EMOTION AND OPINION IN ONLINE TABLOIDS AND BROADSHEETS

This paper is a corpus-based study focusing on implicit evaluation expressed in newspaper discourse, namely, the semantic mapping of emotion and opinion. The corpus, compiled of online “front page” newspaper articles from both selected tabloids ( The Sun , The Ex-press , The Mirror ) and broadsheets ( The Telegraph , The Guardian , The Independent ) was used to answer three research questions: 1) Is evaluation markedly expressed in newspaper discourse? 2) What linguistic means are typical for construing evaluation in newspaper discourse? 3) Is there a difference between the tabloids and the broadsheets regarding the way in which/how evaluation is conveyed/employed? To answer these questions, a pilot keyword study on only six articles was carried out (one article from each of the aforementioned newspapers). The findings confirmed the importance of adjectives in expressing evaluation. Following this, a large study was conducted to detect local grammar adjectival lexicogrammatical patterns, introduced by Hunston (2000) and further amended by Bednarek (2007, 2009). These patterns, which are known for carrying the evaluative load, were analyzed in terms of frequency and function. It was observed that there is a difference in expressing evaluation between the tabloids and the broadsheets. However, more significant differences were found between the broadsheet newspapers themselves.


Introduction
Newspapers report on significant events, both locally and globally, and there is a general expectation that they will be both objective and independent. However, the reality is often different as a newspaper's journalists, editors, and commentators are human, with feelings and attitudes that on occasion override their responsibilities. Newspapers, intentionally or not, express their opinions and emotions towards the events reported, either via direct or subtle linguistic means. Sometimes, opinions and emotions are expressed implicitly, hidden between the lines, at other times explicitly, mainly through expressive lexis although other means can also be employed. Many factors contribute, such as the newspaper itself -broadsheet/tabloid, right/leftwing, state/privately-owned, the newsworthiness of the events reported (Fowler 1991) and the discursive news values (Bednarek 2016: 27). Linguists such as Lemke (1998), Thomson and Hunston (2000), Conrad and Biber (2000), Hoey (2001), Martin and White (2005), and Bednarek (cf. 2006Bednarek (cf. , 2007Bednarek (cf. , 2008Bednarek (cf. , and 2009) have focused on identifying and defining the means by which attitude/evaluation/stance/appraisal are expressed in the language present in various discourses. Lemke (1998), for example, researched evaluative parameters such as importance, usuality, normality, etc., in editorial texts. He emphasizes that evaluation, which acts as a critical function of a language, should be investigated further in all discourses (Lemke 1998: 20). Hunston (2011) explores evaluation from different corpus perspectives, focusing on grammar patterns, local grammar, and evaluation. Her main concern is distinguishing explicit evaluation via certain grammatical patterns. She observes that "while a few patterns with adjectives might be said to perform evaluation, most report evaluation and some do so in a multilayered way" (Hunston 2011: 143). Conrad and Biber (2000) consider stance in speaking and writing discourse, while Martin and White (2005) introduce their appraisal theory, which is considered to be both systematic, and indeed unique, within what could be viewed as the fuzzy field of evaluation. Bednarek (2006) examines evaluation in media by employing a number of Lemke's parameters, but adding her own specific categorization of core and peripheral parameters. She concludes her research by stating that the British newspapers (tabloids and broadsheets) mainly construe the parameter of evidentiality, explaining that, based on her data, "the differences between the two types of the newspaper are smaller than to be expected from our preconceptions" (Bednarek 2006: 203). In her subsequent book, Emotion talk across corpora (2008), she addressed the emotion/affect embedded in the British Register Corpus (BRC) and BRC baby, and later analyzed evaluation in TV blurbs (Bednarek 2014). More recent work includes a consideration of language patterns and attitude (Su and Hunston: 2019), as well as mapping evaluation appraisal and stance in discourse (Li et. al: 2020), with the authors noting that though such research was infrequent at the beginning of the century, it significantly increases after 2008.
The current paper aims to achieve two main goals. Firstly, it intends to identify the explicit 1 expressions of emotion and opinion in British online broadsheets and tabloids (a pilot study, based on a small-size corpus, Section 3.1). It also hopes to 1 The explicit meaning, the so-called inscribed meaning, is directly coded in lexis, e.g. "The children were rudely talking", as opposed to the implicit meaning (which will not be analyzed in this chapter), the so-called provoked or evoked meaning, which has to be decoded as the lexis itself is not evaluative, as in "The children talked while he was presenting the lesson" (Martin and White 2020). establish what lexical means tend to be used to construe the given categories and determine the differences in their use in British online tabloids and broadsheets. Secondly, it focuses on adjectival lexicogrammatical patterns employed in the newspapers and analyzes these in terms of opinion and emotion (a large corpus analysis, Section 3.2). The frequency findings are further investigated to determine the similarities and differences in the newspapers' tendencies to embed emotion and opinion in their news. The specificity of this research lies in the comparable size of the tabloid and broadsheet subcorpora used in the analysis of the adjectival lexicogrammatical patterns, which is relatively large, with 339 broadsheet and 402 tabloid articles. Previous research by Bednarek (2007) used a much smaller sized corpus (70,300 tokens in total, 37,404 for the broadsheets and 32,796 for the tabloids). Alba-Juez (2017) uses a corpus of 200 articles in total, which means the results can be considered more reliable. Another positive aspect of our research is the synthesis of Hunston and Sinclair's (2000) findings on evaluative adjectival lexicogrammatical patterns with those of Bednarek.
Even though the introductory paragraph has mentioned the terms evaluation, stance and appraisal several times, this paper's analyses also employ the terms opinion and emotion. These terms are defined in the following section, based on Martin and White's (2005) appraisal theory. However, the contributions of Bednarek (2008Bednarek ( , 2007Bednarek ( , 2009 and Hunston (2011) are also taken into consideration.

