a better future should the impeachment be approved, whereas their opponents reminded their interlocutors of the then President’s social programs. The combination of techniques used allowed a more finely-grained analysis of the Deputies’ speeches than one based solely on word count.


Introduction
On 2016-04-17, 511 out of the 513 Brazilian Deputies voted the impeachment of the leftleaning Workers' Party (PT) Dilma Rousseff, the 36th President of Brazil. Rousseff was accused of breaking fiscal laws in her management of the federal budget and was definitively ousted from office by the Senate. The petition for impeachment was accepted by Eduardo Cunha of the Brazilian Democratic Movement Party (PMDB, now MDB), who was the President of the Brazilian Chamber of Deputies at the time and a political opponent of Rousseff. With 367 votes for the impeachment drive, the process then moved to the Senate, and on 2016-05-12 this institution decided to accept the charges and begin the trial, resulting in Rousseff's suspension and subsequent impeachment. With the Senate's decision, power was assumed by the then Vice-president Michel Temer, also of PMDB, a party which had played a crucial role in Rousseff's governing coalition, but voted to break with PT, increasing the chances of impeachment.
Besides voting for, against, or simply abstaining from sending the case to the Senate, Brazil's Lower House of Congress's representatives were allowed to use the microphone for up to ten seconds to justify their decisions. And most of them exercised this right, frequently introducing their speeches with pela minha família ('for my family'), por Deus ('for God'), pelo meu país ('for my country'). 1 The more than five-hour voting session was not even over when Internet users started posting memes about the content of the Deputies' speeches. The repeated use of the word Deus, as well as words related to family members, the nation, and the fight against corruption triggered countless memes posted on social networks, which were predominantly critical towards the declarations in favor of the process.
Political scientists and journalists also discussed the frequency of the Deputies' linguistic choices in newspaper headlines, such as 'What do declarations by Deputies favorable to impeachment reveal [about] our democracy?' (Aiuá, 2016), 'Deputies cited "God" 59 times while voting impeachment drive' (Agência Brasil, 2016), 'My, family…: a list of the most cited words in impeachment vote' (Andrade, 2016). Commentators also called special attention to the use of the so-called triad (God, family and nation) generally associating its use with the pro-impeachment votes. This motivated me to carry out a more thorough investigation of the linguistic choices of the Deputies favorable and unfavorable to the process, which is not supported solely by the frequency of the words pronounced. With this analysis, I intend to examine whether the pro-impeachment speeches contained different lexical choices (especially ones related to God, nation and family) to the counter-impeachment ones. By employing the methodology underlying corpus-assisted discourse studies (Partington, 2004), I examined the transcripts of the speeches, made available by the Chamber of Deputies a few days after the voting, in order to answer the following questions: (i) Are the lexical choices of the two major groups of voters statistically different? (ii) If so, to what degree does this choice differ?
Section 2 gives a brief overview of corpus-based discourse studies, as well as an account of how political strategies are revealed in the politicians' speeches. Section 3 describes the methodology used, as well as the quantitative results obtained. Section 4 provides a qualitative investigation of the data retrieved by computational tools, with an analysis of the keywords which recurred in both groups' speeches, followed by a concise account of the keywords particular to each mode of voting. The paper concludes with a summary of the main findings, calling special attention to the importance of combining quantitative and qualitative techniques instead of drawing conclusions based solely on word frequency.

A corpus-assisted discourse analysis
Although the combination of corpus linguistics (CL) and discourse analysis is neither new nor rare, Sanderson (2008, p. 59) explains that these areas have traditionally demonstrated reciprocal reservations: '[i]n the past, corpus linguists have not been particularly interested in discourse, preferring to concentrate on lexical and morphosyntactic analysis. Similarly, discourse analysts have seldom worked with corpora, preferring methods such as introspection, elicitation and the unsystematic collection of anecdotal evidence'. On the one hand, criticisms of critical discourse analysis focus on the practice of investigating a small number of texts, which may reflect the analyst's preconceived ideas (Baker, 2012), a practice usually referred to as 'cherry picking'; on the other hand, CL is accused of focusing solely on numbers, disregarding a deep manual analysis of data. Therefore, a combination of quantitative analyses, enabled by a corpus-based methodology, with a detailed manual investigation of the data may well prove beneficial: discourse studies can rely on semi-automatically retrieved data from whole texts as the starting point, hence decreasing the degree of subjectivity. There has been a growing body of literature that recognizes the importance of this combination, especially over the last two decades (e.g., Partington, 2004;Partington, Morley & Haarman, 2004;Baker, 2006Baker, , 2012Baker et al., 2008). Citing Biber, Conrad & Reppen (1994), Partington (2003, p. 6) suggests that empirical analysis of texts combined with semiautomatic tools enables the retrieval of patterns of language which occur naturally, and emphasizes that 'examples of authentic data can serve to support the researcher's argument or, perhaps even more importantly, as counter-evidence to make them think again'.
In accordance with Baker & McEnery (2005), I believe that semiautomatic analyses of corpora can play an important role in discourse analysis, as they lead researchers to identify patterns in authentic texts with greater objectivity, besides helping them to emphasize patterns of association (collocations), which, in general, surpass the interpretative ability that results from the close reading of a small number of texts. In this study, I employed CL techniques to highlight linguistic patterns in the Deputies' speeches, in order to verify whether -and how -they differed between pro-and counter impeachment Deputies at the Lower House voting.

