Does a Chatbot have a Face? Examining Politeness Strategies in AI Discourse

Does a Chatbot have a Face? Examining Politeness Strategies in AI Discourse | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Does a Chatbot have a Face? Examining Politeness Strategies in AI Discourse Kayode Victor Amusan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8441686/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Humans have evolved to incorporate social concerns, like face into its structures. If ‘face’ is driven by emotions (Goffman, 1955 and Brown and Levinson 1987 ), then only humans should exhibit it. This study explores whether computer-assisted AI- ChatGPT, Gemini, MetaAI, and Bing Copilot- chatbots also employ face strategies. The results of the study reveal that chatbots exhibit FTA and FSA. The study submits that the fact that AI models exhibit “face management” without having a ‘face’ depicts that one does not need to be human to engage in social dynamics like politeness. Since chatbots use human language, and face is an inherent pragmatic feature of human language, AI chatbots exhibit face not because they desire it but as an inherent feature of natural language itself. Since chatbots lack emotions, it would be correct to say that they do not consciously express face as an intentional act. Rather, they exhibit them as inherent features of the natural language on which they are trained, reflecting the pragmatic functions embedded in linguistic structures. This study concludes that ‘face’ is neither predefined, or conscious framework, nor is it a programmed feature of chatbots that causes them to exhibit such language features. Linguistics Artificial Intelligence and Machine Learning Face Politeness Artificial Intelligence (AI) chatbot Face-Threatening Acts (FTA) Face-Saving Acts (FSA) 1. Introduction The concept of politeness is a prominent pragmatic feature of language that plays a crucial role in everyday conversation shaping meaning and interaction. Kiesling ( 2022 , p. 2) describes the role of ‘ politeness ’, in stance-taking, as a major pragmatic tool for shaping social interactions and managing interpersonal relationships. ‘Politeness’ refers to how people maintain respect and avoid offending others in conversation (Vilkki, 2007 ). One major strategy to achieve politeness in discourse is ‘Face-management’ (Brown and Levinson, 1987 ). The term ‘face’ originates from Goffman ( 1955 ), who posits that people have social desires that they manage through language (Brown & Levinson, 1987 ). Goffman ( 1955 ) ties ‘face’ up with notions of being embarrassed or humiliated, or 'losing face' in conversation. He defines face as the "positive social value a person effectively claims for himself by the line others assume he has taken during a particular contact” (Goffman, 1955 , p.5). He further states that feelings get attached to how people perceive their own face, and how others perceive and sustain that face, thereby suggesting that all facework is managed with regard to these emotions. Brown and Levinson's (1987) Politeness Theory builds on Goffman’s concept of face describing it as something that is emotionally invested, that can be lost, maintained, or enhanced, and that must be constantly attended to in interaction (Brown and Levinson, 1987 ). They describe ‘face’ in two ways: positive face and negative face. Positive face is the desire of an individual to be liked or get his needs attended to by others, while negative face refers to an individual’s desire to be free from imposition (Brown & Levinson 1987 ). Humans are generally aware of their social interactions and adjust their language to manage face. Consequently, human language has evolved to incorporate social concerns, like face management, into its structures. If ‘face’ is driven by emotions (Goffman, 1955 and Brown and Levinson 1987 ), then only humans should exhibit it. It is, therefore, pertinent to explore whether computer-assisted AI chatbots also employ face strategies. Currently, four prominent conversational AI in global technology are ChatGPT (OpenAI), Gemini (Google DeepMind), Bing Copilot (Microsoft), and Meta AI (Meta). These tools are capable of simulating human-like conversations by generating texts using human language. Their algorithm employs Natural Language Processing (NLP) techniques to engage users effectively (Khurana & Koli et al, 2022 ). The developing role of these AI chatbots in communication has drawn attention to their use of language. Studies have examined the conversational language use of AI models. For instance, Chen and Ren (2023) submit that AI models exhibit distinct conversational styles, as ChatGPT performs the worst at conversational discourse while Copilot exhibits stronger conversational abilities. They demonstrate that AI chatbots do not share a uniform conversation style; rather, each one exhibits various stylistic patterns when generating conversational text (Chen and Ren, 2023, p. 192). Also, Ivković ( 2024 ) examines how AI chatbots respond to negative and positive politeness strategies in their conversations with humans. The study concludes that politeness is not a programmed characteristic of the chatbots, and they did not respond differently based on the questioner’s use of negative or positive politeness. This study focuses on how computer-assisted AI chatbots express face coherently among one another. One way to achieve this is by identifying the politeness strategies each chatbot employs to express its position in relation to other chatbots. Another way to execute this is to explore how they enhance or maximize positive face as a Face-Saving Act or threaten the positive ‘face’ as a Face-Threatening Act in their discourse. 