Corpus evidence for lexical and genre effects in the metaphorical conceptualization of negative self-evaluative emotions: The case of shame and embarrassment

Abstract

Do different emotion terms trigger different metaphorical conceptualizations of emotions? What are the effects of the discourse context of the genre on metaphor choice in the conceptualization of emotion concepts? Finally, are such lexical and discourse–contextual effects on emotion-targeted metaphor choice quantifiable? Prior discourse-oriented research has demonstrated from a largely qualitative perspective that metaphor use is dynamic and sensitive to discursive contextual variables (e.g., Deignan et al., 2013; Semino 2010, 2011; Semino et al., 2013; Dorst 2015; Caballero 2016; Knapton & Rundblad, 2018). In the present study, these questions are addressed from a corpus-based multivariate perspective, where detailed qualitative analysis of found examples is combined with quantitative modeling. The study examines negative self-evaluative emotions in English, operationalized through their two nominal exponents, i.e., shame and embarrassment, as attested in the discourse context of three genres – fiction, magazine and spoken TV language. The data are first analyzed qualitatively for relevant contextual variables and then modelled quantitatively. The results demonstrate that while both lexical and genre effects are observed in metaphor choice in the conceptualization of negative self-evaluative emotional experience, their combined effect should also be accounted for, as these two variables are found to interact with each other.

1. Introduction

The present study is a corpus-based qualitative and quantitative investigation of the metaphorical conceptualization of two negative self-evaluative emotions in American English. More specifically, the concept under examination is operationalized through its two nominal exponents, shame and embarrassment. Two objectives are pursued here, one methodological, the other descriptive. Methodologically, the study seeks to combine qualitative corpus analysis with multivariate modelling, thus further advancing quantitative figurative language research (cf. Nordmark & Glynn, 2013; Ogarkova & Soriano, 2018; Reijnierse et al., 2019; Rogos-Hebda, 2020; Glynn & Biryukova, 2022). The descriptions thus produced are not only empirically verifiable but also generalizable, being representative of the sociolinguistic context under analysis. Descriptively, the aim is threefold: (i) to map the semantic structure of the two negative self-evaluative emotion terms relative to the metaphorical source domains in the context of three genres. While doing so, the study seeks to identify possible (ii) lexical effects and (iii) genre effects on metaphorical source domain choice in the process of conceptualizing negative self-evaluative emotional experience. In other words, the questions to be addressed here focus on whether there should emerge any lexically-dependent variation in the metaphorical profiling of the emotional experience under investigation and whether the different genres selected should be found to exploit different types of conceptual metaphors.

There is good reason to expect that the answer to these research questions should be affirmative. Prior discourse-oriented studies (e.g., Skorczynska & Deignan, 2006; Semino, 2010, 2011; Semino et al., 2013; Dorst, 2015; Caballero, 2006, 2016; Knapton & Rundblad, 2018; Reijnierse et al., 2019) have shown that contextual variables such as genre or register do influence metaphor use with regard to the frequency of occurrence as well as formal and functional features (Semino et al., 2013: 41). It is therefore only natural that, as Steen (2008: 221) puts it, the metaphor should be studied ‘as part of actual language use, or events of discourse”, where apart from its linguistic instantiation and conceptual structure, it also always performs a particular “communicative function’. The communicative function performed by metaphor is intricately linked to the discourse context of its occurrence, such as genre or register, speaker type (cf. Berger & Jäkel, 2009) and the targeted audience (Deignan et al., 2013: 47).

Given the above points, the hypothesis here is that variation in conceptual metaphor distribution across genres and lexemes will be observed, thus demonstrating that metaphor choice is partially determined by the lexical exponent of the emotional experience as well as by the discourse context of a given genre. With regard to genre-related effects on metaphor choice, fiction and magazines are expected to show greater abundance and heterogeneity in metaphors than the section of spoken discourse. This is based on prior findings concerning metaphor distribution across registers (e.g., Reijnierse et al., 2019), where the discourse of fiction or news has been found to be considerably richer in metaphors than conversation. A further motivation for this hypothesis comes from the general genre-induced differences in emotion communication between fiction and casual conversation observed by Bednarek (2008). Regarding lexical effects on metaphor choice, it is hypothesized that the conceptual metaphors observed for shame and embarrassment will give rise to different representations of the emotional experience of negative self-evaluation, with the primary difference having to do with the perceived level of severity (in terms of both the longevity of the emotional experience and its detrimental effects for the experiencer). Should such an effect be observed, it would be attributed to the more serious and long-lasting nature of shame, as postulated in psychology (see Section 2.1). The expectation of lexical effects on metaphor choice for encoding emotional experience, even when expressed through closely related near-synonyms, is further motivated by supporting evidence obtained in prior research. For example, Ogarkova and Soriano (2018) have empirically demonstrated lexically-induced variation in the metaphorical conceptualization of anger.

In what follows, we first discuss previous research on negative self-evaluative emotions and their metaphorical profiling (Section 2). Next, we turn to the presentation of the method (Section 3), which is followed by a detailed discussion of the data, qualitative analysis and study limitations (Section 4). Finally, the quantitative results regarding lexical and genre effects on metaphor choice are presented and discussed (Section 5).

