The readability of corporate sustainability narratives in a crisis: an analysis of linguistic features using Coh-Metrix

Even if there is no universally accepted definition of a crisis (Coombs, 2010), the term is associated with an unpredictable (albeit not improbable) major event that threatens to harm a company and its stakeholders (Massey, 2001). Crises can take many forms, including natural disasters (tsunamis, earthquakes and wildfires), industrial accidents (spills, explosions and product defects) and intentional events (workplace violence, product tampering and terrorist attacks) (Seeger, 2006). Regardless of the type, during a crisis a company faces a legitimacy issue (Deegan, 2002) that implies a threat of reputational damage (Bebbington et al., 2008; Firoozi and Ku, 2023).

Corporate reputation has become a crucial resource for all companies in recent decades (Mohd Sofian et al., 2023). Being a “perceptual and social” asset (Fombrun and van Riel, 2004, p. 32), the management of corporate reputation is a sensitive issue, as the occurrence of a reputational crisis can quickly erode all the efforts undertaken by a company to establish its reputation (Mohd Sofian et al., 2023). Indeed, the development of information and communication technologies has dramatically increased the speed at which information circulates on digital platforms and social networks, making it increasingly difficult for companies to hide or limit the diffusion of damaging information (De Wolf and Mejri, 2013). In a reputational crisis, companies are expected not only to control the flow of information related to the event but also to feed this flow in a proactive manner (Hale et al., 2005).

The importance of corporate communication in managing complex and multifaceted phenomena like organizational crises has led researchers to develop different perspectives of analysis and propose various methods of investigation (Marsen, 2020). The impression management (IM) framework is among the most used tools to investigate how companies manage post-crisis narratives in an attempt to limit reputational losses (Samkin and Schneider, 2010; Michelon, 2012; Florio and Sproviero, 2021). Our study aims to build on the previous evidence by focusing on one specific IM technique, reading ease manipulation. Specifically, we focus on sustainability reporting narratives to investigate any change in readability after a reputational crisis. The manipulation technique of reading ease has generally been investigated to test IM in poorly performing companies (Merkl-Davies and Brennan, 2007), while evidence in the context of a reputational crisis is lacking. During a reputational crisis, a company is motivated to engage in IM, due to both internal and external pressures. On the one hand, pressure is exerted by shareholders fearing a loss in the company’s value; on the other hand, external pressures, primarily in the form of media pressures, require a prompt and effective response by the managers to reassure all stakeholders. Since readability is a main aspect of disclosure, managers experiencing a reputational crisis may have an incentive to use manipulative strategies to influence readers’ perceptions of the company. The use of sustainability reporting as a means to repair reputational damage is supported by the fact that it is among the most highly scrutinized public disclosures (Keusch et al., 2012) and that the narratives included in a sustainability report fall within a “communication exercise” voluntarily provided by the company to accompany financial statements and other regulated reports (Hrasky, 2012).