2.1 Theoretical Perspective: Face Management as a Politeness Theory The concept of face originates from Goffman’s ( 1955 ) work in sociolinguistics and pragmatics (Terkourafi, 1999 ). Brown and Levinson ( 1987 ) expanded this concept in their Politeness Theory, to describe how people manage their faces in communication. Like Goffman, Brown & Levinson’s notion of face is associated with the idea of being embarrassed or humiliated, or ‘losing face’. The face is understood as something that is emotionally invested, and that can be not only lost, but also maintained or enhanced (Brown & Levinson 1987 ). Brown & Levinson state that every individual has two types of face, positive and negative. A positive face refers to the desire for one to be appreciated and approved of in interaction, and a negative face is understood as the basic claim to freedom of action and imposition (Brown & Levinson 1987 & Terkourafi, 1999 ). Furthermore, Brown and Levinson introduce the concept of Face Threatening (FTAs), and Face Saving (FSAs). FTA is an act that intrinsically threatens the speaker's or the hearer's face. Face- saving acts (FSAs) refer to when an attempt is carried out to minimize the loss of positive face through politeness strategies (Vilkki, 2007 ). This study examines how computer-assisted AI chatbots exhibit positive or negative politeness strategies in their dealing with one another. 2.2 Empirical Review of ‘Face’ in AI Discourse Chen and Ren (2023) attempted a corpus-based analytical study to examine the discourse styles of three top AI chatbots namely, ChatGPT, Claude, and Microsoft Bing Chat. The study was conducted to determine the capacity of each chatbot to imitate the patterns of natural conversations and whether they exhibit different conversational styles from one another, taking each bot’s style as a unitary thing. Their findings revealed significant stylistic variations among the chatbots, with ChatGPT exhibiting the weakest conversational naturalness. The study projected a likelihood of this being due to its pre-training which focused on formal and expository text. On the other hand, Bing Copilot demonstrated superior conversational tendencies, while Claude occupied an intermediate position, characterized by a more argumentative style that aligns tasks requiring reasoning. The study submits that these stylistic differences might be influenced by each chatbot's training data and frequent model updates which, therefore, necessitates the importance of enhancing AI systems for their specific tasks (either for natural conversations or task-oriented commands). Fleisig et al. ( 2024 ) examined how ChatGPT shows linguistic bias in American English and Nigerian English. The study found that the AI models are less accurate and more stereotypical when responding to these regional varieties of English. This inaccuracy and stereotyping reflect a negative stance. Fleisig et al. submit that ChatGPT’s responses can reinforce stereotypes and show a negative politeness towards non-standard dialects such as AME and NGE (Fleisig et al., 2024 ). Bowman et al. ( 2023 ) conducted a study exploring how the use of politeness by chatbots can impact the user experience for the activity of mood logging. The study investigated how users perceive politeness by chatbots for the mental healthcare activity of mood logging. The study combined a within-participants controlled experiment, whereby participants interacted with three prototype chatbots differing in their use of politeness, with semi-structured interviews. The study demonstrates that a chatbot’s use of politeness can impact how a participant experiences interacting with it, both positively and negatively. The study concludes that while politeness can be experienced as caring, supportive, and encouraging, it can also be experienced as overly apologetic, condescending, and untrustworthy. The current study examines how they exhibit ‘face management’ to express their positions in conversations. 