2. Theoretical background

2.1. Negative self-evaluative emotions

Before we consider the metaphorical structuring of negative self-evaluative emotions, let us examine the notion itself as well as the selected target concepts affected by the structuring. Self-evaluative emotions, also referred to variably as social, self-conscious or self-regulatory emotions (e.g., Barret, 2005; Beer and Keltner, 2004; de Hooge et al., 2011; Lewis, 1995, 2008, 2011; Taylor, 1985; Zinck, 2008) constitute complex cognitive constructs which presuppose self-awareness as well as self-representation and the capacity for self-evaluation from the perspective of an actual or imagined other (e.g., de Hooge et al., 2011: 198). This evaluation, in turn, requires an appreciation of culturally determined social standards and norms, whose potential violation may lead to irrevocable damage to the perception of the self (e.g., Lewis, 1995). Self-conscious emotions, such as shame, embarrassment, guilt, humiliation, regret or pride, are critical to the regulation of how we behave and feel and, ultimately, what we think (e.g., Tracy & Robins, 2004: 103). In fact, these emotions are so fundamental to our existence that “we are virtually always in a state of either pride or shame” (Scheff, 1988: 399). In other words, the underlying motivating and modulating force in our daily existence is the prospect of possible shame and our desire to avoid it.

In the present study, it is the emotion of shame alongside its close associate, embarrassment, that we inspect. It is this close relation between the two that was the primary reason for choosing these emotion terms. In fact, the two emotions are so closely intertwined that embarrassment is often considered but a ‘low-level version of shame’ (Barret, 2005: 955), representing a transient and fleeting emotion. Embarrassment is perceived as dependent on the presence of a judgmental audience and as directly linked to a specific triggering situation (Tangney et al., 1996:1258; Lewis, 2008:750). The antecedent events of this emotion normally concern ‘self-presentation issues or minor violations of social rules’ (Barrett, 2005: 955). Shame, on the other hand, is viewed as the ‘grimmer and weightier’ (Tangney et al., 1996:) of the two emotions. It is linked to a ‘sense of devalued and subordinated self’ as a consequence of contravening some implicit social norms (Gilbert, 2003:1222). It represents what is sometimes phrased as ‘a global attack on the self’, which can be experienced either privately or publicly (Lewis, 2008: 748), with quite dire consequences for self-perception, including the experiencer’s desire to ‘hide, disappear, or even die’ (Li et al., 2004:768).

2.1.1. Linguistic treatment of social emotion concepts

Social emotions have received considerable attention in linguistics, particularly within corpus-based approaches, with synchronic and diachronic studies, intra- and inter-linguistic inquiries and investigations into both positively and negatively valenced self-evaluative emotions. A line of investigation into self-evaluative emotions that has paved much of the future empirical avenues of research within linguistics is Wierzbicka’s Natural Semantic Metalanguage (NSM). Relying on the semantic primitives proposed in her NSM, Wierzbicka (1992a:133; 1992b: 574ff.; 1999: 110ff.) captures the very essence of the concepts of shame and embarrassment. In the most general terms, they are described as having to do with the experiencer not wanting other people to think something bad about them. There is, however, a fundamental difference between the two emotions, which has to do with how shame, unlike embarrassment, is likely to be morally grounded and to exert long-lasting consequences on self-evaluation (Wierzbicka, 1999:110).

Another line of research is pursued by Krawczak (2014a, b, 2015, 2018) or Kumamoto (2019, 2020), where detailed qualitative/discourse analysis is combined with statistical modeling to map the cross-cultural structure of negative self-evaluative emotion concepts. Among the findings that are relevant here, Krawczak (2014a, b, 2015, 2018) establishes a clear distinction between the situations that are encoded through the adjectival exponents of shame and embarrassment. More specifically, ashamed, as opposed to embarrassed, is found to refer to more grievous antecedent events, whose negative impact on self-evaluation may be long-lasting and may well appear in complete solitude. These findings provide an empirical basis for our hypothesis that the metaphorical conceptual profiles of shame and embarrassment will be different, with more negative source domains (e.g., involving actual or potential physical damage and destruction) being associated with the former emotion term.

Other methodologically parallel studies include Glynn (2014a) or Soares da Silva (2020). The former focuses on a basic emotion concept, i.e., anger, but is concerned specifically with its social structuring in British and American English, while the latter investigates the cultural conceptualization of pride in European and Brazilian Portuguese. Yet another line of corpus-based research into social emotions is followed by Lewandowska-Tomaszczyk and Wilson (2014) and it involves collocational analysis of shame and guilt in English and Polish. Among the results that are relevant here, the authors find that shame in English is frequently associated with such collocates as blushing, shamefaced, die, cover, great, bloody, damn, deepest (Lewandowska-Tomaszczyk & Wilson, 2014: 135f.). Even though the study does not focus on metaphorical conceptualization, the above collocations might be expected to be linked to such source domains as container (deepest), disease (die), or the conceptual metaphor identified in Tissari (2006) shame is in the face (blushing, shamefaced).

Finally, noteworthy are also elicited-data-based investigations into the cross-cultural conceptual structure of various emotions (e.g., Ogarkova et al., 2010; Fontaine et al., 2013a; Ogarkova & Soriano, 2018; Soriano & Valenzuela, 2022). Some of these studies employ the GRID questionnaire-based method (e.g., Silfver-Kuhalampi et al., 2013 on shame and guilt), while others use triangulation of methods, also involving corpus-based analyses (e.g., Ogarkova & Soriano, 2018). The GRID instrument measures the probability of the contribution of a wide range of properties (e.g. appraisals, responses) to the meaning of frequent emotion words (Fontaine et al., 2013b: 2). For example, in a GRID-based study contrasting shame and guilt as constructs averaged across 34 diverse cultures, operationalized through the translational equivalents of the English terms shame and guilt across the various languages, shame was found to be associated with semantic components such as (i) the tendency to retreat; (ii) relative unpredictability; (iii) relative powerlessness; or (iv) negative valence (Silfver-Kuhalampi et al., 2013: 392f.). In another study employing the GRID method and examining the Dutch emotion term schaamte corresponding to English shame and embarrassment combined, against the data for US English shame, the Dutch lexeme was found to be linked to features that would be considered more typical of embarrassment such as (i) unintentionality of the trigger; (ii) unexpectedness; (iii) lower levels of controllability; and overall (iv) less negatively valenced experience (van Osch et al., 2013:489).

