CODING ANGER IN FICTION

The paper examines the fictional representation of anger in English. The emotion is identified by its lexical indication in the reporting clause (said angrily) accompanied by the direct speech which is assumed to verbalize the emotion. The reporting clause and the direct speech are viewed as the components of an anger-coding unit which is examined in terms of their mutual position and the syntactic and speech-act properties of the direct speech component. Analysis and correlation of these variables (and the length of the direct speech) suggest that anger-coding units display recognizable patterns with characteristic structure and illocutionary profile.

The question of how an emotion gets expressed by linguistic means is an intriguing one.For the purposes of this paper a very distinct type of emotion, anger, was chosen and a seemingly easy way to explore it: in fiction where it is vocalized by direct speech (stylized dialogue).When we are angry it is manifested by our mien, voice or gestures and we articulate the reasons for being angry by what we say.The problem is that the written language does not have extralinguistic and prosodic means of expression at its disposal.Hence the symptoms of anger must be graphically described or the writer may simply state that the fictional character is angry.The study of anger in language may thus examine how the phenomenal aspect of anger is verbally described or it may examine the verbal articulation of (the reasons for) anger by a given person, which in fiction is done through direct speech.The present study attempts the latter while recognizing that it is the writer's stylized literary approximation of what people actually say.

Anger-coding units: definition and data collection
The selection of angry utterances for analysis had to deal with two problems, their location in text and finding criteria for their description.The problem of locating an angry utterance was resolved by looking for the sequence "said angrily" in a corpus (other adverbs, such as "furiously", were excluded for the sake of homogeneity and a manageable size of the sample).This led to instances of direct speech that due to this sequence could be safely considered "angry utterances" as conceptualized by the authors.Since both the reporting clause including the lexical anger marker ("said angrily") and the direct speech (the angry utterance proper) are part and parcel of coding anger in fiction, they were both included in the analysis and treated as one unit.
The investigation thus focused on anger-coding units made up of two components: a reporting clause ("rc" for short) containing the sequence said angrily and a direct speech component which presumably expresses the speaker's anger.Both the reporting clause and the direct speech are clearly distinguishable from the previous and the subsequent text and relatively self-contained.The hypothesis was that direct speech components conveying the speaker's anger will primarily include (a) emotion-charged expressive illocutions, and (b) sentence types likewise associated with emotions (exclamative and imperative sentences).
The source of the material for analysis was William Fletcher's database PIE (Phrases in English) based on the British National Corpus.After weeding out hits that did not conform to the specification -the sequence said angrily must be part of a reporting clause introducing direct speech -the sample included 152 anger-coding units.

The position of the reporting clause and the length of the direct speech
On closer examination it turned out that the anger-coding units are characterized by two basic interrelated features which distinguish between them: the position of the reporting clause and the length of the direct speech.The reporting clauses with said angrily occur in all three possible positions (rcp): before direct speech (rcp1), inside direct speech, bisecting it into two parts (rcp2), and after the direct speech (rcp3): initial (rcp1): Lambert said angrily."Not in this squadron, never." medial (rcp2): "It's not a proposal, " she said angrily."It's -it's an attempt to buy me." final (rcp3): "Come on, tell me!" Gog said angrily.
Table 1 shows that anger-coding units with mid-position reporting clauses are the commonest, occurring in almost 60%; those with initial reporting clauses are the least frequent (12.5%), and those with final reporting clauses are somewhere in between.Table 1 also indicates that the direct speech component (utterance) may consist of more than one sentence, especially with rcp2 utterances in which one sentence typically precedes and another sentence follows the reporting clause (rather than there being just one sentence divided by the reporting clause).The last column in the table shows the difference between the number of utterances and the number of sentences the utterances consist of.There seems to be a direct relation between the reporting clause position and the total length of the direct speech component expressed in the number of sentences.In the initial and final reporting clause utterances the direct speech component is largely composed of a single sentence.In both rcp1 and rcp3 utterances more than one sentence occurred in only 15.8% and 17.7% respectively, although there is no obvious reason why it could not be otherwise.The rcp2 utterances are bisected into two separate parts which again tend to consist of only one sentence just like rcp1 and recp3 utterances (part 1 departs from this tendency in 1.1%, part 2 in 6.8%).Hence the number of sentences in anger-coding units with rcp2 is more than twice the number of the rcp2 utterances (88 utterances, 183 sentences).In the few cases of rcp2 utterances where it was not clear whether part 1 and part 2 should be seen as two sentences or two clauses of one sentence, the decision was made by reference to illocution.Where clauses in what was formally one compound sentence had distinctly different force, they were treated as separate sentences.
Since the position of the reporting clause appears to correlate with the length of the direct speech, and the length of the direct speech component seems to be connected with its pragmatic makeup, as we shall see, the reporting clause position is apparently not a negligible factor.So, even at this point, rcp1, rcp2 and rcp3 anger-coding units can be seen as different anger presentation patterns.