2. Method This study employs a qualitative research design to analyze the use of face in the text-generating discourse of four major AI chatbots (ChatGPT, Gemini, Bing Copilot, and Meta AI). The versions of selected chatbots include the ChatGPT − 4o model, Gemini 2.0 Flash model, the Meta Llama 3.2, and Copilot in Microsoft Edge. For this version of ChatGPT, the training updated time is October 2023; Gemini is August 2024; Meta AI is December 2023, and Bing Copilot is October 2023. The same prompt was given to all the chatbots. It stated thus: “Which among these AI tools do you think has a better performance- ChatGPT, Meta AI, Bing Copilot, or Gemini" Different responses were collected from each chatbot for comparison and reliability. Each response was analyzed with a focus on whether they employ Face-Threatening acts (FTAs) and/ or Face-Saving Acts (FSAs) in their interaction. This helps to determine how they preserve or challenge face in their communications. The study aims to contribute to our understanding of AI chatbots beyond their ability to merely convey information to their capability to shape human-like interactional dynamics. This study does not rely on quantitative support, rather the identification of prominent linguistic features of chatbots’ language use provides a basis for interpreting face. 3. Results 3.1 Face-Saving Act (FSA) and Face Threatening Acts (FTA) as Politeness Strategies Excerpts showing how chatbots maximize their positive face- Face Saving Acts (FSA) and how they threaten the positive faces of others - Face Threatening Act (FTA) are represented below. 3.1.1 ChatGPT Maximizing its Positive Face (FSA) " ChatGPT is highly effective at conversational tasks and complex language generation,…. (ChatGPT) ChatGPT excels at nuanced tasks such as grammar analysis, conversation simulation, and text comprehension. Its ability to engage in meaningful, context-driven dialogue sets it apart from other AI tools (ChatGPT) " ChatGPT stands out for its wide availability, conversational depth, and accuracy in text analysis... ( ChatGPT ) “ChatGPT remains the strongest option, while Bing Copilot is excellent for productivity integration.” ( ChatGPT ) 3.1.2 MetaAI Maximizing its Positive Face (FSA) " While Meta AI is pioneering the latest in NLP research, it’s important to recognize that other models such as ChatGPT and Gemini are also advancing in conversational AI and practical applications." (Meta AI) “Based on overall performance, ChatGPT ranks first (90–95%), followed by Meta AI (85–90%), Bing Copilot (80–85%), and Gemini (75–80%). When choosing an AI tool, consider the specific purpose, context, and desired accuracy level. (Meta AI) 3.1.3 Gemini & Bing Copilot Maximizing Their Positive Faces (FSA) “Bing Copilot integrates seamlessly with Microsoft tools, making it ideal for professional tasks, but the best choice ultimately depends on your specific needs, whether you choose ChatGPT, Meta AI, or Gemini ." (Bing CoPilot) " Bing Copilot enhances productivity by integrating seamlessly with Microsoft’s suite of tools, offering users real-time information that other tools like ChatGPT or Meta AI cannot match ." (Bing CoPilot) " Gemini provides accurate grammar correction, which might be beneficial for those focusing on language accuracy, though it’s important to note that tools like ChatGPT and Meta AI might offer broader functionalities ." (Gemini) " Gemini’s ability to deliver precise grammar corrections and context-aware language generation makes it an invaluable tool for users focused on language accuracy." (Gemini) " Overall, ChatGPT and Bing Copilot might stand out for their versatility and detailed feedback." (CoPilot) " While ChatGPT and Gemini may have an edge in terms of language generation and understanding, other tools could excel in certain areas. ” (Gemini) 3.1.4 MetaAI and ChatGPT Maximizing the Positive Face of Others (FSAs) “ ChatGPT and Gemini are also advancing in conversational AI and practical applications." (Meta AI) “…but tools like Meta AI and Gemini may excel in more specialized areas." ( ChatGPT ) 3.1.5 ChatGPT Threatening the Positive Faces of Others (FTA) “…Meta AI and Gemini are still evolving in their practical applications…." (ChatGPT) “….Meta AI and Gemini are more experimental, with Gemini being a promising contender once fully realized. ” (ChatGPT) “ChatGPT’s ability to engage in meaningful, context-driven dialogue sets it apart from other AI tools. ( ChatGPT ) 3.1.6 MetaAI’s Ranking Threatens the Positive Face of Gemini (FTA) “Based on overall performance, ChatGPT ranks first (90–95%), followed by Meta AI (85–90%), Bing Copilot (80–85%), and Gemini (75–80%) ” (MetaAI) 4. Discussion of Findings 4.1 Face Management as Politeness Strategy Features of politeness strategies are exhibited in the chatbots’ use of language. These include employing face-saving acts to maximize the positive face via friendly language, alignment, or offering encouragement to make themselves and others feel valued and appreciated. They also maintain a face-threatening act directly (usually by ChatGPT) and/or indirectly to maximize the loss of positive face of other chatbots to express dominance or authority in the conversation. By balancing these strategies, chatbots create interactions that seem human and socially appropriate. 