2.2. Conceptual metaphor theory

Theoretically, the present study employs what arguably belongs to the most widely researched linguistic theories, the conceptual metaphor theory. It has revolutionized our understanding of the role and importance of metaphor in language and thought. Popularized in the seminal work of Lakoff and Johnson (1980) and further enhanced by researchers such as Kövecses (e.g., 1986, 2000, 2005, 2015, 2020), this framework perceives metaphor not only as a figure of speech (as in bathed in shame, from example (15)), but primarily as an underlying pattern of thought (e.g., shame is liquid in a container), where an abstract target domain (e.g., shame or emotion) is conceptualized in terms of another more concrete and readily comprehensible domain, the source domain (e.g., liquid). Thus understood, conceptual metaphors have manifold possible instantiations in all manner of expressive behaviors, be it verbal or non-verbal, which, as with all human activity, are context-sensitive. The CMT has been subject to some criticisms, with major problems involving insufficient inductive evidence and inadequate accuracy in domain classification or specification. While a detailed discussion of the intricacies of the CMT goes well beyond the purview of this article, some of its challenges are discussed in Section 3.1 from the perspective of the problems encountered while conducting the analysis in the present study.

2.2.1. Metaphorical profiling of negative self-evaluative emotions

Since the focus of the present investigation is on the metaphorical structuring of negative social emotions, let us discuss prior research in this regard. The studies that have inspired much research in this area are Kövecses’ (1986, 1990) detailed qualitative analyses of pride as well as his (1998, 2000) discussion of shame in present-day American English. Among the studies that followed and which considered negative social emotions from a historical perspective, we find Tissari (2006) or Fabiszak and Hebda (2007). The former study examines the evolution of the metaphorical structure of shame in English on the basis of corpus data spanning a period of over 500 years (1418–1991). Fabiszak and Hebda (2007), in turn, reconstruct the metaphorical scenarios for two negative social emotion lexemes, i.e., shame and guilt, in Old English. Table 1 presents the source domains identified in these studies for shame, with color coding for convergent findings. In addition, the metaphors that have also been identified in the present study are underlined.

Table 1. Source domains for the conceptual metaphors of shame attested in prior research

Most of the metaphors identified for shame are not peculiar to this emotion concept, with the exception of (i) blocking out the world, (ii) decreasing in size, and (iii) having no clothes on, as proposed by Kövecses (2000:147). Interestingly, the degree of convergence between the source domains identified in the three studies is not substantial. The only two commonalities observed across all three studies have to do with some physical ailment (illness, physical injury or damage) and financial gain or loss (economic value, expense, paying for one’s misbehavior, valuable commodity). Naturally, such a relatively low level of overlap may be attributed to the underlying differences (both cultural and historical) in the source material and its respective size, which ranges from present-day American English (Kövecses, 1998, 2000), through late middle, through early modern to present day English (Tissari, 2006), to Old English (Fabiszak & Hebda, 2007). Notwithstanding, as we shall see in Section 4.2, many of the categories identified by Fabiszak and Hebda (2007), Tissari (2006) or Kövecses (1998, 2000) are also attested in the present-day American English data employed here, which is indicative of the continuous currency of these metaphorical conceptualizations of negative self-evaluative emotional experience in English.

It is noteworthy that Kövecses (1998, 2000) in the source domains proposed for shame relies on an interview-based inquiry by Holland and Kipinis (1994), whose focus was on metaphors for embarrassment in American English. Based on the survey, these authors identified five main metaphorical conceptualizations of embarrassment in terms of exposure to the public eye, having no clothes on, desire to escape, desire to disappear, being of low status or being small (Holland & Kipinis, 1994). In addition, they found metaphors that were related to either the incidents that engendered the emotional experience or the physiological consequences thereof.

It should also be pointed out that the methodological approach employed in the present study (see Section 3) has not been widely applied to the analysis of conceptual metaphor, in general, and metaphorical conceptualizations of emotions, in particular. The studies that do employ the method for metaphor analysis include: Nordmark and Glynn (2013) on metaphorical construal of anxiety; Ogarkova and Soriano (2014, 2018) on anger in English, Russian and Spanish; Glynn (2018, 2022) on the quantification of contextual effects on conceptual metaphors of anger; Glynn and Biryukova (2022) on boredom in English and Russian; Rogos-Hebda (2020) on immigration in early American magazines; and Krawczak (2022) or Kumamoto (2022) on the metaphorical representation of social emotions.

3. Methodology

To attain the objectives indicated in Section 1, the present study employs usage-feature analysis which is the first step of the behavioral profile approach, a method developed within Cognitive Linguistics (Geeraerts et al., 1994; Divjak, 2006a, b, 2010; Divjak & Gries, 2009; Glynn, 2009, 2010, 2014b; Janda & Solovyev, 2009). Usage-based analysis entails careful manual qualitative analysis of a large number of corpus examples for relevant features, which is followed by quantitative modeling of the annotated metadata. The latter type of analysis enables us to uncover systematic linguistic patterns and, a fortiori, their underlying conceptual structure.

In this methodological framework, regardless of the object of study, the data are retrieved on the basis of preselected keywords or morphosyntactic patterns, which in the case of conceptual metaphor research may constitute a challenge. This is because conceptual metaphors represent mental structures and as such afford no direct observational access (cf. Stefanowitsch, 2006: 64). This challenge can be resolved through rigorous operationalization relying on language use. More specifically, data retrieval may focus on the lexical exponents associated with either the target concept or the various source concepts responsible for the mapped metaphorical structure (Stefanowitsch, 2006). In the present study, following Stefanowitsch’s (2006) conceptual metaphorical pattern analysis approach, it is the target domains that were selected. They are operationalized through their nominal exponents, i.e., shame and embarrassment. The choice of the grammatical category of noun was motivated by the apparent greater amenability of nouns (possibly, next to adjectives) to metaphorization (e.g., Reijnierse et al., 2019).