Political strategies put into words
Political discourse has been an important theme for research due to its central role in organizing and managing society, with the aim to control people's minds (Dylgjeri, 2017). Rephrasing Machiavelli's impression of a leader's characteristics, as stated in The Prince (Machiavelli, 1532(Machiavelli, /2018, Block concludes that 'honesty and loyalty, though noble traits, are not the most expeditious route to getting things done; astuteness, and the ability to deceive […] are ' (2018, p. 71). Here understood as any public, institutional and private talk on politics (Kampf, 2015), political discourse is one of the means politicians use to achieve and maintain hegemonic power (Reys, 2011).
Language and politics are strongly intertwined. According to Romagnuolo (2009, p. 1), 'language is necessary to any form of social activity, but politics is arguably the one that relies on language more than most to accomplish its goals'. After all, politicians choose lexical items not only because of the official decorum of their position, but also to manipulate the listeners. From the analysis of interviews with politicians, in addition to various political speeches, Chilton (2004) concludes that, in order to imbue veracity on their utterances, politicians make use of evidence as a means to legitimize their discourse: 'what matters, from a political point of view, is whether the speaker has "credibility"' (Chilton 2004, p. 32). And, from a political point of view, making a positive representation of the self is equally important as negatively representing the opponents. Thus, political discourse can be understood as a continuous (de)legitimization process.
Mentioning several of Donald Trump's tweets as examples, Block (2018) explains how 'alternative facts' are used by politicians to spread accusations towards their opponents, making use of hyperbole to claim truths without presenting any proof to support their allegations. Paraphrasing 1984 (Orwell, 1949), Block refers to political language as 'doublespeak', as it is used to tell lies and mislead the interlocutors: 'it is a language which makes the bad seem good, the negative appear positive, the unpleasant appear attractive' (2018, p. 83).
Several linguistic strategies are used by politicians in order to reach out to the interlocutor and convince them of the veracity of their statements. But in order to justify their actions, the politicians' speeches must be legitimized. Reys (2011, p. 782) defines legitimization as 'the process by which speakers accredit or license a type of social behavior'. Through legitimization the speaker provides arguments for his or her attitudes, seeking support and approval by presenting the proposals as the appropriate ones. Legitimization involves assumptions about the will of voters, ideological principles in general, charismatic attitudes and positive self-representation. Conversely, the portrait of the opponents through negative characteristics, accusations and offenses is characterized as strategies of delegitimization (Chilton, 2004).
The way the tactics of (de)legitimization are expressed by the Deputies' speeches while voting Dilma Rousseff's impeachment trial will be discussed in the following section, along with examples extracted from the study corpus, the construction and processing of which is detailed below.

Methodology
As was mentioned above, this paper investigates whether words related to religion, family and patriotism were more heavily invoked by pro-impeachment Deputies during the impeachment session, and, if so, to what extent their contexts of use coincide. In order to answer the research questions, I combined quantitative and qualitative analyses of the study corpus, as detailed below. Nevertheless, it seems important to set the scene of the political situation which took over Brazil at the time the impeachment was voted by the Lower House, as it informed the data collection.