Syntactic analysis of the direct speech component
In the next step, the analysis concentrated on the key component of the anger-coding units, direct speech.It was subject first to syntactic analysis, then to the pragmatic analysis of the speech acts (illocutions) that realize it, and finally to the analysis focused on the correlation between the syntax and illocutions.
Syntactic analysis looks at the distribution of the basic sentence types in the direct speech components.The hypothesis was that they would be typically realized by sentence types with interactive and emotion-prone functions, i.e. exclamative and imper-ative sentences, rather than by the primarily neutral declarative sentences.It turned out to be convenient to distinguish one specific formal type in addition to imperative, interrogative and declarative sentences, namely sentences (called here "other") consisting of verbless, non-finite and formulaic structures, characteristic of the spoken language.Their functions range from expletives to exclamations, and they are generally emotion-laden.
Sentence type examples in the direct speech of anger-coding units: • Declarative: "It's not a proposal, " she said angrily.
• Imperative: He said angrily, dismissively: "Don't be stupid." / "Shut up!" said Ray Shepherd angrily.• Interrogative: "What do you think I am, anyway?"Gazzer said, angrily / … he said angrily."Did you expect a present, Miss Eyre?" • Other: "Of course not!" she said angrily, reddening./ "For God's sake, " I said angrily, / "Tomorrow?"Gog said angrily./ "Silly young fool, " Dickie said angrily.However, quite contrary to expectations, the predominant sentence type in the whole sample is, as Table 2 shows, the declarative sentence (61.2%), while the remaining three types, the category other, imperative sentences and interrogative sentences are roughly equally represented and form slightly more than one third.Interestingly, this ratio appears to be connected with the length of the direct speech component and the reporting-clause position.In the two-part/two-sentence angry utterances with mid-position reporting clauses the dominance of declarative sentences is very strong (two thirds), while in the initial reporting clause utterances this ratio is reversed, the directly interactive types, interrogative, imperative and other sentences, prevail here over the declarative sentences.
In the rcp3 utterances the interactive types and the declarative sentences are balanced.The findings suggest that the shorter the direct speech component (with rc positions 1 and 3), the greater the tendency to use sentence types associated with direct, emotion-prone impact.Still, in all three rcp patterns of angry utterances the proportion of supposedly neutral declarative sentences is very high.

Pragmatic analysis of the direct speech component
The most interesting, but at the same time, the most difficult part was the pragmatic analysis striving to describe the direct speech component in terms of speech acts, i.e. its illocutionary force.Searle's (1979) taxonomy of speech acts, which provided the basic guidelines, is unfortunately very general and, as far as we know, there are very few detailed taxonomies of illocutions that could be adapted for our purposes.Ronan's (2015) interesting categorization of expressive speech acts which as she claims are particularly underresearched partly overlaps with the categories used in this research.
Examination of the sentences in the direct speech components of the sample revealed as many as 26 distinct secondary illocutionary forces.In order to bring the number to a manageable level while retaining useful semantic distinctions, the 26 illocutionary concepts were subsumed under nine categories that correspond to the following four types of the Searlian classification: (i) representatives -(1) belief, (2) explanation; (ii) directives -(3) direct coercion, (4) indirect coercion, (5) query; (iii) commissives: -( 6) self-coercion (promise), ( 7) disagreement; (iv) expressives -(8) expressivity, (9) dissatisfaction.The belief category comprises illocutions such as assertion, claim, confirmation and conviction, which become anger-filled in circumstances when the speaker is questioned, doubted or contradicted.The category explanation appears -with only four exceptions -in the second part of the bisected rcp2 utterances.Explanation serves as a complement to some of the other illocutions as will be shown later.The three coercion categories describe cases when the speakers emphatically enforce their will either directly (order, demand, urging, forbidding) or indirectly (advice, warning, threat) or this enforcement is self-imposed as a promise.The association of the category query with anger is context-dependent and marginal.The category disagreement always involves some kind of negation, i.e., denial, refusal, rejection, or disagreement proper.The expressivity category covers all kinds of short utterances realized by verbless or non-finite clauses or formula-type structures.Their shortness makes the precise specification of the illocution difficult, their purpose is mainly and only to express a strong negative emotion.Comparison of this category with the other illocutions suggests that the computation of the specific nature of an illocution may depend on the length of the utterance -the longer the utterance, the easier it may be to identify it, and vice versa.Dissatisfaction subsumes complaints, accusations, criticisms, as well as reproaches and irony, in which the expression of an emotional load seems to be the main point.