Methodologically, the present study also follows the linguistic Metaphor Identification Procedure (MIP) developed by the Pragglejaz Group (2007) and further advanced by Steen et al. (e.g., 2010). Similarly to Stefanowitsch (2006), the MIP procedure takes the lexical unit as the central point of orientation in metaphor recognition and it can be summarized as follows (after Steen, 2016: 79):

1. (i) determining the overall discursive meaning of a given text and its lexical units;

2. (ii) identifying lexical units used outside their basic usage scope, where basic corresponds to ‘more concrete’, ‘more precise’ and ‘historically older’ uses;

3. (iii) establishing whether the contextual meaning of a lexical unit can be juxtaposed with its more basic meaning and whether the former can be interpreted on the basis of comparison with the latter.

4. (iv) If the answer should be affirmative, a linguistic metaphor has been identified.

The analysis in the present study, nonetheless, does not stop at identifying metaphorical expressions. This is the first stage, which is then followed by an attempt to group the myriad of linguistic metaphors into superordinate conceptual categories that represent domains posited as the sources of structural mappings to our target domains. This process of classifying the identified linguistic metaphors is not devoid of problems, which are addressed in Section 3.1.

3.1. Identification of conceptual metaphors: Methodological challenges

This section focuses on some of the problems in conceptual metaphor identification that have also been encountered in the present study. One major challenge in conceptual metaphor identification, which has already been pointed out by Glynn (e.g., 2018; Glynn & Matusevich, 2016, 2017), is that it is not always straightforward to decide whether a given metaphorical expression is a linguistic realization of an established conceptual mapping or whether it represents a creative instance. This problem is illustrated by example (1) and possibly (4a):

Here, the lexical unit shades of are clearly used figuratively, but it remains unclear whether it is an instance of a more general entrenched correspondence between embarrassment and a specific source domain. It could perhaps be linked to the source domain identified in Tissari (2006) shame is in the face, in which case it might be taken to be based on a metonymic mapping related to the physical symptoms of negative self-evaluative emotions, i.e., blushing. This would not be a surprise, since, as Goossens (1990:323) notes, metaphor and metonymy “may be found in combination in actual natural language expression”.

Another difficulty encountered in the annotation procedure followed here concerned examples for which two or even three related source domains seemed possible. In line with the Conceptual Metaphor Theory (e.g., Lakoff & Johnson, 1980), source domains should represent discrete categories, and yet when confronted with actual language usage, the question of source domain classification is not always an either-or case (see also Glynn, 2006). It is not always straightforward to determine whether a given instantiation is a member of one source domain or perhaps another adjacent category. At times, it may be a matter of granularity, a taxonomic question. This issue with domain delimitation has already been pointed out by Croft (2002), Glynn (2006) or Steen (2016: 77). It is illustrated by such instances as those in (2a)-(2e) or (3a)–(3c).

In examples (2a)-(2d), between two and possibly even four related metaphorical source domains seemed possible, representing either adjacent categories (dirt, stain, mess) or different levels of specificity, with substance being a more generic superordinate concept. Ultimately, based on metaphors identified in prior research (cf. Table 1), the category of dirt, which has been attested in Tissari (2006) and Fabiszak and Hebda (2007), was selected in the qualitative analysis for the examples where the substance in question had negative connotations.

In (3a) and (3b), the choice of disease and injury, respectively, for the source domains evoked in the conceptualization of negative self-evaluative emotions in these examples seems relatively straightforward, given the linguistic indices of infect in (3a) and festering in (3b). In (3c), in turn, pain appears the most obvious choice, and yet pain is part of the frame associated with injury or disease, the latter being the most general. Ultimately, then, the choice is again taxonomic.

Finally, another challenge encountered in the domain classification had to do with situations where two metaphorical expressions belonging to distinct conceptual source domains were observed simultaneously, as in (4):

Examples (4a) and (4b) illustrate instances with two distinct metaphorical expressions used to refer to the target emotion concept, i.e., a course of (linguistic metaphor, possibly instantiating motion?) and acute (pain) in (4a) and the sting of (pain) and rising (liquid) in (4b). In such cases, it was the expression closest to the keyword that was selected for annotation. Another possible solution would be to annotate such examples twice for the two metaphors separately.

4. Data and qualitative analysis

4.1. Data

The data in this study were extracted on the basis of the two selected keywords from the Corpus of Contemporary American English (Davies, 2008). More specifically, given one of the research questions concerning the influence of genre on conceptual metaphor choice, the observations were sourced from three components of the corpus, representing three different genres, namely, fiction, magazine and spoken discourse. Importantly, the two lexemes represent the most frequently attested lexical instantiations of shame and embarrassment in the three genres from which the examples were drawn. ¹ Data extraction involved a manual search of all the metaphorical uses in a sample of 600 observations per lexeme in the three genres. This procedure resulted in a sample of 286 metaphorical uses of shame and embarrassment. The data summary regarding the distribution of observations across the three registers and the two lexemes is provided in Table 2.

Table 2. Data summary

In the final sample, 22 different types of metaphorical source concepts were identified, nine of which have also been attested in prior research (see Tables 4 and 5 for details). The procedure for source domain identification was discussed in detail in Section 3, while Section 4.2 is a detailed overview of the source domains.