Similar to Corazza et al. (2020), our analysis focuses on Costa Crociere, an Italian tour operator that experienced a reputational crisis after an incident involving one of its cruise ships on the Isola del Giglio in 2012. The worldwide resonance of this event created a huge reputational threat for the company, making this case appropriate for our analysis. However, while Corazza et al. (2020) analyse various types of post-crisis disclosures released by Costa Crociere, we analyse the contents of sustainability reporting. We exclude any specific disclosures related to the accidents to verify whether the accident resulted in a change in the corporate disclosure’s readability. While Corazza et al. (2020) aim to identify the main features of the company’s crisis communication strategy, our paper investigates whether and to what extent a crisis event can lead a company to manipulate its disclosure by changing the level of readability. A five-level language-discourse framework is used to provide a comprehensive and in-depth analysis of text readability (Kintsch, 1998; Snow, 2002; Graesser and McNamara, 2011). Each level is investigated through a set of readability measures extracted from the multidimensional Coh-Metrix computational tool (Graesser et al., 2004; Graesser and McNamara, 2011). Our results are contrasting: while a decrease in readability is observed in the Environmental Performance section, the opposite trend is found in the Social Performance and the Letter to Stakeholders sections. While these results suggest a tailored approach in adapting the linguistic characteristics of the post-crisis sustainability report, they do not provide a clear understanding of manipulative intent. This exploratory analysis offers a valuable contribution to crisis communication studies by focusing on a relevant and under-investigated aspect of language features represented by linguistic features. Furthermore, it enriches IM studies by exploring the reading ease manipulation technique in the context of reputational crises. From a methodological point of view, we propose the adoption of multidimensional linguistic tools like Coh-Metrix to investigate the manipulation of reading ease in corporate communication. Compared to criticized readability formulas, Coh-Metrix allows for deeper analyses of disclosure by focusing on several levels of language structure. By unveiling the complexity of corporate narratives’ readability, software like Coh-Metrix opens up future research opportunities to explore the more nuanced aspects of language structure and readability in crisis communication strategies.

The remainder of the paper is organized as follows. The next section outlines the theoretical framework. A literature review is provided first, drawing on crisis communication and IM literature, then the research question is proposed. Section 2 focuses on the methodology, introducing the case of Costa Crociere and illustrating the set of Coh-Metrix readability measures used in the analysis. The results of our study are presented in section three, while the final section offers a discussion and concluding considerations.

Organizational crises represent a potentially serious threat as they are untimely but predictable events that have actual or potential consequences for stakeholders’ interests as well as the reputation of organizations (Millar and Heath, 2003). Reputational consequences concern the loss of legitimacy experienced by the company because of the crisis (Coombs, 2007). It has a great potential to impact a company’s competitiveness and performance results through influencing stakeholders’ decisions and actions towards the company. Reputation is considered a vital component of an organization; it is an essential intangible resource that helps reduce uncertainty and build trust with stakeholders (Sawalha, 2020). Thiessen and Ingenhoff (2011) argue that corporate reputation is increasingly becoming a competitive factor, influencing stakeholders’ decisions and actions towards the company. Academic research confirms that corporate reputation is a critical factor for all companies (von Berlepsch et al., 2024). A favourable corporate reputation improves the evaluation of job attributes and, hence, strengthens employment intentions (Cable and Turban, 2003). Further, it has a positive effect on customer engagement (Parray et al., 2023), is positively associated with a firm’s ability to attract qualified staff (Eccles et al., 2007), contributes to the deepening of relationships (de Castro et al., 2006) and builds trust (Van Der Merwe and Puth, 2014).

Corporate communication is generally understood to be one of the most important tools with which a company can manage its reputation (Mohd Sofian et al., 2023). More specifically, corporate communication is considered a strategic instrument to stem the reputational threat associated with a crisis (Benoit, 1997). Thiessen and Ingenhoff (2011) highlight the fact that in situations where reputation is threatened, communication is an aspect of crisis management that gains great importance. Seeger and Sellnow (2016) state that a crisis creates surprise and uncertainty as it breaks an established order and, in doing so, it generates an empty space, named a “narrative space”, that gets filled by stories. The narrative space gets filled by several different stories, as the company is not the only actor to tell its own story about the crisis. According to Heath (2009), many actors, including other corporations, political actors, activists, experts and the media, tell their own stories, creating a contest of multiple voices, that Frandsen and Johansen (2010) define as a “rhetorical arena”. The different narratives circulating in the rhetorical arena influence the way most people experience and understand the crisis, so that they actually frame the story of the crisis (Seeger and Sellnow, 2016). A firm experiencing a crisis takes part in the rhetorical arena to provide its own story. According to Zhao et al. (2018), if a firm neglects to construct a narrative accounting for their side of the story, then the public might formulate their storyline diverging from the organization’s preferred interpretation of a scandal.