Delineation of the salient terms
The terms evaluation, appraisal and stance are used more or less interchangeably, and there is no "agreed-upon conception of these terms among researchers" (Li et al. 2020: 33). Despite such multiplicity, the following paragraphs aim at determining the nuances between the terms.
Evaluation is a broad term that can be viewed from different perspectives, and still, the definition will be incomplete. It can be viewed as "a marked case against the background of the normal and expected" (Daneš 1994: 253), it can be placed on the binary scales of good vs bad, lovely vs hideous (Downes 2000: 104), or in a more complex approach, it is considered a "term for expression of the speaker's and writer's attitude or stance towards, viewpoint on, or feelings about the entities or propositions that he or she is talking about" (Thompson and Hunston 2000: 5). Thompson and Hunston (2000: 6) further develop the concept by defining other functions of evaluation such as construing and maintaining relations, as well as organizing discourse as the writer and the reader are aware of "the boundaries in the discourse and the nature of the connection between its various parts" (Thomson and Hunston 2000: 12).
Appraisal is viewed as a functionally oriented approach to evaluation, which encompasses all the evaluative means of language. Martin and White (2005: 33) believe that appraisal is "one of three major discourse semantic resources construing interpersonal meaning", resources including the domains of involvement and nego-tiation, as well as appraisal. All these aspects are systematized in the concept of the appraisal theory, which is discussed further later in this section.
Similarly to evaluation and appraisal, the definition of stance is broad and varies depending upon the researchers' interests. Biber and Finegan (1989: 93) define stance as "the lexical and grammatical expressions of attitudes, feelings, judgements, or commitment concerning the propositional content of a message. " However, it is common that two researchers "working on similar issues may use different terminology to cover what appears to be the same type of phenomena" (Englebretson, 2007: 2). A researcher from their own perspective can view stance as evaluation, another as appraisal, and yet another could refer to the instances of interpersonal language as subjectivity.
As this paper investigates the semantic dimensions of emotion and opinion in newspaper discourse, it will now be explained how the terms were established and upon what they draw. The terms were created based on the appraisal theory introduced by Martin and White (2005), who built upon Halliday's Systemic Functional Linguistic Framework. Martin and White perceive appraisal as a system of three central issues, which speakers or writers adopt to create a positive or negative stance towards the given information. The central issues are created by attitude (feelings, judgement, and appreciation), engagement (direct or reported speech, monologue or dialogue), and graduation (emphasizing or obscuring the information), as can be seen in Figure 1. These three central issues are both complex and interwoven.  (Martin and Rose 2003: 54) This paper only examines attitude, where, for the purposes of this paper, affect, judgement and appreciation are divided into the semantic categories of emotion (affect, see example (1), which deals with real human feelings; the example suggests that the person is probably sad, and that is why he cried) and opinion (judgement, see example (2), usually connected with the evaluation of human behaviour, stances, and actions; and appreciation, see example (3), which comments on aesthetic issues). This division is based on the suggestions raised by Bednarek (2009: 172). The author is aware that the categories cannot have clear-cut boundaries; however, after reviewing the relevant resources (e.g. Hunston and Sinclair 2000;Martin and Rose 2003;Bednarek 2007), these two dimensions seemed the most suitable for the pur-poses of the analyses. Yet, it needs to be emphasized that no dual coding is used when running the analyses, with each detected item assigned to a single category, despite the fact dual coding could be legitimate.

