Emotive, evaluative, epistemic: A linguistic analysis of affectivity in news journalism

In this article, we introduce a linguistic approach to studying affectivity as a fundamental feature of news journalism. By reconceptualising affectivity beyond emotive storytelling, intentional stance-taking or evaluative expression, we propose a methodology that highlights how conventions related to mediating, modulating and managing affectivity permeate journalistic genres. Drawing from conversation analysis, Bakhtinian theory of language as dialogical and notion of affective meaning-making, we investigate how selected linguistic forms and structures – namely evidential and epistemic modals and lexical items signalling affective intensity (such as emotive and evaluative words and metaphorical expressions) – participate in affective meaning-making in news journalism. A scalable computational methodology is introduced to study multiple linguistic structures in conjunction. In investigating a case study – the news reporting and commentary on a highly charged, year-long political conflict between the right-wing conservative government and the trade unions in Finland (2015–2016) – the approach allows a focus on the ways in which affectivity operates in journalistic texts in response to both generic expectations of the audience and journalistic conventions. Our findings include identification of the intertwining of strategic rituals of objectivity and emotionality, recognition of metaphoricity as a key source of affectivity and detection of different news article types having their own conventions for managing affectivity. We also observe a connection between emotive and evaluative words and the grammatical constructions used to express degrees of certainty, which suggests these modal constructions play an important part in how affectivity informs journalistic texts.


Introduction
In the contemporary context of hybrid media and networked publics, affect is discussed as an increasingly important issue for journalism. What affect entails is, however, not self-evident. Examining the exceptional news ambience of the Egyptian uprising in 2011, Papacharissi and de Fatima Oliveira (2012: 279) introduced the notion of 'affective news' to describe how news, opinion and emotion in the context of political crisis become indiscernible from one another. The resulting 'affective news streams' were offered as illustrations of the evolving paradigms of hybrid journalism beyond their specific case. Beckett and Deuze (2016: 1) coined the notion of an 'affective media ecosystem' to describe how the technological changes and simultaneous pressures on the media economy reframe journalism as 'an emotionally charged networked environment'.
This article engages in this discussion of the importance of affect in journalism by offering a revised understanding of the concept. Rather than confirming or questioning any narrative of change in journalism, this article proposes an approach to affectivity as a quality permeating journalistic language through multiple linguistic constructs and introduces a scalable computational methodology to study it. In the following, we address two theoretical and methodological questions: What is affectivity in journalism? How can affectivity be made empirically observable in big data?
We also ask how affectivity operates in journalistic texts in response to generic expectations, both of the audience and of journalistic conventions. More specifically, we examine how quotes are used to mediate, modulate and manage affect in journalistic texts and how different genres of news journalism (news, analysis, commentary, editorial) vary in their framing of affectivity. We illustrate this approach through analysis of a particular case study: the news reporting and commentary on a highly charged, year-long political conflict between the right-wing conservative government and the trade unions in Finland (2015Finland ( -2016. In this article, however, the details of this political event are not discussed. While our approach highlights the contextuality of communication and meaning-making, the case itself is not of prime importance. The approach strives for scalability in an age of big data and computational analysis. As such, this article contributes to journalism studies by outlining a theoretical and methodological approach to affectivity in news journalism, enabling empirical and historical analysis of changes and continuities. While this article does not address the overall question of an increased importance of affect in contemporary journalism, the proposed approach enables its study. By examining how affectivity is translated into linguistic structures in different genres of news journalism, we also seek to contribute to further studies of news narration and style (Bednarek, 2008;Bednarek and Caple, 2012;Broersma, 2007;Peters and Broersma, 2016).

Studying emotions in journalism: the state of the art
Until recently, emotions were a neglected issue in journalism studies -'an epistemological elephant in the room' (Wahl-Jorgensen, 2019: 167). The general disinterest in investigating the roles, forms, meanings and effects of emotion in journalism can be seen as resulting from the negative connotations that emotions and emotionality have held within the journalistic profession (Pantti, 2010;Peters, 2011;Wahl-Jorgensen, 2019). In their study of UK journalists, Richards and Rees (2011) noted that emotions were typically suspected of compromising impartiality. Finnish and Dutch television journalists who were interviewed around the same time had a more favourable view of emotional elements in the news. These journalists highlighted the importance of emotions in people's lives and hence a subject to be covered; the journalists also stressed the important role of emotion in storytelling (Pantti, 2010). At the same time, much like their British colleagues, they, too, rejected acknowledging themselves as a locus of emotion.
