Abstract
Today, customer service interactions take place in the digital sphere to a considerable extent. In fact, several social networking platforms have been reappropriated to this end, allowing businesses to engage in conversations with their customers. In this study, we investigate the nature of these conversations on Twitter (now X) in three different French-speaking countries. We focus on the national train operating companies in France (SNCF), Belgium (SNCB) and Switzerland (CFF) and study their Twitter interactions with customers in French, using a corpus linguistic methodology. Our study is based on a corpus of more than half a million tweets addressed to and posted by the respective companies between 2018 and 2022. The aim of this contrastive study is to highlight the linguistic features and communicative strategies used by the respective companies and their customers, with a view to uncovering potential features pertaining to the discourse of customer service tweets in French.
1 Introduction
Social media provide consumers with an easily accessible space for sharing their thoughts about and experiences with products and services. This is especially the case in the context of the railway industry (e.g., Kietzmann et al., 2011), as the social networking platform Twitter â now called X â is routinely used by (dis)satisfied commuters to voice their (dis)satisfaction with their travel experience and by train operating companies to provide customer service (Howard, 2020). Against this background, it is important for social media managers of train operating companies to monitor customersâ positive and negative feedback and appropriately respond to their messages. At the same time, an investigation of the similarities and differences with regard to the communicative strategies and linguistic features used by train operating companies and their customers is all the more relevant and timely, especially when considering a comparison between train companies operating in different European countries in which varieties of the same language are spoken.
In this study, we address the interactions between social media managers of national train operating companies and their customers on Twitter in three French-speaking countries, i.e., Belgium, France and Switzerland. This choice is motivated by the fact that, despite some attention having been paid to online Business-to-Consumer (B2C) interactions in the French-speaking domain (e.g., Béal and Grégoire, 2022), linguistic studies on French business communication in the context of the railway industry are few and far between. More precisely, our research aim is to identify the linguistic and communicative properties of both customersâ and service managersâ tweets in Belgium, France and Switzerland, with a view to uncovering possible cross-regional differences between these three varieties of French. Our study, therefore, adopts a contrastive perspective as we will compare the language of tweets addressed to and posted by train operating companies in three varieties of French. The present study sides with recent work in pragmatics, such as Vladimirou et al. (2021), which focused on the features that differentiate customer feedback voiced on social media from its counterparts in face-to-face and traditional written communication. In addition, as our study investigates the language of customer service interactions in three varieties of one and the same language, i.e., French, it complements previous cross-linguistic studies that compared customer feedback in different languages, such as Dutch, English and Italian (Cenni and Goethals 2017) and Chinese and English (Feng and Ren 2020).
A key assumption in our present work is that corpus linguistics is a valuable method for investigating the linguistic features of business discourse on social media (see e.g., Lutzky, 2021; Lutzky and Kehoe, 2022). Our analyses are based on an 11.1-million-word corpus comprising French language tweets which customers addressed to the national train operating companies in Belgium, France and Switzerland and social media managersâ replies. We compiled this corpus through the Twitter API, extracting all relevant tweets that were posted on Twitter between January 1, 2018 and December 31, 2022.1 The results that we will report in this article are based on a combination of corpus linguistic analyses. This includes engaging in quantitative approaches, such as keyword and collocation analyses, as well as adopting a more qualitative perspective when exploring specific instances of language use from our corpus in detail. Although several studies have already examined the topic of online customer feedback and customer service interactions (see Section 2 for an overview), the present study stands out, firstly, because of the large size of the corpus studied and, secondly, for the original contrastive approach involving the Belgian, French and Swiss varieties of the French language.
This article is structured as follows. In the next section, we offer a review of previous research on the language of customer feedback and webcare, zooming in on discourse-linguistic studies of online complaints in the railway context, specifically on Twitter and in the French language. In the Data and Methodology section, we present our corpus and our methodological approach: we address, in turn, the context of the railway industry in French-speaking Europe, the constitution and properties of our dataset, and the nature of the analyses that we will report on and interpret in the Findings and Discussion section. In the Conclusion section, we summarize our findings, identify the key implications of our results, and make suggestions for future work.
2 Theoretical Background
The advent of the online sphere, in particular social media, has dramatically altered the communication between companies and customers. In the current era of digitalization, customers have abundant opportunities to share their opinions on products or services and offer recommendations to their peers (Argenti, 2006), and they routinely make use of these affordances, often in the form of messages posted on social media such as Twitter, a platform that has become associated with negative customer feedback, such as complaints and criticism (Vargo et al., 2019: 1157). As customer feedback may cause lasting damage to a companyâs reputation and may influence the purchasing decisions of other consumers (e.g., Vermeulen and Seegers, 2009; Von Helversen et al., 2018), most companies make use of webcare, a communicative approach defined by van Noort and Willemsen (2012: 133) as âthe act of engaging in online interactions with (complaining) consumers, by actively searching the web to address consumer feedbackâ. In addition to customer care, webcare may also serve the goal of reputation management, that is, an organizationâs desire to establish or consolidate their positive public self-image (Van Noort et al., 2015; Willemsen et al., 2013). In recent years, discourse-linguistic studies have witnessed a growing interest in the communicative strategies employed by companies when dealing with consumer feedback, as well as those used by both satisfied and dissatisfied customers in expressing their feelings (see, e.g., Decock, 2022). Following-up on these recent developments, our present research consists in a corpus linguistic analysis of online customer service discourse. We will be focusing both on customer tweets, which can be considered as a form of electronic word-of-mouth (eWOM), and on social media managersâ response tweets that address customersâ eWOM.