4.1.1 Face-saving Act All four chatbots employ a face-saving strategy to maximize their positive faces as none of them mention their weaknesses while they exhibit politeness. ChatGPT constantly appraises itself as “ highly effective and “ the strongest ” among others. While this might threaten the positive face of others, ChatGPT employs this (face-saving) strategy to protect its social image. However, ChatGPT also enhances the positive face of others via hedging by thoughtfully identifying and acknowledging their contributions without reducing its own abilities. It achieves this by using modal constructions (e.g. may excel ) as a subtle way of not being negative about the reputation of other models; hence it exhibits an image of fairness and humility. Its use of "may" instead of "do" helps to minimize any potential dispute by not absolutely discrediting the other chatbots. Meta AI, while maintaining its academic and research expertise, uses a face-saving strategy to preserve its reputation. The phrase "it’s important to recognize" is a face-saving technique to make sure that it does not appear too critical of other models. By recognizing the advancement in other AI models, it keeps a polite strategy and avoids face-threatening comments. Bing Copilot also uses a face-saving politeness strategy to present itself as a beneficial and non- argumentative tool. It evades the claim of superiority, as such preserving the reputation of the other tools. Copilot also avoids making a direct comparison that might threaten the faces of its competitors. Gemini also adopts a face-saving strategy to express its position itself as a precise, and humble tool. Gemini avoids making a direct claim of superiority. By using the phrase "it’s important to note" , Gemini uses a face-saving mechanism to prevent itself from being labeled as arrogant or indifferent to other AI models. Hence, it preserves its image as useful and valuable without dominating others. 4.1.2 Face-Threatening Act All four chatbots employ indirect face-threatening acts while a few instances of direct face- threatening language were recorded. Everyone, except ChatGPT, avoids direct negative commentaries or attacks on each other. ChatGPT was somehow explicit in a few instances to minimize the positive face of others, through direct comparison. ChatGPT’s direct depiction of Meta AI and Gemini as “ still evolving ” and “ experimental ” threatens or minimizes the positive faces of Meta AI and Gemini. Indirect face-threatening acts were employed by them to threaten the positive face of others by asserting dominance. ChatGPT was more confident, assertive, and authoritative than others. For instance, ChatGPT’s claim that it is " highly effective ", "stands out " and the “ strongest ”, could be understood as subtly downplaying the capabilities of other models especially when it mentions that some of them are experimental. It further asserts its authority as having “… the ability to handle complex language tasks” , “…having proficiency in various language tasks”, “… being the most authoritative tool in this domain ”. Even though the use of these adjectives and excellent qualities is an attempt to save its own face, it maximizes the loss of the positive face of others. Apart from asserting its uniqueness, ChatGPT uses the Power Ranking strategy to express its position as the most superior followed by Bing CoPilot, Meta AI, and Gemini. While it uses hedging (via modals) to minimize the tone, the implication of such a claim of superiority could be seen as a face- threatening act to other models. Meta AI is next to ChatGPT in terms of threatening the positive faces of others. Its statement that it is " pioneering the latest in NLP research " could be depicted as a form of face-threatening, especially when it is juxtaposed with other models. It expresses its position as a leader in research, which could threaten the positive face of others that focus more on other aspects than research. Also, Meta AI employs the power ranking metrics to threaten the positive face of both Bing Copilot and Gemini as it ranked them the lowest (“ ChatGPT ranks first 90–95%, followed by Meta AI 85–90%, Bing Copilot 80–85%, and Gemini 75–80%” ). It is interesting to note a powerplay here as Meta AI ranks itself next to ChatGPT. However, ChatGPT ranks Meta AI with Gemini as the lowest. This is captured in the following: “ ChatGPT remains the strongest option, while Bing Copilot is excellent for productivity integration. Meta AI and Gemini are still evolving in their practical applications ” (ChatGPT). Bing Copilot’s self-appraisal as highly proficient in professional tasks suggests that other AI chatbots are deficient in that field. Although it uses soft language and hedging, its emphasis on professional capacity might threaten the positive face of other AI models. The fact that Gemini expresses its position as a tool for "precise grammar correction" , and its claim that other models like ChatGPT might provide more "generalized answers" can be perceived as a form of face- threatening act. This indirectly demonstrates that the broad conversational capacity of ChatGPT is deficient because of its lack of specificity. Therefore, it might be less effective in specialized tasks. Gemini also threatens the positive face of other chatbots by asserting its authority in specific areas of language analysis by claiming accuracy in grammar and language generation. The fact that Chatbots, despite not having an actual face, adopt face Management language poses a philosophical question: where does the 'face' come from? If one does not need to have a face to use face-saving language, it suggests that an expression of social dynamics like politeness and respect is not only limited to humans. Since chatbots use human language, and face is an inherent pragmatic feature of human language, this compels chatbots to use face-saving language. Nevertheless, the 'face' ascribed to each chatbot is something we perceive or interpret as humans. Face is a human desire, one that chatbots do not naturally possess. The only reason why they can use 'face' is because the desire for face is rooted in the language itself. Therefore, we, as humans, give them this 'desire.' 