For journalism studies, a wider turn to affect and emotion in social sciences, cultural theory and gender studies (Clough and Halley, 2007;Koivunen, 2010), alongside the emergence of a new, networked media ecosystem, have together made the question of emotion more acute than ever. Peters (2011) has highlighted how journalism has always been emotional in terms of attempting to engage the audience by constructing an 'experience of involvement' through evoking common narrative forms featuring heroes and villains. As distinctions between news and entertainment, hard and soft news, and fact and opinion have become blurred in the new media ecology, the rigid juxtaposition of emotionality and involvement versus neutrality and detachment has been questioned. As a result, the diversity of emotional styles has increased and the acceptability of journalistic emotional involvement has become more explicit and accepted (Peters, 2011).
To investigate emotionality in news journalism against this background, we follow Tuchman's (1972) understanding of journalism as a profession in which 'strategic rituals' function as practices of legitimacy in relation to both professional standards and the public. According to Tuchman (1972), the objectivity norm manifests itself in strategies with which a journalist appears to stand outside any subject at hand: to be dispassionate, disembodied and impartial. Operating alongside this 'ritual of objectivity, ' Wahl-Jorgensen (2013a, 2013b argued, is the 'strategic ritual of emotionality' that regulates the presence of affect in journalistic texts. Wahl-Jorgensen and other scholars (Bednarek, 2008;Stenvall, 2008Stenvall, , 2014White, 1998) have used appraisal theory (Martin and Rose, 2003) to examine the collection of conventions and codes for how emotions are incorporated in journalism through various narrative strategies -dramatic tension, detailed descriptions, anecdotal leads, juxtaposition and personalised storytelling -as well as through outsourcing emotions to story protagonists or other sources (Wahl-Jorgensen, 2013a; see also Huang, 2017;Stenvall, 2014 ;White, 1998). The intertwined rituals of objectivity and emotionality appear as 'a tactic of strategic impersonalisation by which the intersubjective role of the author is backgrounded, obscured or suppressed' (White, 1998: 267).
To study emotionality as a question embedded in practices is not merely to study emotion as a social, rather than individual or psychological, phenomenon; it is to study emotion as a question of professional codes, journalistic forms and linguistic structures. Such an approach also regards emotion itself as a norm that is subject to normative regulation, thus enabling (albeit beyond the scope of this article) an analysis of how the 'emotional labour' of journalism persists and changes over time.

Affectivity as meaning-making: Our approach
In this article, we reconceptualise emotionality in news journalism as affectivity. In using the concept of affect, we acknowledge the rooting of the concept in new materialist thought, which aims to reject the focus on language and representation (Clough and Halley, 2007;Massumi, 2002). Following Wetherell (2012: 19-25), we choose to define affect as being intertwined and cooperating with discourse. Underlining 'the relational, dialogic and distributed aspects of meaning-making,' Wetherell defined her object of study as 'affective-discursive practices,' 'affective orders of social life' and 'affective meaning-making' (Wetherell, 2012: 51-53). Affect is thus not something that is reducible to linguistic and discursive practices alone but is entangled with them. Distinct from understanding emotion as expression, such an approach towards affect highlights its character as intersubjective and impersonal. Importantly, this formulation also highlights what Papacharissi (2015: 5) termed 'pre-emotive intensities' and energies, which form and structure the public and which we study by focusing on linguistic structures.
Concretely, our approach to affect is based on two theoretical starting points: (1) that affectivity and emotion 'are ever-present concerns of participants in social situations' (Ruusuvuori, 2012: 330, paraphrasing Goffman, 1961 and (2) that the fundamentally dialogic nature of language (Bakhtin, 1981;Volosinov, 1995) justifies treating any use of language as such a situation. From this viewpoint -and drawing on appraisal and politeness theories (Brown and Levinson, 1987;Myers, 1989;White, 2003) -affectivity then appears as an ever-present quality that all communication must manage in order to be received in the intended manner. This view enables us to analyse affect, as well as its modulation, in quantifiable terms and as something that can be present with more or less intensity.