Traditionally post-crisis narratives have been investigated in the realm of the IM theoretical framework (Elsbach, 1994), focusing on strategic portrayals of responsibility, blame, scapegoating, denial of responsibility, justification and related strategies (Seeger et al., 2005) [¹]. Previous studies have generally investigated companies’ use of ad hoc communication aimed at managing legitimacy crises and restoring reputation (Cho, 2009). According to Tedeschi and Melburg (1984), managers experiencing a reputational crisis are expected to engage in “defensive” IM techniques to avoid potential adverse reactions from stakeholders and minimize undesirable consequences (Moreno and Jones, 2022) and to retain a positive identity and reputation (Ogden and Clarke, 2005). Empirical evidence documents the presence of defensive IM strategies in many crisis contexts. Michelon (2012) finds a decreasing level of optimism and a growth in active language in the press releases issued by British Petroleum after the oil spill in the Gulf of Mexico in 2010. Florio and Sproviero (2021) show that during the 2015 Dieselgate scandal, the Volkswagen Group adopted rhetorical techniques to support an “avoid panic” strategy. Samkin and Schneider (2010) document the presence of IM techniques in the annual reports of a New Zealand public benefit entity; these techniques were aimed at maintaining and repairing the organizational legitimacy that had been weakened by extensive negative media publicity. Beelitz and Merkl-Davies (2012) investigated managerial discourse in corporate communication during the six-month following an incident in a German nuclear power plant, documenting how the CEO strategically used this discourse to obtain consent regarding the continued operation of the plan.

This study focuses on the reading ease manipulation strategy among the different manipulation techniques. According to Courtis (1998), reading ease manipulation consists of a deliberate technique whereby managers manipulate narratives to alter their readability (Courtis, 2004). IM studies usually employ reading difficulty to test obfuscation hypotheses (Merkl-Davies and Brennan, 2007), with obfuscation being defined as “a narrative writing technique that obscures the intended message, or confuses, distracts or perplexes readers, leaving them bewildered or muddled” (Courtis, 2004, p. 292). Obfuscation studies are based on the presumption that their “preparers manipulate transparency by reducing clarity when they wish to disclose less about their underlying circumstances” (Rutherford, 2003, p. 189). A common argument in these studies is that managers have economic incentives to disclose messages conveying good performance more clearly than messages conveying poor performance (Laksmana et al., 2012).

It is well established in the literature that a reputational crisis raises the need for a company to make strategic use of all communication channels (Firoozi and Ku, 2023). This study focuses on the disclosure included in sustainability reports. A sustainability report is considered one of the most important communication tools through which a company manages its legitimacy (Deegan, 2002). Indeed, there is evidence of the importance of sustainability as a driver of reputation (Melo and Garrido-Morgado, 2012) and sustainability reporting as a communication channel and tool for reputation management (Park et al., 2014; Axjonow et al., 2018) even in time of crisis (Park et al., 2020; Arora and Lodhia, 2017). Cho (2009) points out that social and environmental disclosures remain a powerful legitimacy device rather than an effort towards greater accountability, especially in the case of environmental accidents. Numerous studies have related sustainability reporting to critical events, focusing on the oil and gas (Cho, 2009), mining (De Villiers and Van Staden, 2006) and clothing sectors (Islam and Deegan, 2010). Additionally, stakeholders’ interest in corporate social responsibility practices has been growing in the last decades. Since sustainability reporting is considered the most vital source of information on a company’s environmental and social impact, the interests of external stakeholders in sustainability reporting should increase in the case of a reputational crisis.