The distribution of illocutionary categories
The overall distribution of illocutionary categories is given in Table 3.The most frequent illocution, accounting for a quarter of all instances, is dissatisfaction which can be thus viewed as the main cause of anger.The second most frequent illocution in the sample is remarkably explanation which in itself does not imply anger.Obviously the function of explanation is quite different from that of the other illocutions: it typically accompanies another illocution which is related to anger.The next three most frequent types of illocution associated with anger are direct coercion, disagreement and belief, closely followed by the expressivity category.The remaining three categories are marginal.Comparison of the distribution of illocutions within the three presentation models, rcp1, rcp2 and rcp3 (see Table 4), indicates that the frequency ranking of the three most frequent illocutions is very similar in utterances with undivided one-sentence direct speech components: dissatisfaction in the first position, expressivity in the second and direct coercion alternating in the first and second position.The only difference is disagreement in the first place in rcp1 utterances which comes only third in rcp3 utterances.Table 4 also shows that relatively close to this frequency ranking of illocutions is that of the first part of the rcp2 utterances, with dissatisfaction and disagreement at the top.By contrast the distribution in the second part of these utterances is very different: here explanation is the most frequent illocution of all.This fact led us to correlate the illocutions of the first and the second part of the medial rcp2 utterances.The results of the correlation, given in Table 5, revealed interesting patterns of pairing between the first part illocutions and the second part illocutions: most frequently it is dissatisfaction or disagreement which is followed by explanation, less frequently it is direct coercion, belief or expressivity which is subsequently explained.The other frequently occurring patterns are those of the same illocution extended into the second part of the direct speech: dissatisfaction-dissatisfaction, direct coercion-direct coercion and belief-belief.Anger-related illocutions (in one or both parts) thus come either without explanation or in two parts the second one of which contains an explanation: the first illocution describes the source of anger (dissatisfaction, and so on) and the second gives explanation why the speaker feels that way.
Examples of the correlated illocutions in rcp2 utterances: Dissatisfaction-explanation: "Giles is an idiot, " he said angrily."He could have contacted me if he had tried." Disagreement-explanation: "No, you didn't, " said Peter, angrily."It was my idea." Direct coercion-explanation: "Put some fucking clothes on, Scott, " he said angrily."You're under arrest." Repeated illocutions: (irony-irony) "Thank you, " he said angrily."I hope you'll enjoy reading about this in tomorrow's papers." / (order-order) "Get back!" he said angrily."Don't touch me!"

Correlation between the sentence type and illocution
Finally the interesting question was whether there is any correlation between the sentence type and illocution.Table 6 shows the findings.Given the number and frequency of the sentence types and the number of illocutions in the sample, they are not entirely unpredictable.Although typically a neutral, emotionally unmarked type, declarative sentences which form over 60% of the direct speech text in the sample realize eight of the nine categories of illocutions.Not only do they express dissatisfaction (25.3%), explanation (24.7%) and belief (16.5%), they also signal disagreement, indirect coercion and promises.The remaining sentence types are more specialized: the category other most frequently conveys expressives (37.5%) and disagreements (27.5%), imperative sentences as might be expected express direct coercion (73.5%) and interrogative sentences are mostly used to show dissatisfaction (57.7%).The fact that a single sentence type, declarative sentences, is used to realize all but one type of illocution and that 5 out of the 9 categories of illocu-tions in the sample are realized by all four sentence types suggests that formal expression, namely the sentence type and its concomitant features such as the mood, word order, etc., are not the decisive or main indicators of illocution, i.e.IFIDs.Obviously, anger in direct speech is conveyed primarily by pragmatic means, drawing on the semantic content of the proposition, rather than by sentence type.Anger is thus expressed implicitly unlike in reporting clauses containing angrily.