In the context of the travel industry, research has explored webcare and eWOM, for example, in the airline industry. Vo et al. (2019) documented the impact of airlinesâ corporate social responsibility practices on consumer tweets about service delays. In addition, studies have investigated the efficiency of service managersâ responses to customer complaints and requests (Fan and Niu, 2016), compared the frequency and speed of responses for different British and Irish airlines (Lutzky, 2021), and explored the use of (in)directness, politeness, upgraders and downgraders in Twitter interactions between customers and European airlines (Cavalieri and Corrizzato, 2022). In contrast to the airline industry, little research has been devoted to the use of Twitter in the context of train operating companies, despite a growing interest in railway tweets, for instance from an Artificial Intelligence perspective in engineering (see Dong et al., 2022 for an overview) or in management, as shown in Nisar and Prabhakar (2018), who investigated the relationship between consumer relationship management and the content of English-language tweets using sentiment analysis. In a similar vein, Akhtar and Beg (2020) documented the automatic identification of speech acts, such as complaints, suggestions, and praise, in Indian tweets. Within the field of linguistic pragmatics, Orthaber (2019) presented a qualitative analysis of the discourse features associated with negative online customer feedback, focusing on an aggressive humor style in tweets sent by customers to the Slovenian public transport company. Likewise, Lutzky (2021) studied English and Irish train operating companies and their interactions with customers on Twitter, delineating the discourse of customer service tweets in the respective varieties of English.
Building on a discursive analysis of French, German, and Dutch online business complaints, Decock and Depraetere (2018) proposed a new classification of written complaints that is based on the identification of the constitutive components of a complaint situation and whether or not these components are expressed in the complaint message. This new operational definition of the speech act of complaint was empirically confirmed by Depraetere et al. (2021), and experimentally tested by Ruytenbeek et al. (2023a,b), who investigated the impact of complaint (in)directness on the perception of complaints in terms of negative emotions and face-threat. Ruytenbeek et al. (2023a,b) found, for instance, that the presence of a negative evaluation increases perceived customer dissatisfaction. They also showed that the degree of perceived dissatisfaction varies according to how the negative evaluation is achieved: compared to evaluative adjectives, perceived dissatisfaction grows weaker in the case of emoji depicting negative emotions, such as the angry and sad faces emoji.
Depraetere et al. (2021) carried out a small-scale quantitative investigation of linguistic (in)directness in 200 Twitter complaint threads posted in September and October 2018, which involved complaint interactions between the Belgian and French national railway companies and their customers. Their comparative analyses revealed a number of statistically significant differences in terms of the frequencies of complaint component realizations between the two datasets. First, they found that the complainable was more often expressed by means of what and why interrogative sentences in the Belgian tweets compared to the French ones. Second, among the variety of realizations of customer dissatisfaction, ironic utterances and negative emoji/emoticons were more frequently present in the French tweets than in their Belgian counterparts. Third, regarding the reference to the institution responsible for the complainable, Depraetere et al. (2021) observed that this complaint component was more often present in the French tweets which also more often featured multiple realizations of this component. These realizations included, for instance, mentioning the name of the company after the @/# sign and addressing the company by using an imperative construction or interjection (e.g., Dites). When it comes to the discourse strategies used by the service managers of the two train operating companies, the study found that, while displays of empathy were used to the same extent in both datasets, apologizing and explaining what happened occurred more often in the responses from Belgian service managers. However, these differences concerning corporate tweets were not reported to be statistically significant and the specific lexical choices made by customer service managers when apologizing and providing explanations fell beyond the scope of their article.
Depraetere et al.âs study (2021) contributed to cross-linguistic pragmatic research by unraveling differences between French-French and Belgian-French online complaint behaviour and service managersâ response to this negative feedback. As aptly pointed out by Depraetere et al. (2021), the interpretation of such cross-linguistic differences should be done with caution, as it is difficult to distinguish them from differences in the companiesâ customer-oriented policies (Blodgett et al., 2006). It is indeed plausible that different communicative strategies are the result of a complex interplay between, on the one hand, cultural norms and values, and, on the other hand, guidelines regarding customer feedback management. At the same time, the study underlines that it is rather unlikely that Francophone Belgians and French have a culturally different complaint behaviour or a different way of reacting to customer complaints. Instead, the results of Depraetere et al.âs (2021) comparative analyses could indeed be explained by differences in the response strategies adopted by the two professional communities â a hypothesis that should be investigated in a more systematic manner using, for example, an ethno-pragmatic approach probing into the guidelines of train operating companies in different countries.
With the exception of Depraetere et al.âs study (2021), no comparative linguistic research has been conducted either on company or customer tweets in different varieties of French. Despite the insights provided by their study, several questions remain unanswered, mainly due to the small size of their dataset (i.e., a total of 200 complaint Twitter threads). For example, would the differences found between the French and Belgian samples regarding the most frequent strategies used to express dissatisfaction be verified in a larger dataset? Moreover, the analysis of a small number of complaints does not exhaust the variety of linguistic expressions that can be used by (dis)satisfied customers. More generally, while the majority of previous studies have examined the linguistic features of negative customer feedback such as complaints (see e.g., Decock and Depraetere, 2018; Vásquez, 2011; 2014), positive feedback has been largely ignored (an exception is Lutzky, 2021). Finally, most previous linguistic research of the use of eWOM and webcare on social media consists of small-scale, mainly qualitative analyses of messages and C2B interactions (but see Lutzky, 2021), with a need for more quantitative approaches to this business communication genre, especially in the case of tweets from French speaking customers. In the current study, we will address several of these research gaps by exploring Twitter-based customer service in French in three different Francophone countries, using a corpus-linguistic methodology.
3 Data and Methodology
In this article, we focus on the national train operating companies in three European countries where French has the role of an official or national language: Belgium, France and Switzerland. The SNCB (Société Nationale des Chemins de Fer Belges), also known by its Dutch name NMBS (Nationale Maatschappij der Belgische Spoorwegen), is the national railway operating company in Belgium, which was founded in 1926 by the Belgian government. By 1958, the railway had been fully nationalised and has since been under government operation,2 with the companyâs headquarters in Brussels. SNCF (Société Nationale des Chemins de Fer Français) was formed in 1938.3 Operating out of their headquarters in Paris, SNCF has developed into SNCF Group, which is the top rail freight operator in France and ranks second for rail logistics business in Europe.4 Swiss Federal Railways (Chemin de Fer Fédéraux Suisse, CFF) was founded in 1902,5 and has since been operating, with its headquarters in Bern.6 Upon being founded, CFF was a government institution, but in 1999, the company became a special stock corporation, with its shares being held by the 26 Swiss cantons and the Swiss Confederation.7 At present, it is the largest rail and transport company in Switzerland.8
In this study, we explore the customer service interactions of these train operating companies on the social networking platform Twitter, which was renamed X in July 2023, after its takeover by Elon Musk (Milmo, 2023). Twitter is a platform that allows its users to exchange short messages of 280 characters in length, which were originally known as tweets and are today referred to as posts, that are public by default. Before the rebranding as X, Twitter had become established as a popular site for customer communication that has been used for customer service interactions, with its communicative affordances such as retweets and hashtags increasing the spread of eWOM and the reply feature facilitating interactive exchanges.