The concept of 'genre' in linguistic genre analysis (e.g. Swales, 1990) emphasises how the norms of specific types of linguistic communication stem from their different communicative purposes and, further, that the genres emerge from these predominantly latent normative structures. Looking at distinctions in linguistic conventions regarding affectivity between journalistic genres thus allows one to make observations about the genres' communicative functions. Conversely, if the distributions of different linguistic features related to affectivity display clear and consistent patterns within and across the genres, one can argue that the handling of affectivity is something important that all journalistic genres specifically take into account.

Data: The Finnish case of the 'competitiveness pact'
Our dataset comprises all content published in 2015 and 2016 that mentions the competitiveness pact from four different news sources that were chosen to represent widely read news sources with different profiles and terms of production in order to provide a broad overview of the national news coverage (Newman et al., 2019: 82-83). The dataset is summarised in Table 1 as follows: (1) HS comprises all articles published by Helsingin Sanomat, a widely read national daily, either in its paper version or on the web; (2) YLE contains all articles published on the website of the Finnish public-service broadcaster, the most trusted news source in Finland; (3) IL contains the online articles published by Iltalehti, one of the two national tabloids, which reaches online more than the half of the Finnish population every week; and (4) STT consists of the news articles produced by the Finnish news agency, which serves all Finnish media but is especially important for the regional press. In addition, we collected, as a separate baseline data set, all articles published in the politics section of Helsingin Sanomat between 13 June 2016 and 11 November 2016 (HS-REF). As seen in Table 1, the vast majority of articles from all news outlets were factual news stories. Indeed, their volume was such that any analysis that did not differentiate between the categories would have spoken only for factual news.

Methodology: Identifying aspects of affective language use
Previous studies, especially corpus-based studies based on appraisal theory (e.g. Huang, 2017) and those applying computational methods, such as sentiment analysis, have commonly approached affectivity and emotions through overt emotive vocabulary and words that signal clear negative or positive evaluations. The role of emotive language in constructing and evaluating news values in journalism has been studied particularly by Bednarek (2008). Building on this experience, we started by automatically extracting all 4732 unique adjectives that appear in the corpora. Our linguist reviewed this list manually to remove non-evaluative adjectives. The 1875 non-evaluative words that were excluded consisted mostly of technical and temporal qualifications and other classifications, such as nationalities. What remained were adjectives that expressed evaluations or characterisations of people, objects or events. Beyond adjectives, however, almost any linguistic structure can express affectivity. Therefore, media scholars in the project close-read a sample of news articles in our corpus to obtain a data-informed understanding of their emotive and evaluative vocabulary beyond the adjectives. We were struck by the large amount of metaphorical languagean evidently affective practice (Wetherell, 2012), but one that was both fundamentally different from the evaluative expressions and seemed to appear particularly in response to inherently affectively charged segments of text.
To give an example, in the following passage, many metaphorical expressions (italicised) occur in quick succession: '[. . .] all expenses must come down. Regardless of how current expenses stand in international comparison. Everyone must suffer. The question is of principles and before them all rationality dissolves. There is storm and mayhem ahead, when the new course is set.' 1 Here, the metaphorical use of storm can lend some conceptual features of the source domain (climate) to the target domain (economic policy). Future economic and societal turmoil is framed in terms of a natural phenomenon that is characterised by independence from human intervention and a degree of inevitability. At the same time, because the metaphors overall come from different domains with incompatible features, this piling-up impedes a reading focused on propositional content. Instead, it invites an understanding of the passage in more poetic terms, in which the metaphors primarily build up affective intensity.
Such prominent, foregrounded metaphors are, however, content-specific and cannot be identified using general-purpose word lists nor described using syntactic rules. Yet, they all have something in common: the more foregrounded they are, the more distant their source domain is from the target domain. This enables the recognition of metaphors in a particular discourse on the type level: each occurrence of the word storm is highly likely to be prominently metaphorical in a discourse on economic policy. The narrower the discourse investigated, the easier it is to enlist domains that are distant to it.
To operationalise the identification of metaphors for our particular case regarding the competitiveness pact, we manually analysed a sample of the articles and marked all passages with metaphorical language. Our linguist proceeded to examine the central vocabulary of these annotations and extracted a seed-word list of 970 words that marked metaphorical discourse. Using pre-trained 'word embeddings' 2 to discover other words that were used in similar ways, we automatically expanded this list to 1486 further metaphor-marking candidates. After further filtering, we ended up with 2100 metaphormarking words. From an evaluation against a manually annotated gold standard, the resulting list was found to adequately locate the metaphorical language in the corpus.