Brazilian political scenario before Dilma Rousseff's impeachment
Under the current Constitution, Brazil has held direct presidential elections since 1989, when Fernando Collor was elected. In 1992, two years before completing his five-year mandate, the neoliberal Collor resigned in an attempt to evade his impeachment trial. Twenty-four years later, another Brazilian President underwent a similar process. Although the impeachment is provided for in the Brazilian Constitution as a legal means of dismissing politicians from their duties due to misconduct, this time left-wing sympathizers considered the process a coup initiated by Rousseff's opponents and endorsed by the media. Globo, the leading media group in Latin America and the only Brazilian conglomerate to rank among the 30 largest media owners of 2017, which was already criticized for having admittedly supported the military coup of 1964, received most accusations.
After having critically analyzed 16 headlines and 18 of the daily published editorials of the conglomerate's newspaper O Globo released from March to April 2016 about Dilma, her predecessor Luiz Inácio Lula da Silva, and the PT, van Dijk (2017, p. 200) concluded that 'this impeachment was the result of massive manipulation by Globo, the largest media corporation in the country and the voice of the conservative middle class'. He claims that the process was mostly driven by the hatred of former-president Lula, as the conservative (white) elites, under the influence of O Globo, never surpassed the fact that a lowerclass metallurgical worker won the 2002 presidential election, defeating the PSDB, the party that would also lose the following three elections to PT. As well as PT and supporters, van Dijk (2017) believes that Rousseff's impeachment trial was in fact a maneuver initiated by defeated Aécio Neves -and supported by Globo -, who had lost the election to Dilma Rousseff again in 2014. Actually, we should note that the PSDB candidate in 2010 was José Serra.
The conspiracy theory involving Globo, the unsatisfied elite and the defeated party and its supporters took over counter-impeachment media and population. Meanwhile, the population showed its dissatisfaction -either towards the then government or the process -by participating in countless demonstrations taking place during the months that preceded the voting. This was the scenario at the time the impeachment was voted by the Deputies, whose speeches form the study corpus.
After heated debates that lasted over four months, Dilma Rousseff's impeachment trial was voted on a Sunday, notably the day of the week when most citizens are at home. Hence, Brazilians could watch the Deputies' voting live, as the almost six-hour-long session was broadcast by major open television channels, in addition to cable TV channels, radio stations and internet sites. Globo alone accounted for 82 million viewers (Castro, 2016) out of a population of nearly 210 million (Instituto Brasileiro de Geografia e Estatística, 2018).

The study corpus
As is the customary practice on memorable occasions, the Chamber's Department Escrevendo a história ('Writing the History') made the entire session available a few days after stenographers trained in shorthand recorded the Deputies' speeches. It seems important to emphasize that this transcription method is rather idealized, since it reproduces the standard written Portuguese texts, dismissing fillers and other markers characteristic of oral interactions, which could be indispensable to other kinds of analysis. As in this study I am interested in linguistic choices, the result of the method adopted by the Chamber was judged appropriate.
The research data in this paper is drawn from the material provided by the Chamber as a Google Spreadsheet (Câmara dos Deputados, 2008), formed by six columns for each Deputy, including: (i) name, (ii) party, (iii) state, (iv) vote, (v) gender and (vi) transcript of each speech. For the analysis I considered the texts of the sixth column, whereas the contents of the other five were kept in a header that precedes each speech. The texts were saved in plain text format, in order to be processed by Wordsmith tools version 7 (hence-forth WS7, Scott 2016), and were subdivided into three subcorpora, according to the modes of voteyes, no and abstain. The number of votes, and consequently the number of words, is quite different in each mode: votes favorable to the process account for 71.82% of the speeches (70.29% of the tokens), while the ones unfavorable to it and abstention account, respectively, to 26.81% (28.61% of the tokens) and 1.37% (1.09% of the tokens). Due to its very limited size, the subcorpus of abstaining votes was discarded. Besides, I was interested in confirming whether the lexical choice used by the Deputies who were favorable to impeachment differed from the words chosen by the Deputies contrary to the process. Thus, the focus of this paper lies on the yes and no subcorpora, henceforth Y and N, which account for the 367 pro-impeachment and the 137 counter-impeachment votes respectively.