We are interested in the linguistic features of customersâ and social media managersâ tweets, as well as in possible cross-linguistic differences between the three train operating companies studied. Using the Twitter API, we therefore retrieved all French language tweets customers addressed to the CFF, SNCB and SNCF,9 as well as the companiesâ replies which were posted between January 1, 2018, and December 31, 2022. Retweets were excluded from the data. As we apply a corpus linguistic methodology in this study, our focus is on the verbal characteristics of tweets, and so multimodal content such as images and videos does not form part of our corpus. It should be noted that not all tweets that we collected contained verbal material, especially tweets from customers, some of which consisted of punctuation marks or emoji only.
As indicated in Tables 1 and 2, The French Train Twitter Corpus (FTTC) comprises 523,992 tweets and a total of 11.7 million words, which were posted by the train operating companies and their customers between 2018 and 2022. The largest subcorpus consists of the tweets directed at and posted by the SNCF: it includes 433,415 tweets and a total of 9.6 million words; the majority of these tweets originated from the company SNCF, i.e., 54% while the remaining 46% were posted by customers. This difference is strongest in the case of the SNCB subcorpus, with 60% of the total being tweets shared by the company. While customers tweet less than the three companies studied, their average tweet length is similar (22.5 words for corporate response tweets vs. 22 words for customers, respectively). As is visible from Tables 1 and 2, the FTTC is not optimally balanced, as the SNCF subcorpus is more than five times the size of the SNCB one, which itself is more than ten times the size of the CFF subcorpus. These asymmetries may be due to the respective density of the railway network in these three countries, but it could also be linked to the fact that France features among the top twenty countries worldwide when it comes to the total number of Twitter users,10 whereas Belgium and Switzerland do not. While these differences were unavoidable given the corpus compilation method, which aimed at including all tweets posted by and addressed to the specific companies during a certain timespan, they should nevertheless be borne in mind when considering the results of the following analyses.
Word count for the French Train Twitter Corpus
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
Tweet count for the French Train Twitter Corpus
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
Using the software WordSmith Tools (Scott, 2022), we carried out different types of corpus linguistic analysis centred around an analysis of keywords. A keyword analysis involves a target corpus that is compared to a reference corpus with the aim of uncovering words that appear more frequently than expected in the target corpus and are thus âkeyâ (Scott, 2010; Scott and Tribble, 2006). In other words, keywords allow insights into those words that characterise the target corpus and distinguish it from the reference corpus. This analysis enabled us to identify the linguistic features typical of customersâ and companiesâ tweets, providing insights into the language use of customers and social media managers when interacting with each other.
4 Findings and Discussion
To uncover the main linguistic and communicative focus of our three corpora comprising interactions between customers and the train operating companies CFF, SNCB and SNCF on Twitter, we carried out keyword analyses. To this end, we created two subcorpora for each train operating company: one with all of the corporate response tweets and one including all customer tweets addressed to the respective company. We chose not to study companiesâ original tweets, which include, for example, updates posted by the train operating companies or marketing related posts, as we wanted to focus on the interactions between companies and their customers. In Section 4.1, we first discuss the top twenty keywords for the Swiss corporate tweets, posted by CFF. To uncover cross-regional similarities and differences between the Swiss, Belgian and French corporate responses, we then compare these keywords to those of the Belgian (SNCB) and French (SNCF) corporate tweets in turn. In Section 4.2, we adopt a similar approach for customer tweets.
4.1 Company Tweets
In our first keyword analysis, the results of which are displayed in Table 3, we used each of the three subcorpora of corporate response tweets in turn as our target corpus and compared it to the respective subcorpus of customer tweets, acting as our reference corpus. For example, to uncover the keywords of CFF corporate tweets, we used the CFF corporate response tweet subcorpus as our target corpus and compared it to the CFF customer tweet subcorpus, acting as our reference corpus. We repeated this same procedure for all three corpora and the results of our analysis are given in Table 3, which shows the top 20 keywords in the three corporate response subcorpora.