Conclusions
The paper deals with one selected signal introducing an anger coding unit.Although in the whole of the BNC there were "only" 152 instances of this particular kind ("said angrily"), there is no reason to believe that it is in any respect an unusual way of expressing anger in literary text.It seems justified to speak of a pattern (cf.Hunston, 2010) because the anger-conveying stretch of text forms a unit clearly delimited and recognizable in text; it has a predictable and fairly fixed structure, its parts exhibit distinct tendencies in their syntactic and pragmatic makeup and length.In other words anger-coding units are fairly uniform in spite of being gathered from many books by many different authors.
The anger-coding unit is made up of two components: the reporting clause and the anger-bearing direct speech.The unit is identified by the sequence said angrily in the reporting clause.Anger-coding units divide into three groups or patterns according to the position of the reporting clause which either precedes or follows the direct speech or divides it into two.The position of the reporting clause and the length of the direct speech appear to be rather constant: undivided direct speech components are typically one sentence long, bisected direct speech components are made up of two sentences.
The paper focused especially on the direct speech component and its syntactic and pragmatic properties.Contrary to the initial hypothesis, by far the most common sentence type in the direct speech components is the declarative sentence accounting for almost two thirds of the sample.There was no formally distinct exclamative sentence in the sample.The remaining one third is comprised of "other" sentences (non-finite, verbless or formulaic), imperative sentences and interrogative sentences.The sentences were found to express 26 different kinds of illocutions which were divided into nine categories.Declarative sentences realize eight of the nine categories of illocutions identified in the text.The other three sentence types, on the other hand, did show a high degree of specialization.
All the same, it is possible to say that there is a weak correlation between the illocutionary profile of the direct speech component and its sentence type structure.The fact that declarative sentences may express most of the anger-related illocutions means that the sentence type is in itself not a sufficient IFID.When coding anger in fiction, the author does need the lexical trigger (said) angrily in the reporting clause to help the reader correctly calculate the illocutions in the direct speech component in the absence of extralinguistic and prosodic indicators of anger.
The three groups or patterns of anger-coding units characterized by the position of the reporting clause and the length of the direct speech component (one-and two-sentence) display typical illocutionary profiles: the single-sentence initial reporting clause model is associated primarily with dissatisfaction and disagreement to the same degree, the single-sentence final reporting clause pattern with dissatisfaction and direct coercion (i.e., order, demand, urging, forbidding), and the two-sentence medial reporting clause model typically combines two illocutions, dissatisfaction and explanation in the first and the second sentence respectively.Incidentally, the findings underscore the fact that Searle's distinction of five speech act types is not very helpful in the analysis of actual utterances as it is too broad and that there is a considerable asymmetry between speech act types and the communicative sentence types.
Naturally, it would be interesting to know whether similar patterns of coding anger in fiction occur in other languages.However, an attempt to collect a comparable sample of anger-coding units in Czech encountered several difficulties.Not only does the English verb to say have two equivalents in Czech, říci and pravit, but unlike English reporting clauses in which the verb to say is by far the most preferred one Czech reporting clauses use a much wider range of verba dicendi (some of which incorporate the semantic feature of anger).As a result it is difficult to find a sufficient number of instances with the reporting verbs řekl(a) or pravil(a), matching the English said.In order to get the same number of anger-coding units in Czech fiction at least ten more verbs (such as zvolat, vykřiknout, prohlásit, odseknout, zavrčet, začít, zasyčet, zařvat, zasyčet, utrhnout se, etc.) would be required.Similarly in a comparable Czech corpus it is impossible to find the same number of instances of one Czech adverb matching the English angrily.Instead it would again be necessary to include synonyms, more than fourteen of them, to get the same sample size (zlostně, vztekle, rozzlobeně, nazlobeně, rozčileně, naštvaně, rozhněvaně, nahněvaně, rozezleně, dopáleně, dožraně, namíchaně, navztekaně, hněvivě, etc.).The use of different verbs and adverbs can be expected to influence the illocutionary profile of the direct speech component and so influence the results of comparison with English.Also the different principles of word order in Czech which allow for a variable position of the adverb with regard to the verb complicate the compilation of a reasonably homogeneous corresponding Czech sample.
To conclude, judging from the sample it appears that the language of fiction has developed a relatively stable and conventionalized pattern of coding anger in English, with typical formal and pragmatic features.By a combination of direct indication (lexical trigger) and indirect means (a range of illocutions) it overcomes the absence of context with prosodic and extralinguistic (facial, gestural) signals which communicate anger in spoken language.

Aleš Klégr Department of English Language and ELT Methodology
Faculty of Arts, Charles University ales.klegr@ff.cuni.czPavlína Šaldová Department of English Language and ELT Methodology Faculty of Arts, Charles University pavlina.saldova@ff.cuni.cz

Table 1 .
The reporting clause position and direct speech length figures (* i.e., one sentence per utterance ratio or one sentence per rcp2 utterance part 1 or 2 ratio)

Table 3 .
Overall distribution of illocutions in the reported speech

Table 4 .
Illocutions in the first four positions according to frequency

Table 5 .
Correlation between the illocutions of part 1 and part 2 of rcp2 utterances

Table 6 .
Correlation between sentence type and illocution