Going beyond emotive and evaluative vocabulary -and assuming the operation of the strategic rituals of objectivity and emotionality (Tuchman, 1972;Wahl-Jorgensen, 2013a, 2013b) -we also wanted to study how affect was modulated or 'hedged' in the material. Here, we turned to grammatical and lexical structures used to express 'evidentiality' and 'epistemic modality.' In communication, these structures are typically used to express reservations or commitments or to communicate degrees of uncertainty or conviction, and they have been shown to be heavily involved, especially in academic genres, in politeness strategies (Brown and Levinson, 1987;Myers, 1989;White, 2003). To give an example, in the following citation, almost all the statements are hedged: 'It is likely that the other parties will follow suit and a week-long stalking period will begin. Risky play, I say. It is quite possible that the agreement will fail for the fifth time. This would be a great loss for the Finnish economy, with its less-than-good ratings.' 3 The example has two epistemic modal constructions -it is likely and it is quite possible -and one modal verb would (which here stands for the morphological conditional mood of the Finnish original). Their function is, firstly, to shield the writer from committing too strongly to the proposed scenario. Too much conviction and commitment would invite a challenge to the writer's assumptions that (1) the other parties would follow suit and, (2) as a result, the negotiations would fail and (3) the economic consequences of the failure would be dire. By creating distance from these assumptions, the writer can appear more neutral and impartial. Secondly, hedging also serves an affective purpose; it allows the actual content of the statements to carry more affective meaning (relating here to mistrust between the participants and fears over economic hardship). This interpretation is further supported by how other affective markers co-occur in the passage: metaphorical stalking, play and fall plus evaluative great (loss) and less-than-good all combine with hedging to amplify the affective meaning of the passage. In our work, we gathered these grammatical structures that expressed evidentiality and epistemic modality in Finnish based on a review of established research (Hakulinen et al., 2004;Kangasniemi, 1991;Laitinen, 1992).
In summary, the first set of linguistic markers are evaluative adjectives and metaphorically used nouns and verbs. For the sake of brevity, these are referred to as affective vocabulary in the course of reporting our findings. Evidential and epistemic modal structures were also studied; this group of markers is referred to as hedging markers or simply as hedging. As the main measure, we use the ratio of markers per words in a particular section of text. Thus, the amount of hedging refers simply to the number of hedging markers per word in a piece of text. Likewise, the amount of affect or affective intensity refers to the proportion of words marked affective out of the whole.
After listing the linguistic markers, we determined the material divisions along which to compare them. First, we wanted to study one hypothesis associated with the ritual of emotionality (Wahl-Jorgensen, 2013a, 2013b) by investigating whether affectivity -not merely in prize-winning stories, but in news journalism in a wider sense -was outsourced from journalists' discourse to sources in quotes. We thus developed methods to identify both direct and indirect quotations.
As a second step, we examined possible external factors that could have influenced the quantity of evaluative and emotive words. We tested a 'click-bait' hypothesis, which posits that affectively intensive content will be located at the beginning of an article as a means to solicit reader interest. We also investigated whether the affectivity and intensity of the actual events influenced the journalism by (1) comparing the reporting on the competitiveness pact with baseline political journalism of a randomly chosen earlier period and (2) comparing articles written during periods of peak activity with those written at other times.
We proceeded to analyse the differences among different news media, ranging from the Iltalehti -which, in terms of its journalism, is more a newspaper than a tabloid in the British sense of the word -to the Finnish news agency STT, whose direct clients are not the public but other news outlets, which may choose to republish or extend their stories on their own sites.
Finally, we examined affectivity in various types of news article: (1) news reporting focused on facts; (2) news analyses with an emphasis on contextualising, explaining and interpreting events; (3) news commentaries or columns where the writer's voice and opinion are foregrounded; and (4) editorials and leading articles, either signed or nonsigned, that express the viewpoints of the paper. To distinguish between these types, we drafted classification rules for each news source based on structured metadata. While this caused some heterogeneity in the types of story categorised as news -including longer feature articles, personal interviews and viewpoint articles -the number of such stories was marginal in comparison to factual news reporting, although it does testify to the blurring of genre boundaries.