Top 20 keywords in the CFF, SNCB and SNCF corporate subcorpora
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
The examination of the top twenty keywords for corporate response tweets in the CFF corpus reveals that apologies are prominent in social media managersâ replies to customers. The top two keywords are navré and désolé (both lemmatised forms including also the feminine and plural variants, which are used to say âsorryâ). In addition, the verb forms (je mâ) excuse and (nous nous) excusons for the first person singular and plural (âto apologiseâ) indicate that staff apologise, also in the name of the company, the adverbial malheureusement (âunfortunatelyâ) expresses regret, and the noun désagréments (âinconveniencesâ) at rank seventeen is a typical collocate of apology constructions, denoting the inconvenience caused. Examples (1) and (2) illustrate several of these keywords, which tend to be followed by the causes of complaint, such as train delays and the lack of available seats, and often include an explanation as to why they occurred, such as a technical problem (see also the keyword technique at rank thirteen).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
In addition, the top twenty keywords in CFFâs corporate response tweets include several forms that pertain to customer service interactions. Customers are politely asked to take action (see the keywords veuillez âpleaseâ and prie âaskâ) and are advised to contact a specific service or staff directly (see directement âdirectlyâ and contact). In example (3), for instance, a passenger is advised to ask about the departure times of replacement busses directly at the station. As illustrated in (4), social media managers also tend to acknowledge the feedback they have received from customers and promise to forward it to the parties concerned (see the keywords feedback, transmettre âto pass onâ and concerné âconcernedâ). Apart from that, they also request information from customers, for example when asking about the specific service they are referring to, as in (5) including the keyword agit, which signals aboutness, and they provide customers with advice, for instance by using the keyword recommande ârecommendâ. Finally, example (5) also illustrates the keywords clientèle and dialogueclientele, which are used to refer to CFFâs customer service staff that customers can contact via a hotline to discuss their specific travel needs and issues.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
When comparing the top twenty keywords for CFF to SNCB in Table 3, one notices a difference concerning keywords related to the speech act of apologising. While there is some overlap in that the keyword désagréments features in both lists, for SNCBâs corporate response tweets words such as désolé, navré or sâexcuser are not among the words with the highest keyness values. However, the number one keyword hélas is an interjection that is used to express regret, similar to English âalasâ, and that comes close to the meaning of malheureusement found among the top keywords for CFF, which corresponds to English âunfortunatelyâ or âsadlyâ. Examples (6) and (7) illustrate the use of hélas in the SNCB subcorpus and they include the phrase suite à (âfollowing fromâ) and the negation pas (ânotâ), with which hélas has a strong collocational relationship (a z-score of 22.92 and 7.60 respectively).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
When studying the customer service related keywords in Table 3 in more detail, it turns out that the form prie is also a top keyword for SNCB, having the second highest keyness value. However, when expanding the context and exploring the use of this form in concordance lines, one notices that it almost exclusively appears as part of the construction je vous en prie (âyou are welcomeâ). While it is used mainly by social media managers to ask customers to take specific action in the CFF subcorpus, in the SNCB data it is used by social media managers to acknowledge customersâ appreciation, as illustrated in (8).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
A keyword that is used in a similar manner in both the CFF and SNCB subcorpora is feedback. In corporate tweets for SNCB, its top collocates include the verbs remonter and transmettre (âto pass onâ; see the keywords remonte, remonté, transmets and transmis) and it is thus used by social media managers to assure customers that they will pass on their feedback, for example, to the respective team as in (9). Similar to the CFF, SNCBâs social media managers also engage in requesting information from customers and providing them with details about their travel experience. This is reflected in the keywords parlez (âspeakâ) and souhaitez (âwishâ) which are used by social media managers to ask customers questions, for example, about the exact route they want to travel, as in (10).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
At the same time, the verb souhaitez appears frequently in the construction si vous souhaitez (âif you wishâ), which is the most frequent three-word cluster in which the keyword occurs. Si vous souhaitez introduces a conditional clause explaining how customers can carry out a specific action. In addition, the keyword trouverez (âfindâ) indicates that social media managers inform customers where they can find specific information, for example about possible changes to their train journey, as reflected in the keywords possibles and adaptations (âchangesâ). In addition, SNCBâs social media managers also tend to communicate that they will enquire about the issues passengers are experiencing with the promise to get back to them in a couple of moments, as illustrated in (11) through the keywords renseigne (âenquireâ) and instants (âmomentsâ). Their following tweet, in which they share the respective information, is often introduced by the phrase me revoici (âme againâ) which conveys that they are back to update passengers on a specific concern, such as a train not stopping at a station due to a technical problem, as in (12).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
The top keywords in corporate response tweets by SNCF share some similarities to CFF and SNCB as Table 3 shows. They include forms that pertain to the speech act of apologising, such as désagrément (âinconvenienceâ) and regrettons (âwe regretâ), as well as the verb transmettons (âwe transmitâ) which social media managers use when promising customers to pass on messages to the dedicated team. In addition, the construction mettons (quelquâun) dans la boucle (âwe put someone in the loopâ) appears frequently in SNCFâs corporate responses to inform customers that they are forwarding their tweet to a specific department, such as SNCF Gares et Connexions represented by the handle @ConnectGares in (13), which manages SNCF stations.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
Concerning differences between SNCFâs top twenty keywords and those identified for CFF and SNCB, the keyword restons in the construction restons à votre disposition (âremain at your disposalâ) indicates social media managersâ willingness to answer further questions from their customers, as in (14). Furthermore, the keywords ensemble (âtogetherâ) and confidentialité (âconfidentialityâ) reveal that SNCF explicitly encourages customers to contact staff via a private direct message. They do so, on the one hand, by suggesting that they could investigate an issue, for instance relating to their ticket pass, together via direct message, as in (15). On the other hand, they also ask customers to contact them privately to ensure confidentiality and allow for an issue to be investigated in more detail, as in (16).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
Several of SNCFâs top keywords listed in Table 3 cluster and co-occur together in tweets. This is illustrated in (17), where a social media manager explains that requesting a reimbursement (see the keyword demander âto requestâ) or changing a ticket to a different day (échange âchangeâ) is possible free of charge (sans frais) and without additional cost (surcoût). This tweet was posted in December 2019, when SNCF experienced a longer period of industrial action (Chrisafis, 2019) that resulted in social media managersâ having to repeatedly answer customersâ questions about these topics. These strike related words are therefore key because they were used more frequently than expected in SNCFâs tweets during the time of industrial action in late 2019 and early 2020. Likewise, the top four keywords in Table 3 cluster in the tweet quoted in (18), which was repeatedly posted in June and July 2019 when SNCF engaged in supporting the organisation Capital Filles, an NPO that aims to offer equal educational and professional opportunities to young women from disadvantaged backgrounds. Similarly, the keyword polarSNCF in example (19) does not pertain to the train operating companyâs core business but to an annual award for a crime novel that was launched by the SNCF workforce and that went to the novel receiving most votes from the public.11 The keywords in SNCF corporate response tweets thus include next to customer service related keywords also ones that pertain to exceptional circumstances, such as industrial action, and the companyâs engagement with topics and values pertaining to diversity and culture.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
4.2 Customer Tweets
In our second keyword analysis, the results of which are displayed in Table 4, we replicated the method previously applied to customer service managersâ tweets and used it to analyse the linguistic content of customersâ tweets. Table 4 thus lists the top 20 keywords in the CFF, SNCB and SNCF customer subcorpora.