Methodology and corpora
The study consists of two analyses. First, a pilot study, "the small corpus analysis", was carried out to ascertain the range and classes of evaluative linguistic means used and identify whether emotion or opinion are more embedded in news reporting. The pilot study was followed by "the large corpus analysis", focusing on adjectival evaluative patterns and the dimensions of emotion and opinion. For the purposes of clarity, it should be stated that the analysis does not aim to distinguish who expresses a particular instance of opinion/emotion, whether it is the actual author of the article or the participant of the reported event, even though it is understood that this could have a significant impact on the results. The analyses were intended to identify in general terms whether, and to what extent, emotion or opinion are directly embedded in the researched discourse. Distinguishing between reporting and expressing an opinion/emotion is possible, yet it is not the goal of this research.
Each analysis required a different corpus and different methods. However, in both cases qualitative and quantitative analyses were conducted. The two corpora are described in terms of size and text selection in the following sections, together with further, methodological decisions regarding both analyses.

The small corpus
The small corpus consists of articles downloaded from six British daily online tabloid and broadsheet newspapers. The newspapers chosen are The Sun, The Daily Express, and The Daily Mirror for the tabloids, and The Independent, The Guardian, and The Daily Telegraph for the broadsheets. Two subcorpora were created, each consisting of six articles. An article on the same issue on the same day was downloaded from each of the chosen online newspapers. This procedure was applied to reduce topic-related differences between the individual newspapers to a minimum (see Bednarek 2006). The Qatada corpus contains articles on a Muslim cleric who was allegedly involved in terrorist acts. Britain wished to extradite him to Jordon, but this was impossible due to human rights considerations. He remained in Britain, supported by the state. The Baby corpus comprises articles collected on the day Prince George was born. 2 The two corpora (negative × positive) were analyzed separately in order to identify any different tendencies in the use of emotion or opinion in the given newspapers. Table 1 represents the quantitative characteristics of the small corpus. Since the subcorpora are not of the same size, normalized frequency (per 1,000 words) is used to compare the results from the tabloids and the broadsheets. To conduct the small corpus analysis, the concordancer AntConc 4 was used to help extract the data. The analysis of evaluation expressing means in the small corpus was executed by running a keyword analysis and detailed qualitative analysis, employing the categories of emotion and opinion. Two types of comparison were performed. Firstly, the broadsheet newspaper articles were compared with those from the tabloids. Secondly, the Qatada subcorpus of negative news was compared with the positive Baby subcorpus. The small corpus study indicated that keyword analysis was not suitable for identifying differences in the use of evaluative lexical expressions between the two subcorpora because the differences were not statistically significant. The small corpus size, however, allowed a detailed qualitative analysis of evaluation in the texts studied. Although the results confirmed the assumption that various classes of lexical words are evaluation-loaded, adjectives are the means that incorporate evaluation the most frequently.Therefore, they were chosen as a starting point for the large corpus analysis.

The large corpus
Based on the small corpus analysis results, it was decided to focus specifically on adjectives. The decision was taken for two reasons; firstly, they were the most frequent part of speech employed to express opinion and emotion in the pilot study; secondly, both Hunston and Sinclair (2000) and Bednarek (2009) had previously investigated evaluation-laden adjectives, so the large corpus analysis is based on the occurrence of the adjectival evaluative lexicogrammatical patterns established in their research. (iii) Comparative o: …it will be much harder to achieve that…  The software programme AntConc was used to detect the analyzed patterns in the corpus. Unfortunately, the programme does not morphologically tag the texts as with other similar programmes/platforms, e.g. Sketchengine and Lancsbox. Therefore, the part-of-speech tagger Q-Tag was used. For each pattern presented in Table 2, a specific query was written, which was used to search for potentially evaluative constructions employing adjectives in the tabloid and broadsheet subcorpora. The detected results were sorted manually to select those instances of each pattern that served an explicit evaluative function. The results for each pattern in the broadsheet and tabloid subcorpora were further classified according to whether they expressed emotion or opinion. The type/token ratios were calculated and compared across the subcorpora where relevant. The tabloid and the broadsheet subcorpora were subsequently compared in terms of the patterns and lexical items most frequently used in each subcorpora and the corpus as a whole.