Managing affect by utilising quotes
In the following analyses, all reported differences were determined to be statistically significant to the degree that equal differences could appear randomly at most one time in a thousand (p ⩽ 0.001). Where the data was clearly normally distributed, the significance was ascertained by a t-test. Some feature distributions were, however, heavily non-normal and zero-inflated. To verify the significance of differences for these distributions, a negative binomial regression with a binomial logit hurdle at zero was applied, using the R package pscl. Besides the differences we report, many other differences in the data were also statistically significant, but we have opted to report only those results where the difference was noticeable. Examining how affectivity was located and modulated in the articles, we first tested the claim associated with the ritual of emotionality that, in journalistic texts, emotion is often outsourced into quotes. Here, we found remarkable complexity in how the affective and hedging markers behaved with regard to content type, as shown in Figure 1.
Focusing on the overall averages, direct quotes contained more affective markers (median 8.0%) than either indirect quotes (7.1% affective) or the non-quote parts of the articles (6.1% affective). This result is indicative of the operation of the ritual of emotionality and of the convention of using quotes as a platform for statements with high affective intensity. Further comparing the distributions for direct and indirect quotes, indirect quotes have a much larger proportion of completely 'affect-free' quotes than direct quotes. This suggests that, while most direct quotes carry affective meaning, indirect quotes serve a dual function not only as a locus of emotive expressions, but also to merely attach a source to factual information, thus serving the ritual of objectivity.
When we looked at the amount of hedging in the differently quoted and non-quoted portions of the articles (right side of Figure 1), we noted that, just as with affective vocabulary, direct quotes featured the most hedging, followed by indirect quotes and finally non-quote content. At the same time, these structures were markedly rarer than affective vocabulary as a whole, which shows that emotive and evaluative words also freely occur without reservations or softening. Particularly for indirect quotes, the portion of quotes that did not contain hedging structures was significantly larger than the portion of quotes that did not contain affective language. Combining these observations, we posit that, in these texts, direct quotes were used to frame the most controversial content, which the sources themselves also commonly hedged, while indirect quotes were much more varied, ranging from non-affective to moderately affective to highly affective and hedged.
To further explore the relationship between affective vocabulary and hedging, we graphed the behaviour of the hedging markers in differently affective portions of the texts (Figure 2). Here, we can see that, when the text contains no affective vocabulary, it also almost never contains hedging, which further corroborates our hypothesis of how these two phenomena are linked in language. In contrast, when affective language is present, the amount of hedging can vary by a large amount, but with the median amount of hedging remaining unchanged as the intensity of the affect increases; however, in the The curve height depicts the proportion of articles that exhibited a particular percentage of markers. Individual observations are plotted as points inside the curve, thus highlighting the discontinuity (and therefore also errors in the density estimate) around zero. The red bars are proportion markers, to the left of which ¼, ½ and ¾ of the articles lie.
two types of quote in the very top brackets for affective intensity, this relationship changes. The affect becomes less hedged, further raising its perceived intensity. In summary, most overt affect in the texts is hedged, but there is also a small proportion of highly charged quotes whose intensity is further increased by the lack of hedging.
Another interesting observation is that the general order of the content types in terms of the amount of hedging persists across all intensity brackets. This means that, while direct quotes are generally used to present the more affective content, when this is done in other parts of the texts, the amount of hedging follows the general conventions for the type of content instead of the baseline given by direct speech. In effect, this may further heighten the perceived intensity of affect for these uncommon ways of framing.