The limited available evidence supports the idea that managers strategically massage the content included in sustainability reports after a crisis. Park et al. (2020) examine the sustainability reports of two airlines before and after crises and find relevant changes in the reports’ central keywords, in the social issues supported by the companies and in the stakeholders who were given priority. Arora and Lodhia (2017) analyse the social and environmental information published on the British Petroleum plc website during the period of the Gulf of Mexico oil spill incident and find extensive use of image restoration strategies. Under pressure to limit reputational damage and restore corporate image (Benoit, 2014), we assume that managers might manipulate not only the content of the sustainability report but also the language used in it. Thus, they might engage in reading ease manipulation practices to influence readers’ perceptions. For instance, an increase in narrative readability could be strategically used to boost corporate crisis storytelling, making it more understandable and, thus, more trustable for stakeholders. Conversely, obfuscation strategies might be implemented to reduce the readability of corporate narratives, with fragmented syntax, complex sentences and technical vocabulary used to hamper the understandability of events or divert attention from personal responsibilities. Since predictions cannot be made regarding the expected changes in the readability of sustainability reporting disclosures after the crisis, the following open research question is formulated:

Because of its exploratory nature, this study is a case study analysis (Yin, 2018). According to Lune and Berg (2017), qualitative analyses like case studies are especially appropriate for investigating in-depth unexplored and novel research topics. In crisis communication studies, case studies serve as a foundation for examining how organizations develop and execute their post-crisis responses (Coombs, 2007).

Following Eisenhardt (1989), this study examines the crisis faced by the Italian tour operator Costa Crociere after the Costa Concordia accident. Launched in 2005 at a cost of €450m, the 114,500-ton cruise ship measured 290 m in length and 35 m in width. On January 12, 2012, near Isola del Giglio off the Tuscan coast, the Costa Concordia drifted away from its usual sea route and hit a submerged rock, causing one of the most cruise disasters since the Titanic. The impact created a large hull breach, flooding the ship and causing 32 fatalities and 193 injuries among the 4,229 people onboard. The disaster stemmed from two key factors: the captain diverted near Giglio Island to honour a former captain, and the helmsman misinterpreted an order, turning the rudder “hard to starboard” instead of “hard to port” 20 s before impact. In April 2013, Costa Crociere agreed to a plea deal of one million euros to cover its liability for the offences committed by its employees and the multiple injuries and deaths caused by its non-compliance with both regulations and laws. In May 2017, the Costa Concordia’s captain was definitively convicted by the Italian Supreme Court.

This case represents a critical example for analysis due to its severe reputational, social and environmental impacts. It affected a wide range of stakeholders, including employees, customers, authorities and the Giglio Island community. This is considered the worst accident in the cruise sector directly attributed to the captain’s negligence and human error. This fact, together with the high number of victims of different nationalities, gave the tragedy a great and durable resonance all over the world. Further, the accident caused severe environmental and ecological damage, as the wreck remained in the sea for more than two years (Giustinano et al., 2016). Due to the captain’s misconduct, the disaster’s impact and its global resonance, the Costa Concordia case draws parallels with the Titanic disaster (Schröder-Hinrichs et al., 2012; Marko et al., 2020). All these factors make the Costa Concordia disaster a valuable context for examining whether corporate communication was adapted after the crisis.

The aim of assessing readability is to predict the reading difficulty of a text through the study of its linguistic characteristics (François, 2015). This is a complex issue, because several textual characteristics (e.g., frequency of words, abstractness of concepts, type of syntactic structures, etc.) affect comprehension, and no unified explanatory model of text readability has so far been produced.

Readability formulas are the measure most used to assess the readability of a text, as they are easy to calculate and interpret (Crossley et al., 2008). However, the use of readability formulas has been severely criticized, as they only focus on a few aspects of readability (Crossley et al., 2008). They are even considered to be inadequate for analysing corporate disclosures because of the technical language of such reports and their high percentage of complex words (Loughran and McDonald, 2014).