Small corpus results
Figure 2 shows how opinion and emotion were expressed in the relevant subcorpora. The Qatada subcorpus indicates the prevalence of opinion expressed in both subcorpora (tabloid and broadsheet). It was presumed that the tabloids would rely more on emotion than opinion, yet this assumption was not confirmed, with an expression of opinion still playing a central role in newspaper articles. It is primarily construed by nouns (e.g. terror, threat, disgrace) in both the corpora. Further, the broadsheets more commonly employ adjectives (e.g. fair, frustrating, angry) and then verbs (e.g. torture, roam), while the tabloids use mainly verbs and then adjectives. See examples (4), (5), (6).
(4) …, we believe he is a threat to our country. (Tab) (5) It is extremely frustrating and … (Broad) (6) …, Qatada is allowed to roam the streets for eight hours… (Broad) Emotion, as expected, is more frequently expressed in the tabloids. Perhaps surprisingly, adjectives are not the most prominent means to convey emotion, with verbs (e.g. fume, smirk, blast) appearing almost three times more frequently than adjectives, see examples (7), (8), (9). The tabloids also make use of nouns, which interestingly are not applied in an attributive role as premodifiers, but primarily in the Vn pattern. Adjectives apparently are the least important means to construe emotion in the tabloids. On the other hand, the broadsheets primarily rely on adjectives, with the other means employed to a lesser extent. In fact, emotionally-laden verbs rarely appear. The results of the positive Baby subcorpus show completely different findings. The tabloid newspapers have a greater tendency to report on positive news, with almost 30 instances of evaluative items, both opinion and emotion, per 1,000 words, more than are detected in the broadsheets. The most prevalent part of speech employed to express any evaluation of positive news are adjectives, followed by verbs in both kinds of newspapers. Notably, nouns do not play an important role in expressing positive news. However, the tabloids express opinion almost twice as frequently as the broadsheets, mainly by exploiting adjectives (e.g. warm, wonderful, and special). Emotion is similarly embedded mainly via adjectives in both broadsheets and tabloids (e.g. grateful, proud, delighted). The overall findings are displayed in Figure 3. Opinion is more frequently expressed in the newspapers than emotion, particularly when reporting on negative news. Adjectives are the category most likely to construe evaluation, yet it is worth mentioning that verbs also express both opinion and emotion. The tabloids are aware of this fact and use them successfully: it would be interesting to undertake a more detailed study on the evaluation of loaded verbs. A positive evaluation is expressed reluctantly; the negative prevails. In addition, it is vital to emphasize that lexical items are not used in isolation; they often create clusters, and these clusters, as well as individual words, are highly context-dependent.