Table 3. Distribution of embarrassment and shame across the genres

Obs. Inspected: Observations inspected for semantic relevance to the study, i.e., encoding emotional experience, with the subsample size established based on the power analysis with the confidence of 90% and p-value of 0.05.

Relevant obs.: Observations estimated to be semantically relevant (with the confidence of 90% and p-value of 0.05) from among all the found observations, based on the results of the subsample analysis of inspected observations, where 73%, 70% and 58% of the examples from fiction, magazine and spoken TV discourse, respectively, were identified as encoding emotional experience.

Table 4. Raw frequencies of the conceptual metaphorical source domains across registers

Table 5. Raw frequencies of the conceptual metaphorical source domains for the two lexemes

4.2. Qualitative data analysis: Source domains classification

This section discusses and exemplifies the qualitative data analysis. Five source domains (stain/dirt, substance, disease, injury, pain) have already been considered in Section 3.1 and so are excluded from the discussion below. The remaining source domains are discussed in alphabetical order. The lexical units that were identified as indicative of metaphoricity are highlighted. In addition, further examples of lexical expressions associated with the source domains are provided.

The domain burden has been found to provide a metaphorical source of structural mappings for shame by Kövecses (1998, 2000) and Fabiszak and Hebda (2007). In the data forming the basis of this study, other lexical indices of burden included the following expressions: carry, bear, crumble from the weight of, attached to, tied with, lay down the bricks of or burdened with. Burden conceptualizations focus on how psychologically inescapable or, figuratively speaking, heavy the negative self-evaluation associated with embarrassment and, in particular, shame may be for the experiencer.

The next set of three source domains was marked by low frequency in the dataset (cf. Tables 4 and 5) and is excluded from the quantitative modeling. None of them seem to have been identified in prior research:

The temperature-related source domains of cold and heat were combined due to the overall low number of observations that instantiated these concepts. It might be argued that they are linked to possible physiological symptoms of the emotions and thus represent metonymic rather than metaphorical mappings. However, as already pointed out, the two figures of speech and thought may well coincide in use. It is also unclear whether the cold- and heat-related sensations are actually physically experienced, or whether they are brought up by the speakers as domains of comparison. The domain of confinement accentuates the fact that negative self-evaluative emotions trap the experiencer, blocking normal social functioning. It is noteworthy that the desire to escape encoded in this metaphor corresponds closely to the model of the two negative social emotions proposed in psychology, where the ‘desire to escape or to hide – to sink into the floor and disappear’ is emphasized as an important characteristic (Tangney et al., 1996: 1257). Beyond the expressions highlighted in blue in sentences (6)–(8), the following lexical expressions were observed for the three source domains: tremble with, chill of or sweaty with (cold/heat); be sentenced to (confinement); escalated into (conflict).

Sentence (9) illustrates the source domain container, which has already been attested by Tissari (2006). It represents a generic level concept ², comparable to those of object or person, which is not clearly associated with any specific mappings, other than that of a concrete ontological status it affords for the emotion concept. In addition to its combination with the adjective deep, in the present study, container was also found to be realized through the adjective profound as well as the prepositions in and out of.

Cover and crime, exemplified in (10) and (11), respectively, both represent relatively infrequent metaphorical source domains in the dataset. The former concept is related to Kövecses’ (1998, 2000) more specific domain of having no clothes on, as clearly evidenced in sentence (10). However, there were also other lexical indices that resulted in a more schematic category than that proposed by Kövecses. These expressions included: the prepositions over, under and on, or such phrases as a shred of, a net of, or a dense cloak of. Crime, in turn, was also linked to the verb cover up, which additionally points to the idea of hiding one’s misdeeds.

The next source domain revealed in the analysis – danger – as illustrated in (12) does not appear to have been priorly attested in the context of negative self-evaluative emotions. It emphasizes the undesirability of public exposure linked to negative self-evaluative emotions and the potential destructive or even deadly consequences. Through this, danger is related to the domains of disease, injury or pain. In addition to the verb loom, it has also been associated with such expressions as prevent, spare, save oneself/someone from, protect someone from, be braced for, skid away from, avert, expose someone to or befall someone.

The source domain of fire, exemplified in (13) below, has been previously found to be associated with shame by Tissari (2006). In this study, beyond the verb burn with, it has also been linked to such phrases as burn in, burn someone up, spark or flare up. This source domain is closely linked to the temperature-oriented domain of heat. The connotations, however, are substantially different, with fire being linked to ultimate annihilation.

The source domain force has not been previously attested in the context of either concept. The emphasis in this conceptualization is on the uncontrollable and potentially destructive nature of the experience, as evidenced in (14). It has also been instantiated through such phrases as drive someone, be driven by or fueled by.

liquid has been found to be one of the more frequent source domains in the metaphorical conceptualization of negative self-evaluative emotions. It has been identified in both Kövecses (1998, 2000) and Tissari (2006). This source domain, whose applicability scope is not shame-specific, but rather encompasses any emotion concept, is closely linked to the container concept. The expressions revealed as its lexical instantiations included rise in someone, fill someone with, a wave of, steam with, be full of, a wash of or pouring,

The next, generic-level source domain, i.e., object, was the most frequently attested one in the data. Interestingly, while reification does not seem to have been identified as a general metaphorization process, Kövecses (1998, 2000) and Tissari (2006) do propose more specialized cases of this general source domain, i.e., incomplete object and unwanted entity or substance, respectively. Among the lexical exponents of objectification, the following phrases have been found: hide, conceal, cover, export, bring, take, take away or such adjectives as big, huge.

The source domains of opponent and person, as illustrated in (17) and (18), respectively, are closely related, with the former being a more specialized instance of the latter concept. Neither appear to have been priorly attested. Among phrases associated with these two domains we find battle, overcome, face, stand up to or win out (opponent) and come, go, be gone, join, survive or silent (person).