Top 20 keywords in the CFF, SNCB and SNCF customer subcorpora
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
Most of the top 20 keywords in the CFF customer tweets refer to the variety of issues customers experience. The keyword with the second highest keyness value is the negative particle rien (ânothingâ), which indicates that customers deplore the absence of certain features, as in (20), where rien appears twice in a tweet criticising that CFF is not implementing enough safety measures to protect its passengers during the COVID-19 pandemic. In addition, customers complain about the fact that they have to stand on crowded trains, as indicated by the keyword debout (âstandingâ, rank 9 in Table 4). The verb payer, in the majority of its uses, pertains to customers complaining about high prices that do not match the service provided. This is illustrated in (21), where payer co-occurs with the keyword honte (âshameâ). Other keywords related to issues with the railway services are gens (âpeopleâ) and semaines (âweeksâ), as in example (22), where a customer complains about train delays that have been going on for several weeks and crowded trains with people having to sit on the floor, also in first class. These keywords show that customers are concerned about time and the duration of certain issues and they include their fellow passengers, which allows them to amplify the scale of their complaint.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
In addition, Swiss customers routinely express their disappointment with CFFâs service provision using the keyword dommage (âpityâ) and they evaluate their travel experience. When studying the concordance lines in which the keyword bravo appears, it turns out that this expression, which is generally associated with the speech act of congratulating, is used in a sarcastic manner in the majority (80%) of its uses in the CFF customer subcorpus. This is the case in (23), where the noun bordel (âmessâ) and the phrase en avoir marre (âto be fed upâ) contribute to the interpretation of bravo as sarcastic. This finding for Swiss Twitter data aligns with Depraetere et al.âs (2021) study of Belgian and French data, which found that customer dissatisfaction toward railway companies is regularly expressed using sarcasm.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
While around half of the top twenty keywords for CFF in Table 4 relate to issues customers are experiencing and the complaints they voice, it also features the somewhat unexpected keyword semble (âseemsâ), an epistemic modal of uncertainty, which in the context of customer tweets, serves to mitigate the severity of the complainable. In example (24), for instance, a customer points out that neither of the two ticket validators seems to be working (see also the keyword quai âplatformâ). By including the modal semble, the message conveyed is softened and it is rendered more polite.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
The top keyword in CFF customer tweets is comment, which indicates that CFF customers are interested in the way a certain situation came about or may be resolved. Comment is thus used frequently in requests for information and so are the keywords savoir âto knowâ and its second person plural form savez, which ask social media managers to share specific information (e.g., jâaimerais savoir âIâd like to knowâ). At the same time, these keywords appear in the context of complaint situations, as in (25) where a customer asks how they may be reimbursed.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
In addition, the keywords parle (1st or 3rd person singular of infinitive parler âto talkâ) and dit (3rd person singular of infinitive dire âto sayâ) indicate that customers clarify what they are talking about (e.g., je parle dâun billet de train CFF âIâm talking about a CFF train ticketâ) or report on the information they have received (e.g., le site CFF dit âthe website saysâ). Dites is mostly used to address service managers and functions as a discourse marker with a meaning similar to hey, while the keyword ok acknowledges receipt of information.
When comparing the top 20 keywords for CFF to those for SNCB in Table 4, it turns out that there is some overlap. The top keyword for CFF and SNCB is the same, the adverbial comment, and so are several keywords referring to the issues experienced by customers, i.e., gens, debout, dois and dommage. Customers tweeting SNCB also use the negatively evaluative word honte to describe the companyâs service provision and the exclamation bravo, which as before also shows sarcastic uses in this data set (in more than 60% of its occurrences). As for the CFF, the keyword min at rank 3, which is an abbreviation for minutes, indicates that time is also a concern for SNCB customers, who inquire, for example, if trains that are delayed considerably will be running at all, as in (26).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
This is also reflected in the keyword combien which appears in the three-word cluster combien de temps (âhow much timeâ) in almost half of its occurrences (49%). In addition to keywords signalling the speech act of requests, such as combien, the top 20 customer keywords for the SNCB include further interactional features, such as the abbreviation bjr, which stands for the greeting bonjour (âhelloâ), as well as the discourse marker hein (âisnât itâ), which is used to elicit a response to a question or ask for confirmation of a point made.
Further causes for complaint that the top 20 keywords reveal include the SNCB app (see appli at rank 6) that is not always working reliably, as well as the shortage of train carriages (see wagons at rank 7) which leads to crowded trains (see bondé at rank 15). In (27), for instance, a customer complains about a train being crowded as it consists of only 3 instead of the usual 9 carriages during rush hour. They underline their frustration through the use of the English phrase as fuck, describe their experience as une dinguerie (âlunacyâ), and express their hatred for the company (je vous hais âI hate youâ). In addition to honte, the frustration that passengers feel about these complainables is reflected in the keyword marre, which expresses customer dissatisfaction (en avoir marre âto be fed upâ) and the derogatory merde (âshitâ), as illustrated in (28), where a customer deplores constant problems with and rising prices for SNCBâs services.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
Sympa, which appears at rank 18 in Table 4, does not unanimously convey a positive connotation in SNCB customer tweets. In fact, it is used in around one fifth of its occurrences to make suggestions and indicate which steps could be or could have been taken by the company, as in (29). At the same time, around 40% of all uses of sympa appear in a negative context (e.g., pas sympa) or with a sarcastic meaning, as in (30), where a customer expresses the opposite of what they are saying as they clearly did not have a good start to their day.
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
For the SNCF, the top 20 keywords in customer tweets show some similarities to those identified for CFF and SNCB. In fact, twelve keywords also appear among the top keywords of the CFF or SNCB, indicating that they contribute towards the discourse of customer service tweets in French. These twelve keywords, several of which were illustrated and discussed above, pertain to train travel and complainables, such as gens, usagers, wagon, contrôleur, paye and payé, verbs indicating what customers want or have to do (see veux, dois) or reporting on the information they have received (see dit), as well as examples of derogatory or sarcastic language use (e.g., honte, merde, bravo). A striking difference between SNCF customer tweets and the other two subcorpora concerns the higher keyness value of derogatory expressions, with honte appearing at rank 5 in Table 4 and merde at rank 8. In addition, bande (de) at rank 9 is a productive nominal expression which translates as âbunch ofâ and is usually followed by an insult, as in (31).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
In this example, a customer refers to SNCF as a bunch of dogs, complaining about trains being delayed or cancelled, and the company going on strike regularly. In fact, the keyword grève (âstrikeâ) at rank 3 in Table 4 introduces a new complainable that is unique to the SNCF when considering the top keywords across the three companiesâ subcorpora. In addition, the keywords rentrer (âreturnâ) and attends (the first person singular of attendre âto waitâ) pertain to further consequences or causes of complaint.