An alternative approach to assessing readability is represented by computational linguistics, as this allows the investigation of elements of disclosure that are not quickly identified with other tools (Berger, 2011). Computational linguistic tools provide a detailed analysis of language and cohesion features through the integration of the language metrics that have been developed in this field (Jurafsky and Martin, 2000). These tools address many of the criticisms of traditional readability formulas because the language metrics they report usually include text-based processes and cohesion features that are integral to cognitive reading processes such as decoding, syntactic parsing and meaning construction (Graesser and McNamara, 2011). This integration facilitates the examination of the deeper linguistic features of the text that are related to text processing and reading comprehension (Crossley et al., 2008).

The computational linguistic tool used in this study is Coh-Metrix (Graesser et al., 2004; Graesser and McNamara, 2011). This is a large collection of software sub-systems that computes a range of language and discourse measures based on the empirical and theoretical foundations of text feature extraction (McNamara et al., 2014). Coh-Metrix has been developed to look at five language-discourse levels proposed in multilevel theoretical frameworks (Kintsch, 1998; Snow, 2002; Graesser and McNamara, 2011): word information, syntax, textbase, situation model and genre (Figure 1). A set of specific measures is proposed for each language-discourse level, allowing a complete and articulated readability assessment. This multidimensional approach results in an in-depth analysis that identifies differences in the language beyond the surface-level features usually proposed by traditional readability measures. The five language-discourse levels, as briefly described below, are used in this paper as an analysis framework to compare the readability of sustainability reporting disclosure before and after a reputational crisis.

Word information level represents the surface level in the language-discourse framework. It primarily refers to the vocabulary used in the text, assuming that the words composing the text have a substantial impact on reading time and comprehension (Perfetti, 2007). After words are recognized (by meaning), syntax links them together into meaningful phrases and constituents (Crossley and McNamara, 2011). These links define the second language-discourse level: the syntax level. Theories of syntax assign words to part-of-speech categories (e.g., nouns, verbs, adjectives or conjunctions), group words into phrases or constituents (e.g., noun-phrases, verb-phrases, prepositional-phrases and clauses) and construct syntactic tree structures for sentences (Graesser et al., 2011). The complexity of the syntactic structure involves the reader’s working memory, thus negatively affecting the readability of the text. The syntactic structure of written texts can show different levels of complexity, depending on the type of text. In general, a low level of complexity is associated with shorter sentences, fewer words before the main verb of the main clause and fewer words per noun-phrase (Jurafsky and Martin, 2000).

The textbase level considers a mental representation of the text (van Dijk and Kintsch, 1983), which consists of the explicit ideas (the meaning) in the text and also covers the semantic content of the text (Seger et al., 2021). A textbase is created using the connections between the discourse constituents that emerge at the sentence, paragraph and text levels to develop a coherent mental representation of the text (Crossley and McNamara, 2011).

The situation model level considers a mental representation of a text but is different from the textbase level (van Dijk and Kintsch, 1983). While the textbase consists of the explicit ideas expressed by the text, the situation model is a coherent representation of the situation referred to in a text that is constructed by drawing inferences activated by the explicit text and encoded in the meaning representation (Graesser et al., 2011). According to Zwaan and Radvansky (1998), a situation model of a text can be depicted according to five different dimensions: causation, intentionality (goals), time, space and protagonist. A break in one or more of these dimensions causes a discontinuity in the cohesion of the situation model that affects the readability of the text, as it increases reading time (O’Brien et al., 1998). The presence of specific signalling devices (particles), such as connectives, transitional phrases and adverbs, can assist the reader by making the discontinuity clearer. Thus, the investigation of these particles can be used to evaluate the level of cohesion of the situation model of the text (Graesser et al., 2011). Cohesion is facilitated by these particles, as they clarify and stitch together the actions, goals, events and states conveyed in the text (Louwerse, 2001).