Automatic retrieval of keywords
Using WS7, I first applied the keyword technique in order to identify the most salient words in the study corpus. This was done by comparing each subcorpus with a 76-million-word corpus of transcripts from the Brazilian Lower House sessions, which is part of the Corpus Brasileiro ('Brazilian Corpus', henceforth CB) (Berber Sardinha et al., 2010), a general language reference corpus of Brazilian Portuguese. Keywords were calculated through a combination of effect size (log ratio) (Hardie, 2014) and statistical significance metrics (log likelihood, henceforth LL). Log ratio was used to determine the difference between the frequencies of a given word or a semantic domain in the two corpora. The higher the log ratio, the larger the difference between the corpus under investigation and the reference corpus. Log likelihood was calculated to identify large frequency differences that were also statistically significant. The threshold for the size of frequency difference was a log ratio score of 2.0. The threshold for statistical significance was a log likelihood score of 6.63 (p < 0.01).
In order to avoid selecting words restricted to a handful of examples, the analysis focused on words occurring at least five times in N and 12 times in Y, as Y is approximately 2.5 times bigger than N. According to the criteria established, the quantitative analyses resulted in 65 and 101 keywords retrieved, respectively, from Y and N (see Appendix for a complete list of keywords). Tables 2 and 3 show the 20 highest-rated items by log ratio.   Since Portuguese nouns, adjectives, articles and verbs are mostly inflected, and because the corpus is not morphosyntactically tagged, for subsequent analyses different word forms were manually lemmatized and grouped with the one with the highest effect size score. For example, in Y keywords, brasileiro ('Brazilian') (log ratio 2.20) and brasileiros ('Brazilians') (log ratio 2.36) were joined under brasileiros (see Table 4). Compound proper names were also grouped, so that analyses were facilitated. Homonyms were manually distinguished. One example is the word voto, which may refer to the noun ('vote') or to the verb votar ('to vote') conjugated in the first person of the present indicative. The next section presents the similarities between Y and N keyword lists, i.e., keykeywords, along with a discussion of how these words conform to the strategies fre-quently used by politicians with the aim of (de)legitimizing their actions. I also provide a concise analysis of the keywords which differ in Y and N, and what they reveal about the Deputies' intentions towards the process in focus.

General picture: an overview of keywords
As previously explained, the fierce dispute between Rousseff and her opponent Aécio Neves in the 2014 election resulted in a strong polarization, marked by countless pro-and counter-government demonstrations across the country, which culminated in the impeachment process, first voted by the Lower House. Judging by the regular audience of TV Câmara, which occupied the 76 th position at the time (Soares, 2016), Brazilians are not usually interested in the regular sessions. But that Sunday the citizens' eyes were turned towards the Lower House to check how their representatives would vote. Obviously aware of the population's interest in the process, Deputies certainly prepared themselves to cause an impression on their voters, as this visibility could influence their reelections. After all, rhetoric is a powerful weapon used by politicians to persuade interlocutors (Reys, 2011).
At the Chamber, the long-lasting session reflected the polarized political spectrum that took over the country. Although the voters belonged to 25 different parties, grouped according to diverse priorities, such as health, human rights, evangelical community, etc., two antagonistic blocs -pro-and counter impeachment -were formed inside the Chamber. The mood was tense and one of the most troubled moments was performed by proimpeachment Jair Bolsonaro and counter-impeachment Jean Wyllys. After praising a colonel accused of torturing civilians during the Brazilian military dictatorship, Bolsonaro was spat at by Wyllys. Two years later, the former became the President of Brazil, whereas the latter decided to resign from his re-election and leave the country, claiming to have received death threats. Given the decisive nature of this moment, it was expected that the rivalry would be translated into words.
Next, I analyze the keywords which recur in Y and N, along with their co-texts, in order to verify whether, and how, the lexical choices of pro-and counter-impeachment Deputies differ. Following this analysis, I provide a brief account of the keywords which are restricted to Y or N.

Pro-and counter-impeachment keywords
The lemmatized lists account for 53 keywords in Y and 93 in N. Out of these, 28 coincide. Table 4 shows the recurring keywords in alphabetical order, along with their frequency and keyness (as log ratio).  After being analyzed in their contexts, these keywords were grouped according to the meanings they convey in the Deputies' speeches, with reference to what has been widely discussed by theorists on political discourse.

Being altruistic
Public speakers tend to express their intentions as if they were not driven by personal interests. By presenting themselves as altruistic, politicians legitimize their decisions as a response to their voters' well-being (Reys, 2011 It is important to emphasize that, although the word FAMÍLIA is key both in Y and in N, their effect sizes are quite different -4.48 in Y and 2.04 in N. Besides, the investigation of its co-text demonstrated that, apart from its canonical meaning -a social group consisting of parent(s) and child(ren) -, used in all occurrences of the word in Y, FAMÍLIA is used in three out of its eight occurrences in N as part of the proper name Bolsa Família, a welfare program created by former President Luiz Inácio Lula da Silva to provide financial aid to poor families in Brazil. A centerpiece of da Silva's administration, the program certainly played a central role in the election and re-election of his successor, Dilma Rousseff. Therefore, it comes as no surprise that the Deputies contrary to the impeachment appealed to the program to justify their votes, as illustrated in Excerpt 3: ( As mentioned above, during the weeks that preceded the voting at the Lower House, pro-and counter-impeachment demonstrations occupied the streets of several cities around the country. Allusion to the protests with RUA ('street') is frequent in Y and N, in which the word log ratio scores practically coincide -4.06 and 4.08, respectively. Excerpts 4 and 5 illustrate its use: (4) levando em consideração também a legitimidade dos protestos, as vozes das ruas, a legalidade do processo e a governabilidade do nosso País, eu voto sim, Sr. Presidente. ('also taking into account the legitimacy of the protests, the voices of the streets, the legality of the process and the governability of our country, I vote yes, Mr. President.') (5) Saudando os trabalhadores e trabalhadoras do Brasil que foram às ruas em defesa da democracia ('Greeting Brazil's workers who went to the streets in defense of democracy.')