However, the top twenty keywords in SNCF customer tweets do not exclusively include forms pertaining to complaints and expressing negative evaluation. In addition, there are also interactive keywords, such as the second person singular pronoun tu, which is generally used to address the company in an informal manner. The keyword coup appears, in almost half of its occurrences, as part of the two-word cluster du coup, a discourse marker with a meaning similar to âthat is why.â The keyword super, finally, may be used with a sarcastic tone, similar to the keyword bravo, as illustrated in (32), where a customer refers to a train delay as super and underlines their discontent through the use of angry emojis. At the same time, super can also be part of a thanking expression in response to a tweet from a service manager, with merci being its top collocate in one position to the right (z-score of 18.15), as illustrated in (33).
Citation: Contrastive Pragmatics 6, 3 (2025) ; 10.1163/26660393-bja10130
5 Conclusion
In this study, it was our aim to identify the main linguistic and communicative characteristics of the Swiss-, French- and Belgian-French Twitter interactions between European train operating companies and their customers. To this end, we compiled three corpora of tweets posted by and addressed to the CFF, SNCB and SNCF between 2018 and 2022, creating subcorpora for corporate and customer tweets. Using a corpus linguistic methodology, we carried out keyword analyses to uncover the words that characterise customer tweets addressed to the three train operating companies as well as those that are typical of corporate replies.
Studying the top 20 keywords of corporate response tweets revealed several tendencies regarding the French-speaking discourse of customer service tweets. We found that, while all three subcorpora include expressions of regret and apologies, such as désagréments (âinconveniencesâ), these are more prominent for the CFF, which indicates that the service managers of the CFF have a stronger tendency to apologize for inconveniences caused in comparison with their French and Belgian French counterparts. Several keywords pertain to customer service interactions and indicate that social media managers working for the CFF and SNCB use features of polite language use when requesting information from passengers and providing advice. While all three companies promise to forward customer feedback to the relevant teams, the SNCF differs from the CFF and SNCB in several regards. The keyword analysis showed that SNCF staff often encourage customers to contact them via direct message and thus attempt to move the conversation from a public to a private level. This finding aligns with the results of Orthaber (2023: 285â325), who also observed this practice in the responses of a Slovenian train company to its customers on Facebook. We also saw that the SNCF uses standard responses, for example during times of industrial action, with several keywords clustering in the same tweet, and they do not only use Twitter for customer service purposes. In addition, they engage in the webcare goal of reputation management (van Noort et al., 2015) by discussing CSR initiatives and aim to generate user engagement through, for example, issuing awards for the best crime novel.
For the top 20 keywords in customer tweets, our keyword analyses revealed that there is some overlap between tweets addressed to the CFF, SNCB and SNCF. Customers of all three companies tend to mention specific complainables in their tweets, such as crowded trains or high ticket prices, they express negative evaluation through the keyword honte (âshameâ), and they use sarcasm in their tweets (e.g., bravo). While Depraetere et al. (2021) observed sarcastic utterances to be more frequent in French than in Belgian customer tweets, we provide evidence that positive evaluative words, such as bravo and sympa, are often used with a sarcastic intent to convey dissatisfaction in all three subcorpora. However, the SNCB customer tweets feature more keywords that are derogatory in nature, compared to the CFF and SNCB, and they also have higher keyness values. The CFF and the SNCB, on the other hand share the same top keyword comment (âhowâ), which indicates that their customers are interested in knowing how an issue came about and may be solved (see also combien âhow muchâ for the SNCB). Our results thus confirm and complement previous findings by Depraetere et al. (2021), who found that complainable-related questions including what and why were more frequent in the Belgian (SNCB) customer tweets compared to the French (SNCF) ones.
Building on Depraetere et al. (2021), the present study contributes to cross-linguistic pragmatic research by uncovering differences between Swiss-, French- and Belgian-French online feedback and social media managersâ responses to this feedback. As aptly pointed out by Depraetere et al. (2021), one ought to be careful when trying to interpret such cross-linguistic differences, as they cannot easily be disentangled from differences in the companiesâ policies bearing on service interactions. Nevertheless, our study makes an important contribution to practice, with a view to improving customer service provision. For example, by focusing on the most significant complainables (as shown in our keyword analyses) rather than having to navigate through a very large number of tweets, companies will be able to identify and address those aspects of their service provision that customers deem most problematic. Our findings also show that the complainables are similar across the three companies and French-speaking countries studied, indicating that customers share the same concerns and highlight comparable issues when contacting customer service on Twitter. In addition, our analyses of the keywords used by the service managers highlight the relative prominence of the communicative strategies they adopt, allowing the companies studied to potentially adjust their response policies in the future, for example by paying more attention to empathic language use or interactional dynamics. Several of the keywords uncover corporate policies of asking customers to contact them via direct message or get in touch with a specific service team directly. As previous research pointed out (e.g., Van Herck et al. 2020), such communicative strategies are often not appreciated by customers, thus opening up a potential opportunity for adopting more efficient means of interaction. At the same time, our results have revealed that while companies such as the SNCF engage in the full range of approaches to webcare, including marketing and reputation management, the CFF and SNCB mainly focus on using Twitter for customer service interactions. There is thus room for expansion in their practice to benefit from the diverse possibilities that webcare has to offer.
While our research has provided new insights into the French-speaking customer service interactions of three European train operating companies online, there is a need for future studies to expand on our approach. Further research could, for instance, explore the perception of specific words and expressions, such as derogatory language use, by fellow customers or customer service managers. Using examples from our dataset, experimental studies could also investigate the impact of different realizations of service managersâ apologies on dependent variables such as, for instance, perceived sincerity and the effectiveness of the communicative strategy. In addition, our corpora could facilitate the testing of sarcasm detection algorithms (bravo, super) by using hashtags (#honte âshameâ). Finally, we believe that additional linguistic insights into online customer discourse could be gained by running specific queries (e.g., the modal uses of savoir in Belgian French compared to pouvoir in France) and incorporating other types of corpus linguistic analyses, such as multi-word expressions, to explore the use of template responses in French-speaking customer service interactions.