The fifth level is the genre level. A genre consists of the organization of verbal and written communication based on external, non-linguistic criteria such as intended audience, purpose and activity type (Biber, 1988). Distinctive characteristics of language signal different text genres, such as narration, exposition, persuasion or description (Graesser et al., 2011). The relationship between text genre and readability consists of the distinctive characteristics of the text: narrative texts are substantially easier to read, comprehend and recall than informational texts (Graesser and McNamara, 2011). A robust approach for assessing the genre consists in measuring the level of narrativity. Narrativity measures are usually used to identify and distinguish different genres, while narrativity analysis is generally based on the examination of the words and syntax. Many measures capture the characteristics of oral language (Biber, 1988), which tends to be used for familiar topics and involve sentence constructions that are easy for the audience to comprehend. This results in the prevalence of short, high-frequency words and pronouns, shorter noun phrases, fewer words before the main verbs of the main clause and fewer passive constructions.

To investigate the readability of Costa’s disclosure before and after the disaster, our analysis focuses on the sustainability report for the fiscal year 2010, published in 2011 before the disaster and the sustainability report for the fiscal year 2012/2013, published in 2014 after the disaster. We downloaded the English versions of these reports for other research purposes in 2017 before they were removed from the company’s website. We focused on the English-language reports, as this language is the only one supported by Coh-Metrix.

Considering that the reputational issues caused by the accident were mainly related to social and environmental impacts, we selected the texts to analyse accordingly. Specifically, we focused on the letter to stakeholders (LtS texts), the section on social performance (SOC text) and the section on environmental performance (ENV texts). Narratives concerning social and environmental performances are closely related to the reputational threats caused by the event. The letter to stakeholders is also included in the analysis as it is considered the most important narrative to promote the image and reputation of organizations (Craig and Brennan, 2012; Civera et al., 2023). According to Cong et al. (2014), the language of a letter is critical also to respond to a crisis.

Table 1 provides some descriptive details of the sustainability reports analysed together with the number of words and characters of the analysed texts. The length of texts is based on Coh-Metrix’s maximum limit of 15.000 analysable characters.

Among the 106 measures automatically computed by Coh-Metrix, we selected 46, drawing on the work of Graesser et al. (2011). According to these authors, these measures are the most effective in emphasizing variations in text characteristics, and they do not show a high correlation. Additionally, they “ensure sufficient coverage of the different language and discourse levels” (p. 228). The measures are reported in the first column of Table 2.

Table 2 reports the results of the Coh-Metrix measures used to compare the linguistic features of the selected sections of the Costa sustainability reports before and after the crisis. The discussion follows the five levels included in the language-discourse framework adopted for the analysis.

The analysis of the five-level framework starts from the word level. Results reported by the three psychological ratings (n. 1–3) do not show substantial changes after the crisis, supporting the idea that the company did not make a strategic change in the words used in its sustainability narratives as a consequence of the reputational crisis. Reported scores are in the neighbourhood of the middle score of 350 [²] both before and after the crisis, indicating that, although the sustainability report is a technical document with informative content, it makes wide use of common and accessible vocabulary.

At the syntax level, the texts are characterized by a high density of noun phrases and very low sentence syntax similarity, as indicated by indices n. 7 and n. 9, reflecting the predominantly descriptive nature of the texts. The indices also reveal notable differences before and after the crisis. Specifically, LtS texts—and to a lesser extent, SOC texts—exhibit an increase in readability. This improvement is primarily evidenced by a reduction in left embeddedness (n. 8) and, for LtS texts, a decrease in sentence length (n. 4). In contrast, ENV texts show scores that indicate more complex syntax, suggesting that their readability became more challenging after the crisis. Specifically, the increase in the average number of words per sentence (n. 4) indicates the presence of longer phrases that are more difficult to read. The decrease in sentence syntax similarity (n. 5 to 6) suggests that the syntax has become less straightforward to process. Additionally, the increase in left embeddedness (n. 8) is associated with a higher cognitive load on the reader’s working memory.