Assuming and sharing responsibilities
In his analysis of political interviews, Karapetjana (2011) concludes that the first person singular pronoun expresses personal involvement, and it is strategically used to announce good news and to indicate that the speaker's role is important, his or her decision is individual, and he or she is not driven by the collective choice of the peers. Eu ('I') has a large effect size and occupies the 19th and 9th position in Y and N keyword lists, with a log ratio of 5.30 and 5.36, respectively. After presenting their reasons, making it clear that their priority is the common good, the Deputies assume the responsibility for their choices by positioning themselves: eu voto/digo [sim/não] ('I vote/say [yes/no]'), eu vou votar ('I will vote'), eu sou contra ('I am against'). Usually combined with aqui ('here'), the explicit or implicit first person singular pronoun calls attention to the speaker's attitude for the common good, as shown below, with excerpts from Y and N speeches: The first person plural, on the other hand, is strategically used 'to induce interpreters to conceptualize group identity, coalitions, parties, and the like' (Chilton, 2004, p. 56). It shortens the distance between the Self, the speaker, who is here, and the interlocutor. Karapetjana (2011) explains that the plural pronouns are especially used when controversial and unpopular decisions are taken because they invoke shared responsibility, convey a sense of collectivity and contribute for trustworthiness. Given the importance of that voting, the Deputies unsurprisingly presented their choices as if they were serving their voters' will, albeit with a lesser extent if compared with the singularnós ('we') has a log ratio of 2.56 in Y and 2.30 in N. Let us look at the Excerpts 8 and 9 below, which represent the use of the plural in both keyword lists, respectively: This recurrent strategy corroborates Capone's (2010Capone's ( , p. 2967) explanation of the reason why the politicians speak in the name of we: 'After all, to win an election a speaker must become the representative of a community of people (an aggregate of social groups), and to become such a representative one must show/prove that one's voice is the voice of the nation, or at least expresses the voice(s) of the nation'. Proctor & Su (2011) also explain that personal pronouns perfectly serve rhetoric because they are used to convey different meanings in an easy way: by using the first person plural, politicians take distance from they, the ones who do not share the same ideology.