Large corpus results
There were 1,179 occurrences of all the listed patterns found in the corpus, with 620 detected in the broadsheets, and 559 in the tabloids. The results indicate that evaluative adjectival constructions are slightly more frequent in the broadsheets, however, the difference in the total numbers of evaluative adjectival patterns between the two subcorpora is not statistically significant 6 (this applies to both the non-graded and graded patterns). Figure 4 visualizes the occurrence of all the patterns across the large corpus in normalized frequency per 10,000 words. Pattern 4 v-link ADJ prep is the most frequent pattern, followed by pattern 3a v-link ADJ toinf, and pattern 1a it v-link ADJ finite/non-finite clause. These three patterns seem to play a prominent role in expressing the dimensions of opinion and emotion in newspaper discourse. Figure 4 also displays the non-occurrence of patterns 1d it V n ADJ that and V it as ADJ /V it ADJ. It would be interesting to attempt to determine whether these patterns are used to construe evaluation in discourses other than newspaper discourse. The frequency of the graded adjectival patterns were limited in the analyzed texts, but seen more often than patterns 1b, 1c, 5 and 6. Peldová (2017: 157-158) states that the graded evaluative adjectival patterns can eventually be unified with the basic patterns introduced by Hunston and Sinclair (2000), and further amended by Bednarek (2009 Pattern 4 v-link ADJ prep seems to be the most prominent pattern for expressing opinion and emotion in this newspaper discourse study. It appears 7.0 times per 10,000 words in the broadsheets and 9.0 times per 10,000 words in the tabloids. As mentioned above, the pattern itself can express both opinion and emotion, however, the tabloids seem to employ it mainly for emotion. The most frequent adjectives used in this pattern to embed emotion are concerned, worried, and proud in the tabloids, with the broadsheets, in addition to these adjectives, employing the negative lexical expression unhappy. Via these lexical expressions, the newspapers aim to suggest insecurity, unhappiness and the level of satisfaction of the agents of the reported news. If TTR (type-token ratio) analysis is applied, we can see that despite the pattern being used the most frequently in the corpus, the lexical richness is limited (146 tokens, 39 types, TTR 27%), which highlights the fact that both types of newspapers tend to use the same adjectives. As the tabloids mainly use this pattern to promote emotion, they have a tendency to use, apart from personal pronouns, first names (e.g. John, Patrick), proper nouns and noun groups (e.g., cops, family, parents) in the role of the evaluator, while the broadsheets include groups who have the power to influence the country (e.g. investors, conservatives, lawyers). The difference could be assigned to the nature of the reported news in the respective newspapers. Examples of emotion can be seen in (14) and (15). Pattern 4 is also used to construe opinion, to a much lesser extent but still relatively frequently throughout the corpus. The adjectives primarily utilized by this pattern are responsible, good and guilty. Interestingly, the lexical density determined within the semantical dimension of opinion is higher than observed within emotion (119 tokens, 66 types, TTR 30%).
Does this mean that the dimension of emotion has a narrower repertoire than opinion? This is possible, taking into consideration that the dimension of opinion in this study covers the dimensions of appreciation and judgement in Martin and White's terminology (2005). Examples (16) and (17)  it v-link ADJ finite/non-finite clause Pattern 1a is a complex pattern that includes many varieties, yet it is a pattern that is solely used to express opinion; no instances of emotion construed by this pattern were found in the corpus. Finite and non-finite clauses can complement the pattern. The non-finite dependent clauses that follow the it v-link ADJ core part of the pattern are to-infinitives (20), (21) and gerund -ing complementation (18), (19). Thatclauses and interrogative dependent clauses represent the finite clauses. The gerund -ing complementation is rare in the corpus, with only two instances detected. However, the to-infinitive was found in 107 cases, representing around 44% of all the complementation. The finite clause complementation (with all its variants) was detected in 55% of all the complementation. Further analysis shows that the that-clause complementation is used the most frequently (117 occurrences, representing 48% of all complementation) of the finite-clause complementation. Therefore, to-infinitive clauses and that-clauses are the two most frequent variants of this pattern. A qualitative analysis focusing on whether these two variants indicate differences in the content reveals that the to-infinitive in the tabloids tends to embed information on the good-bad scale using adjectives such as right, interesting, wrong, wonderful, unfair (21), (22). At the same time, the broadsheets rely on the dimensions of important and good, using adjectives such as right, important, vital, and appropriate (23). A significant fact is that the broadsheets tend to report that something is possible, while the tabloids eliminate such an approach. The broadsheets utilize finite and non-finite complementation more or less equally; therefore, no versions is marked, while a thatclause is marked in the tabloids. A possible reason could be an editorial decision, because to-infinitive non-finite clauses tend to be shorter than full that finite clauses. In this pattern opinion does not rely upon a wide range of adjectives, and also the lexical diversity is not relatively low at 23%. Overall, the main adjectives embedded in this pattern are clear, right, and important, see (22) and (23). An interesting finding is that the adjective clear, the most frequently utilized adjective in this pattern within the corpus, plays a more important role in the broadsheets and is not included in the to-infinitive variant. In other words, the that-clause complementation is frequently linked with the adjective, whereas there is no evidence of this construction in connection with the to-infinitive. Further, interrogative dependent clauses (if /whether, wh-clauses, how-clauses) only follow when the adjective unclear or negated adjective clear is used (24), (25). This is a possible area for further research. To conclude this section on pattern 1a and its evaluative load, the tabloids use it to mainly pursue what can be termed the goodness and badness of the information with a relatively high TTR (44% and 50%), while the broadsheets focus more on epistemic stance, mainly evaluating clarity, possibility, and importance. A more detailed description of this pattern and the adjectives used can be found in Peldová (2016Peldová ( , 2017