Finally, the last set of three source domains, illustrated in (19)–(21), are among the less frequent source domains attested in the dataset. Only one of them, i.e., precious object, has been revealed in previous research (Kövecses, 1998, 2000; Tissari, 2006). This domain happens to be the least expected of all the domains, given its positive connotations. In addition to the expressions highlighted in the examples below, these domains were also realized through such phrases as wide, distance from, beyond, a cell (of) (place), grow and reap, (plant), and be worth, afford, be the cost or earn (precious object).

Among the metaphors that have been priorly identified in the literature, but which have not been attested in the data, we find the following: Kövecses’ (1998) (i) being worthless, (ii) being a divided self, (iii) blocking out the world; (iv) decrease in size; (v) hiding from the world; Tissari’s (2006) (vi) obstacle; (vii) shame is in the face; and Fabiszak and Hebda’s (2007) (viii) amending shame is paying for the misbehavior; (ix) darkness and (x) shame is down. This, on the one hand, may be due to a relatively small data sample. We may have well found many more metaphor types if we were to have accounted for more observations. On the other hand, the operationalization of the concept of negative self-evaluative emotions on the basis of the two selected nominal exponents means that we could not have found many of the metaphorical linguistic expressions related to shame proposed by Kövecses (1998, 2000), including I felt this big or I wanted to bury my head in the sand, for the source domains listed in (iv) or (v), respectively (from Kövecses, 2000: 32). With regard to Tissari’s metaphor shame is in the face, in this study, any instances referring to facial symptoms of the emotion such as blushing or being red in the face were treated as more likely candidates for metonymy and were not taken into account.

4.3. Limitations

Before we continue, some methodological limitations need to be mentioned. It should be indicated that manual qualitative analysis for semantic features suffers from at least two drawbacks: (i) relatively small sample sizes and the concomitant representativity issues; and (ii) the subjective nature of the annotation and the ensuing difficulty in its operationalization, which, in turn, affects replicability. In regard to the former point, it is a direct consequence of the non-automatic/manual annotation procedure involving close reading of examples. It is, therefore, essential to control for possible variation. By focusing on a restricted number of genres, we can increase the sample size for the preselected genres, thus at least partially addressing the issue of data representativity.

With respect to the other limitation, i.e., the subjective nature of qualitative analysis, careful operationalization of the annotation schema, clear presentation and rigorous systematicity in its application resolve this problem to a large extent. To further address the issue, secondary coding can be applied and inter-coder agreement can be checked, which was done in this study. More precisely, a subsample of the data was annotated by another coder. The degree of inter-coder reliability in the classification procedure is indicative of substantial agreement with the Cohen Kappa estimate at 0.73 (within the following confidence boundaries: lower: 0.56 and upper: 0.90). ³ It is also noteworthy that errors in subjective coding constitute noise in the data and as such they do not improve the predictions (Type I vs. Type II error). Statistical modeling, however, is designed to find structure in data despite noise.

Finally, it should also be noted that the non-random character of the data extraction procedure is a study design choice, rather than a limitation. More specifically, the data were extracted in a balanced manner on the basis of the combination of the two independent variables (i.e., lexeme and genre), rather than proportionally with respect to the lexical distribution in the three genres. However, the overall frequencies of occurrence were checked for both lexemes in the discourse context of the three genres and the results of this analysis are presented in Section 5.1 (Table 3). In themselves, these distributional findings represent an interesting insight into the lexical representation of negative self-evaluative emotions in the three genres.

5. Results and discussion

This section reports the results obtained through statistical modeling of the metadata discussed in the previous section. The quantitative analysis was performed in R (version 4.0.3, R Core Team, 2013). First, we inspect the distribution of embarrassment and shame across the three genres (Table 3, Section 5.1). Second, in Section 5.2, we turn to the descriptive statistics of the sample with regard to the distribution of source domains across the three genres (Table 4) and the two lexemes (Table 5). Then, we consider the results of two bivariate tests of independence (Section 5.3) to determine whether there is significant variation in how negative self-evaluative emotions are conceptualized metaphorically across the genres and lexemes. Finally, we turn to a mosaic plot visualization of a loglinear analysis (Section 5.4), which seeks to identify any interactions between the genre and the lexeme with respect to source domains.

5.1. Lexical distribution across genres

This section addresses the question of the distribution of the two nominal exponents of negative self-evaluative emotions across the three genres. Table 3 presents the raw frequencies of occurrence of embarrassment and shame. Given that one of the meanings of shame is not related to negative self-evaluative emotion (as in And it’s a shame that it is inconvenient, COCA), the examples had to be inspected for relevance. This was achieved for a representative sample established on the basis of a power analysis with the confidence of 90% and the p-value of 0.05. ⁴

Of the two lexemes, shame is the more frequently encountered instantiation of the emotional experience under analysis. This is not unexpected given that shame is perceived as the generic-level term for negative self-evaluative emotions or, as Kaufman (1996: 24) puts it ‘the central, integrative concept’. With regard to the genre-specific-distribution of the two lexemes, both embarrassment and shame are most typical of the discourse context of fiction, followed up by magazine and closely by spoken discourse. That these nouns are more commonly used in the written mode, particularly in fiction, should not perhaps come as too much of a surprise, since the nominal encoding (as opposed to adjectival) of emotions does not appear to represent the preferred choice in English (Wierzbicka, 1999), and choices that are linguistically marked would seem to be more probable to occur in less spontaneous discourse.