Appealing to emotion
Appealing to emotions is an effective strategy politicians use in order to imbue veracity to their words. Reys (2011) explains that the choice of words that evoke in the interlocutors' minds a series of previous experiences triggers emotional links with past situations, besides distorting the interlocutors' previous opinion about something. Regarding delegitimization, he concludes that, by resorting to emotion to demonize the opponent, the speaker leads the interlocutor to fear the future in case this opponent wins the contention. In both Y and N speeches Deputies boast of being contra ('against')corrupção ('corruption'), ditadura ('dictatorship'), golpe ('coup'), etc. -, and mutually accuse their opponents of practicing the reprehensible actions. Among N keywords, the use of the word contra is even stronger, accounting for a log ratio of 3.89 versus 2.17 in Y. Delegitimization, i.e., the representation of opponents through negative characteristics, accusations and offenses, is a recurrent weapon used by politicians (Chilton, 2004).
According to Reys (2011, p. 785), the strategy of attributing positive qualities to the 'usgroup', whereas the 'them-group' is depicted negatively, is also a form of legitimization through emotion. In this study, the Deputies' keywords express mutual accusations by approaching similar issues. For example, both parties claim to stand against corruptionits Portuguese equivalent corrupção has a similar effect size: 3.82 in Y and 3.55 in N -, hence showing that the Deputies mutually accuse the opponents of committing this crime. Excerpts 10 and 11 illustrate Y and N claims, respectively: Cometeu ('committed') and crime (de responsabilidade) ('breach (of fiscal law)') are controversial words recurrently used by both sides of voters and which demanded an analysis that goes beyond keyness, a measure which places the words in 13th and 26th positions in N, and 4th and 47th in Y, respectively. The investigation of the concordance lines of cometeu showed that, in the 13 occurrences in Y, the Deputies claim that the then President definitely committed the crime of breach of fiscal law. The only two times não ('no, not') appears in the surroundings of the keyword in Y it is used with só ('only') -forming não só ('not only') -to include other accusations, whereas in N, não is used only in defense of the then President. Excerpts 12 and 13 illustrate the mutual accusations in Y and N, respectively: (12) a Presidente da República não só cometeu crimes como as pedaladas, mas sobretudo mentiu ao País ('the President of the Republic not only committed crimes such as the breach of fiscal law, but above all she lied to the Country').
Orders invoking the dismissal of opponents, introduced by fora ('out'), were frequently used both by pro-and counter impeachment, although to a lesser extent in N, in which the word achieved a log ratio of 2.45, against 3.28 in Y. Another difference identified is the persona non grata, obviously. In Y, the order is frequently followed by Dilma, Lula and PT, whereas in N, by (Eduardo) Cunha and golpistas ('coupists'), as illustrated respectively in Excerpts 14 and 15: (14) Eu voto sim. Fora, Dilma! Fora PT! ('I vote yes. Dilma out! PT out!').
However, the Deputies' speeches express a balance between negative and positive expression of emotion. Although with rather different keyness -2nd and 28th positions in Y and N, respectively -, both pro-and counter-impeachment Deputies repeated the adjective QUERIDO ('beloved') to show empathy towards places, especially their cities and states of origin, or the ones they represent in the Lower House. It is worth emphasizing, though, that the feminine form of the adjective was also ironically used by some pro-impeachment Deputies as a salutation, in allusion to a wiretapped phone call between Dilma Rousseff and former President da Silva, in which da Silva closes the conversation saying Tchau, querida. ('Bye, dear'). After analyzing the political use of language in speeches and interviews, Chilton (2004) concluded that, along with fear, anger, a sense of security and loyalty, protectiveness towards the family is also stimulated as an emotive legitimization strategy, as the family represents the center of social entities, and, as such, it contrasts with the 'outsiders' (Chilton, 2004, p. 52). Justifying the vote on behalf of family members was also identified as a recurrent strategy in the study corpus, being the keywords filhos ('children, offspring') and família ('family') present in both Y and N, albeit with bigger effect size in Y: the former accounts for a log ratio of 4.27 in Y and 2.97 in N, whereas the latter, 4.48 and 2.04 in Y and N, respectively. A closer look at the co-context of filhos indicated a recurring tendency of justifying the vote in the name of the offspring, as pelos ('for') followed by the possessive adjectives meus ('my'), seus ('your, their') and nossos ('our'), collocates with the search word in both subcorpora.

Appealing to emotion
Religious rhetoric is a common practice among politicians. Crines and Theakston (2015, pp. 159-160) claim that British Prime Ministers' political rhetoric is permeated by religious language 'to justify policy, support their ideological positions, present a public persona, and underline their personal ethical appeal to highlight their individual moral suitability to be a national leader'. Once a traditional Republican practice in the United States, the 'God strategy' (Domke & Coe, 2008) has been widely used by left-wing politicians to approach voters. Referring to the United States' presidential campaign in 1992, they note: 'Clinton had well learned what has become perhaps the most important lesson in contemporary American politics: to compete successfully, politicians need not always walk the religious walk, but they had better be able to talk the religious talk' (Domke & Coe, 2008, p. 6). And it seems that the lesson has been well learned. They point out that since the mid-1970s, all United States' Presidents have explicitly invoked God in their addresses, and this is an increasing trend, which seems to affect other countries. After ana-lyzing the speeches of politicians from different nationalities, Chilton (2004) notes that the language of political discourse is commonly intertwined with religious beliefs. Burity (2008) also observes that in central countries there has been an increasing advance of movements and situations involving religious actors -social and political actors whose religious identity is a relevant component. In the case of Brazil, the Roman Catholic monopoly has lost ground to evangelicals, but Christianity still prevails. From the 1980s on, the Evangelical Bench has gained ground in Brazilian politics, and, according to Burity (2008, p. 84), 'either in culture or daily life, religious actors move around and make their language, their ethos, and their demands public in the most diverse directions'. As the Brazilian Evangelical Deputies, who account for approximately 40% of the Lower House representatives (Câmara dos Deputados, 2015), were massively favorable to impeachment, it comes as no surprise that Deus ('God') is a keyword in Y, with a log ratio of 3.84. Nevertheless, it is also among the keywords from counter-impeachment Deputies -although with a lower effect size (2.32) -, formed basically of left-wing politicians, whose ideology traditionally repels religious references. In face of this statistical similarity, I investigated the concordance lines of the keyword and concluded that its contexts of use are not always coincident. In Y, Deus, with 49 hits, collocates with words commonly used in religious contexts, such as abençoe ('bless'.PRS.SBJV.3SG), Senhor ('Lord'), agradecer ('thank') and pedir ('plead'), resulting in appeals such as the one illustrated in Excerpt 18 below: In N, on the other hand, out of its seven occurrences, Deus is used (i) to criticize the speeches of those who supported impeachment (four times), (ii) as an interjection (one time) and (iii) to invoke divine help (two times), as illustrated with Excerpts 19, 20 and 21, respectively: Therefore, I conclude that, in spite of being a keyword in both subcorpora, Deus is not used with the same intention in them, since in Y Deputies invoke the name of God with religious aim, whereas in N there is more criticism towards the other group's speech.
The investigation of the recurring keywords showed that words related to altruism, responsibility, emotion and religion corroborate other researchers' assumptions about political strategies of (de)legitimization. More specifically, of the three words which triggered this study, namely nação, Deus, and FAMÍLIA, the last two recur both in Y and N, albeit with diverse keyness and in different contexts of use. Of these, only the first is recurrent only in Y.
Next I give a brief account of the keywords which are proper of Y or N. It is important to emphasize that investigating every single keyword which characterizes pro-or counter speeches is beyond the scope of this study. Thus, I sought to group the keywords according to the meanings they convey, in a search to identify what they reveal about each group's intentions.