Affect and hedging in various news outlets and article types
After examining how affect was managed through different types of quoted and nonquoted content in general, we investigated whether there were differences across these content types between the news outlets and article types. For direct and indirect quotes, neither outlet nor article type produced significant differences in the proportion of either affective or hedging markers (not shown), apart from direct quotes in Helsingin Sanomat, which featured slightly more affectivity and hedging than the other sources. This indicates a remarkable stability in how quotes are used across all of journalism, from news agency to tabloid. As an individual observation, Helsingin Sanomat appears to have been particularly careful in following the ritual of emotionality: in general, affective discourse resided in quotes, and more hedging occurred overall than in the other sources. This may indicate that control of affective intensity is a more central concern for press journalism than it is for news agencies or tabloid journalism -or even for Finland's public broadcaster. Whereas the amount and modulation of affective intensity in quotes was equal across the outlets and article types, this did not hold for the journalistic narrative voice (Figure 3). For article types, news articles featured the fewest of both affect and hedging markers, which was not surprising as they reflected the rules of the news genre and the ritual of objectivity. Between outlets, the tabloid Iltalehti contained more markers in news articles, while the STT news agency used the least amount of affective language. As a news agency, its news output consists mainly of short, factual stories for publication or extension in other outlets, from tabloids to major regional newspapers and public-service media (cf. Stenvall, 2014). The public-service media YLE took a middle position, not distinguishing itself in any aspect of the analysis.
Returning to the article types, next on the ascending affectivity scale after news were news analyses, including editorials from the major newspaper Helsingin Sanomat. The editorials of Iltalehti, in contrast, contained the highest amount of affective vocabulary -even more than the category of commentary, which otherwise featured the most. Interestingly, in terms of hedging, Iltalehti editorials also contained the most markers (median 2.8%, compared to the overall median of 2.4%). Iltalehti commentaries, on the other hand, while containing highly affective language, had the least hedging (median 1.8%). Of the textual styles, tabloid commentaries appear to be the genre whose authors have the freest reign to write affectively without needing to hedge their words.
Comparing the different modes of journalistic writing against the affectivity of the quotes, the commentaries (including columns) and tabloid editorials included more affective vocabulary than the general level for the quoted content discussed above. The analyses and editorials of Helsingin Sanomat, in contrast, aligned with the level of affectivity we generally found in quotes. Because they contained few quotes, these formats contrasted interestingly with the news stories, in which more than half of the overall content came from an even split of direct and indirect quotes. In these news articles, in contrast to the other article types, we also noted a stark difference between the proportion of affective vocabulary inside and outside of quotes. The intertwining of the rituals of objectivity and emotionality was evident: news articles strove to be perceived as factual, calm and non-partial, while also seeking to engage readers and to represent the tensions and passions of the reported event.
To this end, they featured multiple sources and selected quotes with high affective intensity, but they counteracted this intensity in quoted language by toning down the emotive intensity of the journalistic voice well below the overall averages.

The effect of external factors
In terms of the temporal dynamics of the competitiveness pact reporting, all news outlets followed the same patterns, with their news output peaking in response to real-world events (Figure 4, showing only the sum across sources). Notably, while this overall distribution was dominated by news stories, the distributions of the other types of story also individually mirrored this trend without any statistically significant differences.
To examine whether the external intensity of the events had an effect on language, we used the news output frequency to split our competitiveness pact corpus into three parts: (1) peaks, where at least ten articles were published across the sources daily (which are visible and named in Figure 4); (2) post-peaks, which covered the 2 days after a peak; and (3) non-peaks. We also gathered a reference corpus of articles from the politics section of Helsingin Sanomat from a random period at the end of 2016 with no specific news peaks. In comparing these corpora to each other, we found that the tone of the journalistic portion of the articles relating to the competitiveness pact did not differ from general political journalistic writing. It also did not differ between peaks and non-peaks.
While the journalistic language was not significantly altered by peaks, we detected a difference in the proportion and affectivity of the quotes that were used. While the quantity of direct quotes remained the same, their affectivity diminished during peaks to 8.5%, from The line depicts a left-aligned three-day rolling mean used for smoothing, while points denote actual daily counts. a non-peak median affectivity of 9.3%. With indirect quotes, in contrast, the affectivity stayed the same (at 7.6%), but their overall proportion within the articles increased at the cost of non-quote content. These changes may be interpreted as reflecting the fact that peak news reporting is heavily source-based as well as being primarily focused on stating the events as they occur, with little emphasis on interpretation or framing.