5.2. Metaphorical source domain distribution across genres and lexemes

Table 4 presents all the source concepts revealed in the sample across the three registers, with those already revealed in Kövecses (1998), Tissari (2006) or Fabiszak and Hebda (2007) highlighted in blue. Among them, we find cover, which, as evidenced in example (10), is related to Kövecses’ source domain having no clothes on.

As can be observed, it is the corpus component of the magazine, followed closely by that of spoken TV discourse, which is the most abundant in the metaphorical conceptualization of the two emotion categories. This is true both in terms of types and tokens. There is only one source concept that has not been attested in this register, i.e., conflict. Overall, the most frequent metaphorical conceptualizations of the two lexical categories are in terms of two generic source concepts, i.e., object (37) and container (30), with the latter being particularly typical of shame (see Table 5). There is another frequently represented generic source concept, person (20), which is preceded in terms of raw frequencies by four more specific source concepts, i.e., liquid (26), opponent (25), disease (23), and danger (22). As should be noted, however, the source domain opponent is a more specialized case of personification. As many as 11 conceptual source domains in the sample do not cross the threshold of 10 occurrences.

As Table 5 shows, taking into account raw frequencies, some of the source domains are relatively evenly distributed for the two lexical exponents of negative self-evaluative emotions, including object, plant, fire or force. There are also some metaphors that have been attested considerably more frequently for one of the emotion terms. Such disproportion is particularly noticeable for burden and container (more frequent with shame) and danger or disease (more frequent with embarrassment). It is noteworthy that with person and opponent, where the latter is a more specific case of the former domain, we observe that while embarrassment favors the more specific concept, shame is more frequently associated with the generic concept. In the next section, we will see whether any of the above-discussed associations constitute significant patterns that could predict lexical and genre effects on metaphor choice in the conceptualization of negative self-evaluative emotions.

5.3. Are there any lexical or genre effects on metaphor choice?

The primary research questions addressed in the present study concern genre and lexical effects on the distribution of conceptual metaphors associated with negative self-evaluative emotion constructs in American English. To test our hypotheses corresponding to these questions, the Chi² test of independence was employed. ⁵ For the purposes of the quantitative analysis, care needed to be taken of the smaller cells in the data. Accordingly, some of the source domains that were related taxonomically (stain/dirt > substance; cover > object) or that represented closely related categories (liquid and container; disease, pain, injury and danger) were combined. ⁶ Other infrequent source domains (cold/heat, conflict, confinement, crime, force, plant, precious object) were removed from the sample. Despite these changes, we still observe a relatively high number of degrees of freedom and data sparsity, which is why in the two chi² tests of independence, Yates’ correction was implemented for adjusted degrees of freedom and Monte Carlo permutations were added (n = 1000). As established, the variation in conceptual metaphor distribution is statistically significant both across the three genres (Table 6) and with respect to the two lexical exponents of negative self-evaluative emotions (Table 7).

Table 6. Results of the Pearson’s chi² test of independence: metaphor choice wrt. genre

Table 7. Results of the Pearson’s chi² test of independence: metaphor choice wrt. lexeme

The high number of degrees of freedom in the data warrants caution, which is why the results obtained through the chi² tests of independence were further validated through other tests of independence. Importantly, the first of the tests presented below is specifically designed for datasets with small cells.

1. (i) The Fisher exact test of independence:

   genre effects: p-value = 0.0004998;

   lexical effects: p-value = 0.0004998;

2. (ii) The Approximative General Independence Test of independence, library{coin}:

   genre effects: maxT = 4.2767, p-value = 8e-04;

   lexical effects: maxT = 6.0459, p-value <1e-04;

3. (iii) The G-test/Likelihood Ratio test, library{vcd}:

   genre effects: X-squared = 50.439, df = 16, p-value = 1.9511e-05;

   lexical effects: X-squared = 81.904, df = 8, p-value = 2.0206e-14.

All the results show that both the discourse context of genre and the lexeme have a significant effect on metaphor choice. The specific frequency-based associations and disassociations between the conceptual metaphors and the three genres are visualized in Figure 1, while those between the conceptual metaphors and the two lexemes are presented in Figure 2. Importantly, some of the Pearson residuals visualized in the two figures are indicative of clear differences.

Figure 1. Visualization of the associations and disassociations between conceptual metaphors and genres.

Figure 2. Visualization of the associations and disassociations between conceptual metaphors and lexemes.

Packages: {corrplot}.

As can be observed in Figure 1, ⁷ among the most important positive associations for fiction, we find fire and substance, as well as a weaker one with danger-disease. Given the values of the Pearson residuals, the first two source domains represent particularly strong associations. The genre of the magazine is distinctly associated with the source domain opponent, and less strongly with danger-disease and container-liquid. Finally, spoken discourse has one very strong association, i.e., object, and two less distinct associations, namely, place and burden.

It is noteworthy that the written discourse is associated with more specific source domains that are linked to more vivid imagery and intricate conceptual structure, such as fire or opponent. Spoken discourse, on the other hand, is associated with more generic level domains, such as object and place, which are rather schematic in terms of image or function, mostly providing ontological status. We thus find empirical evidence that fiction and magazine are related to more complex source domains, rather than necessarily exceeding spoken discourse in terms of the sheer number and heterogeneity of the metaphors observed, as was hypothesized. This divergence between the genres in terms of the complexity of the evoked source domains might perhaps be motivated by the degree of spontaneity in language production typical of written and spoken discourse.

With regard to the associations observed for embarrassment and shame, we find danger-disease, place and opponent as the most distinct source domains correlated with the former lexeme, whereas shame is strongly associated with container-liquid, person and burden. Fire, object and substance, for which we have observed some strong genre-related effects, are not significantly associated with either lexeme.