References
Akhtar, Nadeem, and Mira Mohd. Sufyan Beg. 2021. Railway Complaint Tweets Identification. In: Neha Sharma, Amlan Chakrabarti, Valentina Emilia Balas and Jan Martinovic (eds.), Data Management, Analytics and Innovation. Advances in Intelligent Systems and Computing, vol. 1174. Springer, 195â207.
Argenti, Paul A. 2006. How technology has influenced the field of corporate communication. Journal of Business and Technical Communication 20(3): 357â370. http://dx.doi.org/10.2139/ssrn.828926.
Béal, Mathieu, and Yany Grégoire. 2022. How do observers react to companiesâ humorous responses to online public complaints? Journal of Service Research 25(2): 242â259. https://doi.org/10.1177/1094670521989448.
Blodgett, Jeffrey G., Donna Hill, and Aysen Bakir. 2006. Cross-cultural complaining behavior? An alternative explanation. Journal of Consumer Satisfaction, Dissatisfaction and Complaining Behavior 19: 103â117.
Cavalieri, Silvia, and Sara Corrizzato. 2022. âWeâre sorry for any inconvenience caused.â Pragmatic aspects of handling complaints in customer-airline company tweets. Lingue e Linguaggi 53: 101â115. https://doi.org/10.1285/i22390359v53p101.
Cenni, Irene and Patrick Goethals. 2017. Negative hotel reviews on TripAdvisor: A cross-linguistic analysis. Discourse, Context & Media 16: 22â30. https://doi.org/10.1016/j.dcm.2017.01.004.
Chrisafis, Angelique. 2019. Christmas travel in France likely to be disrupted as strikes continue. The Guardian, 19 December. https://www.theguardian.com/world/2019/dec/19/christmas-travels-likely-to-be-disrupted-as-french-strikes-continue.
Collins, Luke C. 2019. Corpus Linguistics for Online Communication. A Guide for Research. London: Routledge.
Decock, Sofie. 2022. Discursive approaches to webcare: A closer look at apologies, conversational human voice, legitimation, and emotion regulation. Discourse Context & Media 45: 100575. https://doi.org/10.1016/j.dcm.2021.100575.
Decock, Sofie, and Ilse Depraetere. 2018. (In)directness and complaints: A reassessment. Journal of Pragmatics 132: 33â46. https://doi.org/10.1016/j.pragma.2018.04.010.
Dong, Kaitai, Igor Romanov, Colin Mclellan and Ahmet F. Esen. 2022. Recent text-based research and applications in railways: A critical review and future trends. Engineering Applications of Artificial Intelligence 116: 105435. https://doi.org/10.1016/j.engappai.2022.105435.
Fan Ying and Run Hong Niu. 2016. To tweet or not to tweet? Exploring the effectiveness of service recovery strategies using social media. International Journal of Operations & Production Management 36(9): 1014â1036. https://doi.org/10.1108/IJOPM-10-2013-0461.
Feng, Wei, and Ren, Wei. Impoliteness in negative online consumer reviews: A cross-language and cross-sector comparison. Intercultural Pragmatics 17(1): 1â25. https://doi.org/10.1515/ip-2020-0001.
Howard, Jeffrey M. 2020. Trains, Twitter and the social licence to operate: An analysis of Twitter use by train operating companies in the United Kingdom. Case Studies on Transport Policy 8(3): 812â821. https://doi.org/10.1016/j.cstp.2020.06.002.
Jansen, Bernard J., Mimi Zhang, Kate Sobel, and Abdur Chowdury. 2009. Twitter power: Tweets as electronic word of mouth. Journal of the American Society for Information Science and Technology 60/11: 2169â2188. https://doi.org/10.1002/asi.21149.
Kietzmann, Jan H., Kristopher Hermkens, Ian P. McCarthy, and Bruno S. Silvestre. 2011. Social media? Get serious! Understanding the functional building blocks of social media. Business Horizons 54(3): 241â251. https://doi.org/10.1016/j.bushor.2011.01.005.
Lutzky, Ursula. 2021. The Discourse of Customer Service Tweets. London: Bloomsbury.
Lutzky, Ursula, and Andrew Kehoe. 2022. Using corpus linguistics to study online data. In: Camilla Vásquez (ed.), Research Methods for Digital Discourse Analysis, 219â236. Bloomsbury.
Milmo, Dan. 2023. Elon Musk reveals new Twitter logo X. The Guardian, July 24 https://www.theguardian.com/technology/2023/jul/24/elon-musk-reveals-the-new-twitter-logo-x.
Nisar, Tahir M., and Guru Prabhakar. 2018. Trains and Twitter: Firm generated content, consumer relationship management and message framing. Transportation Research Part A: Policy and Practice 113: 318â334. https://doi.org/10.1016/j.tra.2018.04.026.
Orthaber, Sara. 2019. Aggressive humour as a means of voicing customer dissatisfaction and creating in-group identity. Journal of Pragmatics 152: 160â171. https://doi.org/10.1016/j.pragma.2018.11.002.
Orthaber, Sara. 2023. (Im)politeness at a Slovenian Call Centre. Advances in (Im)politeness Studies. Cham: Springer. https://doi.org/10.1007/978-3-031-43320-7_7.
Ruytenbeek, Nicolas, Decock, Sofie, and Ilse Depraetere. 2023a. Experiments into the influence of linguistic (in)directness on perceived face-threat in Twitter complaints. Journal of Politeness Research 19 (1): 59â86. https://doi.org/10.1515/pr-2019-0042.
Ruytenbeek, Nicolas, Decock, Sofie, and Ilse Depraetere. 2023b. The influence of linguistic choices on perceived face-threat in Twitter complaints: An experimental approach. Journal of Politeness Research 19 (1): 87â122. https://doi.org/10.1515/pr-2021-0031.