Pro-impeachment strategies
In order to legitimize their decisions, politicians usually connect past, present and future, i.e., they justify that a present action should be taken to solve a problem caused by bad past decisions, and that this action will have a positive impact in the future. Legitimizing present actions by resorting to hypothetical future accomplishments is a recurrent political strategy of persuasion (Reys, 2011) that was unsurprisingly recognized in the pro-impeachment Deputies' keywords. After all, those voters present themselves as the authorities who have the power to correct the course of ongoing problems, supposedly created during fourteen years of PT government. Hence, they aim to convince their interlocutors that they envisage a better future, which is only possible if impeachment is approved. Keywords such as esperança ('hope'), futuro ('future'), mudança ('change'), and melhor ('better') corroborate this legitimization strategy.
Persuasion through emotion was expressed by the recurrent use of words which convey excitement, such as viva ('hurrah'), usually followed by Brasil, a keyword that, along with nação ('nation'), reveals patriotism, a concept mostly associated with rightwing principles (Osler, 2009). Chilton (2004) also recognizes the appeal to patriotism as a politically relevant feeling, as is love. In Y speeches, amor ('love') is recurrently dedicated to the country as a whole, and also to the states the Deputies represent. After all, as Chilton (2004, p. 117) explains, 'The sense of security is related to one's geographical territory'.

Counter-impeachment strategies
While pro-impeachment voters justified their choice by proposing changes, their opponents obviously expressed the wish that things should continue the way they were by appealing to the legality of Rousseff's election process. Keywords such as legitimidade ('legitimacy'), urnas ('ballot boxes'), constituição ('constitution'), respeito ('respect'), soberania (do voto popular) ('sovereignty (of the popular vote')), lutaram (pela democracia) ('fought (for democracy))', defender (a democracia/a Constituição) ('defend democracy/Constitution'), and defesa ('defense') help remind the interlocutors of appropriate past decisions that have a positive impact in the present. Resorting to popular programs created during the years of PT government, such as (Bolsa) Família and Minha (Casa, Minha Vida) can be understood as an attempt to alert the interlocutors of the imminent losses resulting from the impeachment.
Defined by Ekström et al. (2018) as 'both a political discourse, or "thin-centred ideology", representing politics and society as structured by a fundamental antagonistic relationship between "the elite" and "the people"', populism is highlighted in N through keywords such as trabalhadores ('workers'), pobres ('poor'), and classe (trabalhadora) ('(working) class'). The researchers claim that the populists' discourse is permeated with demonstrations of empathy for the ordinary citizen, who would be subjugated by the privileged elite. As the then President belonged to a left-wing party, it was expected that her supporters would demonstrate a more socialist ideology than her opponents. Some keywords in N corroborated this expectation. Counter-impeachment Deputies praised people's accomplishments and defended assistance programs and minorities through words such as (estado) democrático (de direito) ('democratic (rights)'), (reforma) agrária ('land (reform)'), companheiros ('comrades'), luta ('fight'), juventude ('youth') and mulher ('woman'). It is also worth mentioning that, as Portuguese prioritizes the masculine forms when generalizations are made, genre ideologies, boosted by the rise of left-wing parties during the previous decades, also impacted on the Deputies' lexicon, as observed especially in the speeches of those who claim to fight for equality. This trend was also highlighted in counter-impeachment speeches. Examples are the use of both masculine and feminine when generalizingtrabalhadores e trabalhadoras ('workers.PL.M and workers.PL.F') and Deputadas e Deputados ('Deputies.PL.M and Deputies.PL.F') -, besides the feminine Presidenta ('President.SG.F') to refer to Rousseff, although the neutral title Presidente is broadly accepted in Brazilian Portuguese Obviously, praising the then President with words such as honrada ('honorable') and honesta ('honest') was also a recurrent strategy in N, whereas the attribution of negative characteristics to the opponents helps create a distance between the speaker and the ones contrary to their arguments. When the discussion about impeachment began, those contrary to it started associating the process with the military coup d'état of 1964, when President João Goulart was deposed. The slogan 'Não é impeachment, é golpe' ('It's not impeachment, it's coup.') reverberated throughout the country, and it was loudly repeated by the counter-impeachment Deputies while voting. Criticism towards the ones who supported the process was demonstrated through the repetition of words such as golpe ('coup'), golpistas ('coupists'), covardes ('cowards'), farsa ('farce'), hipocrisia ('hypocrisy') and corruptos ('corrupts').