It was hypothesized that shame, as the more serious emotion of the two (see Section 2.1), would be associated with more negatively valenced source domains, linked to more severe consequences for the experiencer. While fire can definitely be linked to annihilation, such categories as danger-disease and opponent, observed for embarrassment, may also represent a grave threat. We thus do not seem to have found much supporting empirical evidence for this hypothesis. However, it is interesting to note that the categories of danger-disease and opponent might be argued to include in their conceptual structure an element of surprise or unexpectedness, which is a more typical property of the antecedents of embarrassment.

Overall, the above results demonstrate that there is significant variation in the conceptual metaphors that are chosen in negative self-evaluative emotions conceptualization relative to the lexical exponent and the discourse context of the genre. The question remains as to the combined effect of both lexeme and genre on metaphor choice. This is particularly important because lexical choice and the discourse context of a specific genre are unlikely to be completely independent variables. Let us therefore turn to an analysis that models the choice of the source domain with respect to the discourse and lexical context simultaneously.

5.4. Interaction between lexeme and genre in metaphor choice

To identify any possible interactions between the lexical exponent of shame and the discourse context, a log-linear analysis was performed and the results are presented in the form of a mosaic plot in Figure 3. In mosaic plot visualizations, the color of the box is indicative of association (blue) or disassociation (red), while its size corresponds to the proportional representation of the combination of the respective features of the three variables in the data (i.e., the smaller the box for a given association of features, the less supported such an association is by the data). Inspecting the associations visualized in Figure 3, we can now gain more insight into the findings obtained in Figures 1 and 2.

Figure 3. Mosaic plot visualization of log-linear analysis results for conceptual metaphors wrt. Lexeme and genre.

Packages: {MASS}, {vcd}.

We can see that embarrassment is indeed associated with disease – danger and place, but this is only true in the discourse contexts of magazines and spoken language, respectively. In addition, we also find that it is associated with substance in fiction, with opponent in magazine, and with object in spoken language, in which it is also disassociated from container – liquid. Shame, on the other hand, is found to be indeed associated with burden and container-liquid, but only in the discourse contexts of spoken language for the former source domain and magazine and spoken language for the latter. Beyond these correspondences, we also find an association with fire in the context of fiction and with object in spoken language. There is also one considerable disassociation for shame, namely disease – danger in the discourse context of the magazine. There is only one source domain that is an association shared by the two lexical exponents of negative self-evaluative emotions, i.e., object in the context of spoken discourse, which is consistent with the observation made in the preceding section that spoken language appears to attract generic source domains. This generic-level source domain was not only the most numerous metaphor type in the dataset, but it was also the strongest association for the spoken genre. It was not found to be particularly strongly associated with either lexeme when genre-variation was not accounted for, thus demonstrating the overriding effect of the genre in this case.

6. Conclusion

The present study constitutes the first attempt at applying the behavioral profile approach to the investigation of conceptual metaphor variation for negative social emotions. Not only does it test findings obtained in prior research through a different methodological framework, but it also expands the inventory of previously proposed metaphors for shame and embarrassment. In so doing, it captures subtle lexical and genre effects on metaphor choice in emotion communication, providing further empirical support for the importance of context in the representation of emotion concepts.

More specifically, the study has pursued descriptive as well as methodological objectives. Descriptively, it sought to establish whether there are lexical and genre effects on metaphor choice in the conceptualization of two negative self-evaluative emotions. It has been demonstrated that metaphor choice does indeed manifest both lexical and genre effects. Regarding the lexical effects, while, perhaps unsurprisingly, there is substantial overlap in the metaphorical conceptualization of the two closely related near-synonymous terms encoding negative social emotions, we have also observed statistically significant variation. Embarrassment has been found to be associated with such specific source domains as disease-danger or opponent and a more generic domain of place, shame, on the other hand, is linked to the specific domain of burden, and more generic domains of container-liquid or person. This slight preference of shame for more schematic domains might be related to the broader, umbrella-like character of the term, representing, as Kaufman (1996: 24) phrases it, the ‘central, integrative concept’ within negative social emotions. As already indicated in Section 5.3, we do not seem to have found enough empirical evidence for the claim that the ‘grimmer and weightier’ (Tangney et al., 1996: 1258) nature of shame should also be reflected in comparatively gloomier (i.e., more negatively valenced) source domains.

With regard to the genre effects on metaphor choice, we have observed that spoken discourse has two weaker associations (place for embarrassment and burden for shame) and one strong association with object (for both terms), which represents the only association that emerges as strongly dictated by genre, when the interaction between genre and lexeme is taken into account. As for the two other genres, fiction has been found to be associated with fire (for shame), substance (for embarrassment), and, less distinctly with danger-disease (but only when genre effects are considered), while magazine with opponent (for embarrassment), danger-disease (for both terms), and, less strongly with container-liquid (for shame). Overall, the written mode is associated with two more metaphor types than the spoken mode (see Figure 1: fiction – fire, substance, danger-disease; magazine – opponent, danger-disease, container-liquid; spoken – object, place, burden) (see Figure 1). However, as we have observed, the lexical and genre effects in reality interact in the metaphorical conceptualization of negative social emotions and so they should be accounted for simultaneously. When this is done, the apparent dominance of the two written genres over spoken discourse in terms of greater abundance in metaphor types disappears. The written discourse, nonetheless, does appear to attract metaphors that are more complex in their conceptual structure, such as fire, disease or opponent.

Methodologically, it has been demonstrated that discursive contextual effects and lexical effects on metaphor choice in emotion conceptualization are indeed quantifiable. Importantly, the method applied here allows us to make generalizations about the sociolinguistic context under analysis, which would not be possible with purely qualitative analysis. Overall, the results provide further empirical support for the need to examine metaphor in the context of ‘actual language use’ (Steen, 2008: 221).