Finally, regarding the 'click-bait' hypothesis, the concentration of metaphor markers and hedging structures was observed to be somewhat higher at the start of the articles than in later sections. Counter to our expectations, however, the evaluative adjectives showed the reverse behaviour and were more frequent in the middle and end sections. Analysing this remains a question for further study. When comparing these tendencies across the sources, we were unsurprised that the tabloid Iltalehti had the fewest zero-affectivity beginnings, thereby corroborating, to some degree, the click-bait hypothesis. However, Iltalehti also had a considerable number of articles with no affective vocabulary, thus suggesting that there, too, the hypothesis could explain only a specific subtype of articles. Examining this further would require a closer examination of the news category to distinguish the feature articles, personal interviews and viewpoint articles that currently disappear in the mass and which may be outliers. To do so, we might use the indicators developed in the present study to further cluster the articles inside the news category in order to identify the subtypes.

Conclusion
Our aim in this article was to develop an approach for studying affectivity as a fundamental feature of news journalism: a question of form and structure, rather than intentional stance-taking, emotive storytelling or evaluative expression. Drawing from conversation analysis (Goffman, 1961), Bakhtinian theory of language as dialogical (Volosinov, 1995) and Wetherell's (2012) notion of affective meaning-making, we developed an approach that allowed us to chart how different linguistic forms and structures participate in affective meaning-making in news journalism.
Methodologically, we sought a computational approach that could be scaled up to allow the study of big data sets and, in the future, long-term developments in news journalism. The approach we developed rests on a flexible technical base (Janicki et al., 2020) combined with an iterative meta-approach to build indicators that capture the phenomena of interest . The analysis is modular: each part can be adapted to new material or discarded, and new parts added. For example, to port this analysis to other corpora would require tuning the quote detection and article categorisation rules, as well as refiltering the evaluative adjective list. Once defined for a particular corpus, such rules would however work mostly automatically for any analysis targeting that corpus. The metaphor markers, on the other hand, are case-specific, and would require at least some tens of manually annotated examples to extract for each individual use-case. Our study nevertheless shows that a highly specified quantitative approach is able to yield new knowledge by both providing novel observations and affirming at scale patterns recognized in previous research. Even if details of the approach need to be amended for other datasets and questions, the basic methodological ideas are widely applicable.
In our attempt to revise how affect is conceptualised in news journalism, we focused on emotive and evaluative vocabulary, which is the most prominent linguistic strategy for affectivity-related functions. The close-reading part of our analysis revealed that most of the words used for affective meaning-making derived their impact from metaphoricity. In this genre, metaphorical transference proved to be such a consistent strategy that the results from this indicator mostly mirrored those from the evaluative adjectives.
In using the indicators for emotive and evaluative vocabulary and for modal expressions, we examined both structural and external influences in a data set that incorporated four different types of news outlet: a leading national newspaper, a public-service news source, an evening tabloid and a news agency widely published in national and regional news outlets. Some of the findings were predictable, such as the positioning of news-agency and tabloidnews journalism at the extremes, with public-service news journalism as a non-distinct average (cf. Stenvall, 2014). More interestingly, in the case being studied, which exemplified heightened political and labour market conflict, the affectivity of journalistic language appeared no different from political journalism in general. However, there were differences in the amount and affectivity of the quotes used between peaks and non-peaks of political activity. Importantly, our findings consistently support the idea of the rituals of objectivity and emotionality being coterminous and interdependent (Tuchman, 1972;Wahl-Jorgensen, 2013a, 2013b but also point to interesting variations in how affectivity is managed between the different article types with the use of the different types of quote and with hedging. Indeed, the accuracy with which the linguistic markers were able to differentiate between the external variables we examined was striking. In other words, the codes of the different established journalistic genres across all news outlets seemed to have distinctive and highly contextualised conventions for modulating affectivity, governed by their own communicational purposes. The fact that affectivity is something that these latent conventions seem to address is clear evidence of the centrality of affect in journalistic practices. Together, these findings suggest that studying affectivity as not only encompassing emotive and evaluative adjectives, but also highlighting the importance of metaphorical nouns and verbs and evidential and epistemic modal structures, offers insights for further study of journalistic forms, styles and conventions (Broersma, 2007), the language of news (Bednarek and Caple, 2012) and the narrative politics of news (Peters and Broersma, 2016). Finally, the emotional turn in journalism studies has highlighted emotional labour as a professional skill and requirement of journalists (Kotišová, 2019;Richards and Rees, 2011;Wahl-Jorgensen, 2019). With our approach, we have demonstrated that this affective linguistic labour -the many forms and structures that mediate and modulate affectivity -extends beyond hiding and disguising subjective emotions.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/ or publication of this article: This work was supported by