The worsening of the readability reported at the syntax level for ENV texts is not confirmed at the textbase level, as readability indices suggest a more cohesive (readable) disclosure. The referential cohesion measures (n. 11–16) exhibit a substantial increase after the crisis, for both noun, argument and content overlaps, thus suggesting a more cohesive text. The lexical diversity type–token ratio referred to all words (n. 18) decreases, signalling a text for which the processing is less complicated. An improvement in text cohesion is also reported for LtS texts, although in this case the changes in the indices are less pronounced. Contrarily, all the indices’ changes for SOC texts show a negative sign, suggesting a reduction in text cohesion after the crisis.

A more pronounced change at the situation model level was observed in ENV texts. The small decrease in causal connectives (n. 21), the reduction in adversative and contrastive connectives (n. 25) and the notable reduction in intentional verbs (n. 32), accompanied by a decreasing incidence of causal verbs (n. 30 to 31), suggest a shift towards a more descriptive and less contrasting (or plainer) situation model. However, the substantial decrease in the incidence of additive and temporal connectives suggests a decline in ease of understanding. A similar trend is registered for LtS texts, which show a reduction in the incidence of connectives, particularly additives and temporal connectives. Causal connectives (n. 22) are the only exception, showing a small increase. This finding, combined with the rise in causal verbs and causal particle indices (n. 31), suggests a shift in the situational model of LtS texts towards a more explanatory approach. Changes in the situational model are also evident in SOC texts. Here, alterations in connectives seem to result in more straightforward and descriptive disclosures. This is reflected in a significant increase in the use of additive connectives (n. 24), alongside decreases in the incidence of adversative and contrastive connectives (n. 25) and intentional verbs (n. 32)

The final level of analysis focuses on the genre level. Indicators such as the average familiarity of content words (n. 46), the average age of acquisition of words (n. 45) and the CELEX log frequency index (n. 43), which measures the use of infrequent or rare words, reveal consistent results before and after the crisis. Similarly, narrativity indicators (n. 34–36) do not reveal significant changes, except for the incidence of adverbs and pronouns (n. 37 and n. 38). These indices exhibit the most notable changes at the genre level, with distinct trends across text types. ENV and SOC texts display a substantial decrease in pronoun incidence (n. 38), particularly third-person plural pronouns (n. 42), while first-person (n. 39 and n. 40) and third-person singular pronouns (n. 41) remain unchanged. Conversely, LtS texts show a marked increase in pronoun use, especially first- and third-person plural pronouns, alongside a slight decrease in first-person singular pronouns. Given that pronoun use is associated with higher readability, these findings suggest a potential negative impact on the readability of ENV and SOC texts and a positive impact on LtS texts. It is worth noting that the magnitude of these effects may be more pronounced in LtS texts, as they exhibit substantially higher pronoun incidence per 1,000 words compared to ENV and SOC texts.

Finally, a contrasting indicator for the readability of ENV texts is provided by the CELEX log minimum frequency index (n. 44), which shows a decline in the use of low-frequency words after the crisis, with a positive impact on readability.

This exploratory study focuses on the reputational crisis experienced by Costa Crociere in 2012 to examine any change in the readability in its post-crisis sustainability reporting narratives. To offer a complete overview of sustainability reporting narratives, three primary sections have been investigated: the Letter to Shareholders, the Environmental Performance section and the Social Performance section. The findings reveal considerable changes across the five levels of readability, with the exception of the word level, where no noticeable changes were observed.

The disclosure in the Environmental Performance section shows the most pronounced changes. After the crisis, the content appears more straightforward and descriptive, with fewer intentional and causal verbs, causal connectives, as well as adversative and contrastive connectives. this disclosure is marked by more complex syntax, featuring longer sentences that require additional processing time. At the textbase level, results indicate an increased overlap of content after the crisis, suggesting a more focused approach to disclosure. Notably, there is a decrease in the use of personal pronouns, particularly third-person plural pronouns, after the crisis. In business communication, such pronouns are often used to foster a sense of community, so this reduction may signal a shift in the intended objectives of the disclosure. Overall, these changes in linguistic features could be better interpreted, considering that, despite the high number of victims, the Costa Crociere disaster gained widespread attention as an environmental disaster because the ship was stranded on the seabed off Giglio Island for an extended period. While environmental responsibility was likely viewed as the most significant reputational issue, the attempt to focus on more straightforward, descriptive and impersonal contents may reflect an effort to provide objective and neutral information—perceived as more reliable—with the goal of restoring trust in the company.