Concluding remarks
This study was motivated by an interest in comparing the short speeches delivered by pro-and counter-impeachment Deputies during the Lower House voting session on the impeachment trial of then President Dilma Rousseff, in order to verify whether, and to what degree, their lexical choices differed. The analysis of Y and N keywords, carried out using a combination of effect size and statistical significance measures, showed that, in spite of representing antagonistic opinions about the process under discussion, both proand counter-impeachment Brazilian Deputies tended to choose similar words to justify their votes, corroborating other researchers' conclusions about legitimization of political strategies through words indicating positive self-representation, altruism, closeness to the voters and to their place of origin, and appeal to religion. The keyness analysis showed that, of the words related to the so-called triad (Deus, família and nação), the first is recurrent in the Deputies' speeches who voted either for or against the process, even though the word was not equally frequent in both modes of votes. More importantly, the analysis of the concordance lines revealed that this word was not always used with the same meaning in the two types of voting, showing that the mere counting of decontextualized words should not suffice to jump to conclusions about the Deputies' ideologies.
Justifications associated with family members also recur in both subcorpora, although they were used in higher proportion in Y, especially if the uses of família as a proper noun in N -Bolsa Família -were ignored. Nevertheless, referring to family members to legitimize the Deputies' votes was a recurrent strategy in the counter-impeachment voters as well.
As for nação, it was key only in pro-impeachment speeches, and was also employed as a strategy for legitimizing votes through reference to patriotism, especially if the word Brasil is also considered as a synonym. However, speaking in the name of their voters, by making reference to the corresponding states and cities, was a recurrent strategy of both pro-and counter impeachment Deputies. The combination of quantitative and qualitative analyses also showed that both Y and N speeches repeatedly resort to mutual offenses to delegitimize their opponents Regarding the keywords which appeared in one of the modes of voting, pro-impeachment Deputies tried to persuade their audience and legitimize their choices through promises of a better future, whereas their opponents defended the then President and reminded their interlocutors of her social programs, which privileged poor people.
CL is fast becoming a key methodology in discourse studies. The quantitative analysis, based on keywords, combined with the manual analysis, helps reveal data that would otherwise be restricted only to randomly chosen examples and biased interpretation of the analysts, who always speak from an ideological position constructed during their lives. Thus, a CL approach allows for a more objective discourse analysis, through the identifi-cation of patterns by computational tools as a starting point. Total objectivity is obviously impossible. In this study, for example, I chose as the reference corpus other political speeches -a different choice would certainly result in different keywords; I also prioritized the analysis of the keywords with the highest log ratio measures, and chose excerpts from the corpus to illustrate the quantitative results obtained. However, a raw word count is not enough for conclusions to be drawn when unbalanced sets of data are taken into account, as the number of votes favorable to the process was 2.68 times higher than those opposed to it. Furthermore, the analysis of the co-text of the words is vital in unveiling the real differences between the speeches.
In line with Chilton (2004), this study has shown that, regardless of their ideology, the Deputies' lexical choices aim at legitimizing their actions and delegitimizing their opponents'. Hence, I could add to the epigraph of this article that politics may be a world of poorly understood, remembered and theorized words (Hart et al., 2013), but it is also a world of strategically used words.