v-link ADJ to-inf
The third most frequent pattern utilized in the online newspaper discourse under analysis is pattern 3a. Unlike pattern 1a, this pattern expresses the semantic dimensions of both emotion and opinion. Hunston and Sinclair (2000: 87) even distinguish two types of opinion, one that is concerned with judging particular behaviour (26), (27) and a second that evaluates objects (28). However, in this study, this division was not taken into consideration. The broadsheets use this pattern mainly to construe opinion (almost three times more often than emotion), while the tabloids use this pattern to create emotion; however, the difference between the occurrence of the dimension of opinion and that of emotion is not statistically significant. This finding supports the results from the pilot study, revealing that the broadsheets tend not to employ expressions of emotion in their news reporting. The most frequent adjectives used to express emotion in the tabloids are sorry (and reluctant). The broadsheets make use of keen and reluctant, which suggests that the broadsheets tend to express dis/inclination towards something (29); the tabloids, apart from inclination (27) As mentioned above, opinion is more frequently expressed in the broadsheets than in the tabloids. Both types of newspapers embed epistemic modality in their news/reports by expressing uncertainty and likelihood when applying this pattern. They use adjectives such as likely, unlikely, and certain. Interestingly, the adjective likely is the most prominent adjective in the whole corpus, followed by clear and right.

Opinion vs emotion
This section attempts to answer questions concerning the differences observed in the employment of opinion and emotion in the tabloids and the broadsheets by means of the adjectival evaluative pattern. The research proves that there is a statistical difference between the subcorpora in the expression of opinion and emotion. Not unexpectedly, when compared to the tabloids, the broadsheets reveal a considerably greater use of the semantic dimension of opinion. However, the most significant difference is not between the subcorpora, but within the subcorpora (see Figure 5). It was assumed that the tabloids would employ more emotion than opinion due to the intrinsic nature of these newspapers, yet the findings underline that the difference between the semantic dimensions of opinion and emotion is not as significant as in the broadsheets. Although the semantic category opinion is expressed 1.5 times more frequently than emotion in the tabloids, the broadsheets rely to a far greater extent on opinion. Nevertheless, the influence of informalization can be detected as the broadsheet news reports also express emotion. As mentioned in the sections on individual patterns, the pattern it v-link ADJ finite/non-finite clause is exclusively employed to construe opinion, with emotion carried by the patterns v-link ADJ prep and v-link ADJ to-infinitive. The most frequent adjectives used to express opinion are the same across the whole corpus, i.e. likely, clear, right, important, and unlikely. Emotion, when expressed, is mainly carried by the adjectives concerned and sorry. The tabloids also include positive emotion via the adjectives happy and proud, this positive load is not typically seen in the broadsheets.

Conclusions
The main aim of this paper was to identify what means online newspapers use to express evaluation and what differences exist in construing evaluation between the tabloids and the broadsheets. To this end, the paper focused on evaluative adjectival patterns and their utilization in online newspaper discourse. The objective was to determine whether these patterns were employed to express the semantic dimensions of opinion or emotion, drawing on Martin and White's appraisal theory. A pilot analysis on the small corpus indicates that explicit evaluation is indeed expressed in the given discourse, and that the most common means of evaluation are adjectives, nouns and, significantly, verbs. It was concluded that in both kinds of newspapers, evaluation tends to occur more frequently in negative news.
Based on the small corpus analysis findings, a decision was made to focus on adjectival patterns in the large corpus analysis. The study affirms that there are close relations between certain lexicogrammatical adjectival patterns and the expression of evaluative attitude in the language of online newspapers. The large corpus analysis shows that newspaper discourse tends to use several adjectival evaluative patterns, such as the v-link ADJ prep, v-link ADJ to-inf, and it v-link ADJ finite/non-finite clause. It was found that there was a statistically significant difference between the representation of emotion in the tabloids and the broadsheets. These differences are apparent in the selection of evaluative adjectives. The tabloids tend to use a wide range of emotion adjectives, such as happy, concerned, sorry, and proud. On the other hand, the broadsheets primarily rely on the adjective concerned. Furthermore, the large corpus indicates a preponderance of the opinion evaluative expressions in both subcorpora; adjectives construing the semantic mapping of clarity, importance, and epistemic stance prevail in both types of newspapers, primarily due to the high representation of the adjectives likely, clear, right, and important.
To sum up, the study confirms that adjective evaluative patterns are embedded in news discourse. The most notable difference is not in the frequency of the patterns in both subcorpora but in their function. It has also been shown that the local grammar pattern approach to the detection and description of evaluative meaning in newspaper discourse is helpful and could be potentially fruitful in further research.