Fewer notable changes were observed in the Social Performance and Letter to Stakeholders sections, for which results show a general increase in readability. This is mainly referable to a reduction in sentence length and in the words before the main verbs, as well as to a decrease in the use of adversative and contrastive connectives. However, some specific features differentiate the trends in these two sections. The most notable difference concerns the use of connectives and pronouns. In the Social Performance section, the incidence of adversative and contrastive connectives decreases while the use of additive connectives increases. Furthermore, the incidence of pronouns, especially the third-person plural pronoun, decreases. As with the Environmental Performance section, the Social Performance content appears more focused on conveying impersonal, descriptive information. In contrast, the Letter to Stakeholders shows an increase in both the use of causal verb and causal particles, along with a remarkable rise in the use of the third-person plural pronoun. These changes positively influence the readability of the section, enhancing its role as key communication tool for fostering a sense of community.

While our results do not provide clear evidence of ease reading manipulation, they reveal interesting changes in the readability features of sustainability reporting after the crisis. Given that language features are a crucial component of narratives and narratives are among the most important tools in crisis communication, our findings suggest that linguistic features are just as relevant to post-crisis disclosure as other, most investigated, language features, like contents, tone and rhetorical devices. These results open up new avenues for future research into the factors influencing readability changes and enhances our understanding of how companies manage post-crisis narratives. In the field of IM, our findings pointed out different trends in the linguistic features of the three examined sections after the crisis. While these results hinder the detection of any manipulative intent, they leave open the possibility that companies, in preparing their sustainability reports, tailor the linguistic features of each section to align with specific objectives and address different stakeholders.

Unlike Corazza et al. (2020), who employed textual and content analysis to identify the main features of Costa Crociere’s crisis communication strategy, our study uses a structured analytical framework derived from linguistics to detect the specific manipulation technique of reading ease manipulation. Multidimensional linguistic software, like Coh-Metrix, offers opportunities for future research for the investigation of corporate narratives. This analytic approach proposed by this software can be particularly useful for exploring the nuanced aspects of language manipulation in greater depth, contributing to the call for research on post-crisis narrative manipulation (Florio and Sproviero, 2021). By including the analysis of deep language structure levels, such as text-base and situation models, the proposed framework allows for a more comprehensive investigation of textual manipulation compared to common readability measures based solely on sentence structure.

By proposing an empirical tool to investigate the language features of the quality of a narrative, this study offers a contribution to companies, professionals and regulators involved in the process of regulating non-financial disclosure. Since non-financial disclosure is mainly narrative in nature, the issue of language features becomes relevant in guaranteeing the quality of corporate reporting. More generally, the set of indices proposed in the analysis could be adopted as a useful tool for investigating the quality of corporate narratives as regards the specific aspect of language features. The growth in the number of narratives voluntarily published by companies in their corporate reports, together with the ongoing regulation process of non-financial reporting, call for more research evaluating the overall quality of disclosure, including language features. Finally, multidimensional tools, like Coh-Metrix, could also be used at the professional level to answer the call of, among others, fund managers, financial analysts and external auditors for new algorithms and tools that can help them assess the quality of non-financial reporting (Beattie et al., 2004), as well as of sustainability reporting (Clarkson et al., 2020). The case proposed in this study does not allow for generalisation due to its unique context and its exploratory nature. However, considering that “reputational crises are becoming more frequent” (Schermer, 2021, p. 82), the linguistic features of post-crisis communication represent a relevant strand of literature that calls for further research.