Scott, Mike, and Christopher Tribble. 2006. Textual Patterns. Key Words and Corpus Analysis in Language Education. Amsterdam: John Benjamins.
Scott, Mike. 2010. Problems in investigating keyness, or clearing the undergrowth and marking out trails. In: Marina Bondi and Mike Scott (eds.), Keyness in Texts. Amsterdam: John Benjamins, 43â57. https://doi.org/10.1075/scl.41.04sco.
Scott, Mike. 2022. WordSmith Tools version 8 (64 bit version) Stroud: Lexical Analysis Software.
Van Herck, Rebecca, Sofie Decock, and Bernard De Clerck. 2020. âCan you send us a PM please?â Service recovery interactions on social media from the perspective of organizational legitimacy. Discourse, Context & Media 38: 100445. https://doi.org/10.1016/j.dcm.2020.100445.
Van Noort, Guda and Lotte M. Willemsen. 2012. Online damage control: The effects of proactive versus reactive webcare interventions in consumer-generated and brand-generated platforms. Journal of Interactive Marketing 26: 131â140. https://doi.org/10.1016/j.intmar.2011.07.001.
Van Noort, Guda, Lotte M. Willemsen, Peter Kerkhof, and Joost W. M. Verhoeven. 2015. Webcare as an integrative tool for customer care, reputation management, and online marketing: A literature review. In: P. J. Kitchen and E. UzunoÄlu (eds.), Integrated Communications in the Postmodern Era. London: Palgrave Macmillan, 77â99. https://doi.org/10.1057/9781137388551_4.
Vargo, Chris, Harsha Gangadharbatla, and Toby Hopp. 2019. eWOM across channels: comparing the impact of self-enhancement, positivity bias and vengeance on Facebook and Twitter. International Journal of Advertising 38(8): 1153â1172. https://doi.org/10.1080/02650487.2019.1593720.
Vásquez, Camilla. 2011. Complaints online: The case of TripAdvisor. Journal of Pragmatics 43: 1707â1017. https://doi.org/10.1016/j.pragma.2010.11.007.
Vásquez, Camilla. 2014. The Discourse of Online Consumer Reviews. London: Bloomsbury.
Vermeulen, Ivar E., and Daphne Seegers. 2009. Tried and tested: The impact of online hotel reviews on consumer consideration. Tourism management 30(1): 123â127. https://doi.org/10.1016/j.tourman.2008.04.008.
Vladimirou, Dimitra, Juliane House, and Dániel Kádár. 2021. Aggressive complaining on Social Media: The case of #MuckyMerton. Journal of Pragmatics 177: 51â64. https://doi.org/10.1016/j.pragma.2021.01.017.
Vo, Tam Thien, Xinning Xiao, and Shuk Ying Ho. 2019. How does corporate social responsibility engagement influence word of mouth on Twitter? Evidence from the airline industry. Journal of Business Ethics 157(2): 525â542. https://doi.org/10.1007/s10551-017-3679-z.
Von Helversen, Bettina, Katarzyna Abramczuk, WiesÅaw KopeÄ, and Radoslaw Nielek. 2018. Influence of consumer reviews on online purchasing decisions in older and younger adults. Decision Support Systems 113: 1â10. https://doi.org/10.1016/j.dss.2018.05.006.
Willemsen, Lotte, Peter C. Neijens, and Fred A. Bronner. 2013. Webcare as customer relationship and reputation management? Motives for negative electronic word of mouth and their effect on webcare receptiveness. In: Sara Rosengren, Micael Dahlén, and Shintaro Okazaki (eds), Advances in Advertising Research (Vol. IV) The Changing Roles of Advertising. Wiesbaden: Springer Gabler, 55â69. https://doi.org/10.1007/978-3-658-02365-2_5.
Biographical Notes
Ursula Lutzky is Associate Professor at the Vienna University of Economics and Business. Her research interests include business communication, discourse studies, pragmatics, and corpus linguistics. Her recent work has focused on the study of language in large corpora of digital discourse, such as blogs and microblogs. Her monograph, The Discourse of Customer Service Tweets (2021), explores the linguistic and communicative features used by companies and their customers when engaging in customer service interactions on Twitter.
Nicolas Ruytenbeek is an assistant professor in digital and multilingual communication at the Department of Linguistics (KU Leuven). In his research activities, he addresses the production and comprehension of speech acts in relation to face-threat and (im)politeness in digital settings, weaving together the fields of philosophy of language, pragmatics, psycholinguistics and business communication.
As our data was compiled before the name change from Twitter to X, we will refer to the social networking platform as âTwitterâ throughout our article.
https://www.infrabel.be/en/history (accessed 24 October 2023).
https://www.britannica.com/topic/Societe-Nationale-des-Chemins-de-Fer-Francais (accessed 24 October 2023).
https://medias.sncf.com/sncfcom/finances/Publications_Groupe/sncf-group-mobility-partner.pdf (accessed 24 October 2023).
https://company.sbb.ch/en/the-company/profile/history.html (accessed 24 October 2023).
https://www.sbb.ch/de/meta/legallines/impressum.html (accessed 24 October 2023).
https://company.sbb.ch/en/the-company/profile/history.html (accessed 24 October 2023).
https://company.sbb.ch/en/home.html (accessed 24 October 2023).
We faced an important challenge when classifying SNCF tweets into company and customer tweets. First, upon visual inspection, we noticed that, on some occasions, customer service managers included the @SNCF username in their tweets. This made it difficult to distinguish tweets that were posted by customers from tweets that originated from social media managers. As the French train operating company (SNCF) appears to use multiple Twitter accounts to respond to customer tweets, we decided to use the IDs of all these Twitter accounts to identify the tweets from service managers. To do this, we turned the usernames corresponding to these accounts into a list of Twitter IDâs using the Twitter ID finder online tool (https://twiteridfinder.com).
https://www.statista.com/statistics/242606/number-of-active-twitter-users-in-selected-countries/ (accessed 2 February 2024).
https://www.sncf.com/en/group/culture/celebrating-comics (accessed 20 February 2024).