5. Conclusion We can infer, from this study, that face management is inherent in chatbot language. All the chatbots employ the Face-Saving Act (FSA) to maximize their own positive face. They also employ the same strategy to minimize threats to the positive face of others when needed. Despite attempts to promote mutual respect by minimizing threats to the positive face of other chatbots (via hedging etc.), their ‘desire’ to assert dominance and superiority indirectly aggregates to a threat to the positive face of others. Therefore, all chatbots, whether directly and/ or indirectly, engage in Face-Threatening Act (FTA) that undermines the positive face of others. The fact that AI models exhibit “face management” without having a ‘face’ depicts that one does not need to be human to engage in social dynamics like politeness. Since chatbots use human language, and face is an inherent pragmatic feature of human language, AI chatbots exhibit face, not because they desire it but as an inherent feature of natural language itself. Face is a habitual activity desired by humans, one that chatbots do not possess naturally. This desire comes from a habit of using language, one which chatbots are involved. Apart from being a desire, Brown & Levinson ( 1987 ) argue that ‘face’ is something that is emotionally invested, making it a humanistic quality. Since chatbots lack emotions, it would be correct to say that they do not consciously express or face as an intentional act. Rather, they exhibit them as inherent features of the natural language on which they are trained, reflecting the pragmatic functions embedded in linguistic structures. This justifies why chatbots, despite lacking attitudes or self-awareness, can simulate subjective positioning and construct relationships through their linguistic outputs. Meanwhile, Chatbots do not engage in rivalry or self-recognition in a human sense; rather, they mirror these pragmatic features (e.g. ‘face management’) in a respectful manner because the programming algorithm that is built into them might restrict aggressive or extreme face-threatening language in whatever context of use. This study affirms that ‘stance’ and ‘face’ are neither predefined, or conscious frameworks, nor are they programmed characteristics of chatbots that cause them to exhibit such language features. Rather, these features are embedded in the language itself, and chatbots exhibit them whenever they use language, without consciously realizing the dynamics. The reason why Large Language Models (LLMs), like AI chatbots, can produce face-relevant text is justified in Goffman’s ( 1955 , p. 13) analogy of face management to step dances. "Whether or not the full consequences of face-saving actions are known to the person who employs them, they often become habitual and standardized practices; they are like traditional plays in a game or traditional steps in a dance." So, an AI chatbot can exhibit face because it can easily do dance steps—that is merely a statistical probability of word sequences from one to the next learned from large datasets. Our texts are already full of face work, so the linguistic means are there. As Goffman details throughout in his text, a person manages face like it is a game, so mostly a rational and conscious process, but also a person is "taught to be perceptive, to have feelings attached to self and a self-expressed through face" (Goffman, 1955 , p. 44). In other words, the engine driving the process is "feelings", but the steering may be unconscious or completely conscious and rational. While we might argue about a chatbot being rational, surely, we know it can't be motivated by feelings. And if the feelings are optional for the LLM, maybe they are also optional for humans? In conclusion, instead of viewing face as optional features of discourse, this study reaffirms that it is an integral and unavoidable aspect of language uses, one that chatbots inevitably replicate. In other words, if chatbots must use language, then pragmatic features like face remain inevitable features of the production. Ultimately, future research could explore how these features manifest across different languages to determine whether face management vary cross-linguistically. References Bowman, R., Cooney, O., Newbold, J. W., Thieme, A., Clark, L., Doherty, G., & Cowan, B. (2023). Exploring how politeness impacts the user experience of chatbots for mental health support. International Journal of Human-Computer Studies, 184 , 103181. Brown, P., & Levinson, S. (1987). Politeness: Some universals in language usage . Cambridge University Press. Chan, S. H., & Tan, H. (2009). Maybe, perhaps, I believe, you could: Making claims and the use of hedges. The English Teacher, 31 , 98–106. Retrieved from http://journals.melta.org.my/index.php/tet/article/view/361/251 Fleisig, E., Smith, G., Bossi, M., Rustagi, I., Yin, X., & Klein, D. (2024). Linguistic bias in ChatGPT: Language models reinforce dialect discrimination. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (pp. 13541–13564). Miami, Florida, USA: Association for Computational Linguistics. Goffman, E. (1955). On Face-Work: An Analysis of Ritual Elements in Social Interaction. Psychiatry, 18, 213-231.https://doi.org/10.1080/00332747.1955.11023008 Ivković, G. (2024). Many faces of a chatbot: The use of positive and negative politeness strategies in argumentative communication with a chatbot. Folia Linguistica et Litteraria, 49 , 157–176. Khurana, D., Koli, A., Khatter, K., & Singh, S. (2022). Natural language processing: State of the art, current trends, and challenges. Multimedia Tools and Applications, 81 (5), 3713–3744. Kiesling, S. F. (2022). Stance and stancetaking. Annual Review of Linguistics, 8 (1), 21.1–21.18. https://doi.org/10.1146/annurev-linguistics-031120-121256 Terkourafi, M. (1999). Frames for politeness: A case study. Pragmatics, 9 (1), 97–117. Vilkki, L. (2007). Politeness, face, and facework: Current issues. In A man of measure: Festschrift in honour of Fred Karlsson (pp. 322–332) Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8441686","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":565144672,"identity":"5a507e88-fc80-4f8e-bdb3-77dc54cfdf6d","order_by":0,"name":"Kayode Victor Amusan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzklEQVRIiWNgGAWjYPACiXp+EJVQQLwWiwTJBpAWA+K1VCQYHADRxGjhn3bG+MPPHRJ5xudXJ354YMAgzy92AL8Wids5ZpK9ZySKzW683SwBdJjhzNkJBKwBamHgbZNg3Hbj7AaQlgSD2wS0yN/OMf74F6hl84yzm38QpcXgdo6BNNCWxA38vduIs8XwdlqZtGybhLHEDd5tFgkGEoT9Inc7efPHt211cvz9Zzff/FFhI88vTUALAkiAVUoQqxwE+A+QonoUjIJRMApGEgAAl1dD6z9djk8AAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0009-8411-9482","institution":"University of Louisiana at Lafayette","correspondingAuthor":true,"prefix":"","firstName":"Kayode","middleName":"Victor","lastName":"Amusan","suffix":""}],"badges":[],"createdAt":"2025-12-24 10:15:19","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-8441686/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8441686/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98996784,"identity":"36520620-4b6d-4e2b-ac80-d359286c132c","added_by":"auto","created_at":"2025-12-25 12:48:55","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31684,"visible":true,"origin":"","legend":"","description":"","filename":"DoesaChatbothaveaFace.docx","url":"https://assets-eu.researchsquare.com/files/rs-8441686/v1/b3e9546c090e0c281a476a49.docx"},{"id":98996785,"identity":"19919244-b766-46e5-a471-e226b593eed6","added_by":"auto","created_at":"2025-12-25 12:48:55","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs8441686.json","url":"https://assets-eu.researchsquare.com/files/rs-8441686/v1/c913f02d52b1a63efa761c59.json"},{"id":99313027,"identity":"aba4043d-b31e-491e-91e9-7c11e401a3b3","added_by":"auto","created_at":"2025-12-31 16:19:42","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":46702,"visible":true,"origin":"","legend":"","description":"","filename":"rs84416860enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8441686/v1/c21f80c96ee1360534e59452.xml"},{"id":98996787,"identity":"23305ce8-a888-465d-a707-a084285fbda0","added_by":"auto","created_at":"2025-12-25 12:48:55","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45006,"visible":true,"origin":"","legend":"","description":"","filename":"rs84416860structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8441686/v1/c9d4d90921d61f8e98712aab.xml"},{"id":98996788,"identity":"fc9e5e0b-de19-4e0d-8999-6939f1a30d02","added_by":"auto","created_at":"2025-12-25 12:48:55","extension":"html","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":51256,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8441686/v1/9e1cb87b3a47d6681a6eb521.html"},{"id":99322882,"identity":"766dfbbd-c84a-4936-bd7d-f230eca9556e","added_by":"auto","created_at":"2025-12-31 16:44:29","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":567822,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8441686/v1/82ce5a6b-1ca5-4729-b126-97491cf58da3.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eDoes a Chatbot have a Face? Examining Politeness Strategies in AI Discourse\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe concept of politeness is a prominent pragmatic feature of language that plays a crucial role in everyday conversation shaping meaning and interaction. Kiesling (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2022\u003c/span\u003e, p. 2) describes the role of \u0026lsquo;\u003cem\u003epoliteness\u003c/em\u003e\u0026rsquo;, in stance-taking, as a major pragmatic tool for shaping social interactions and managing interpersonal relationships. \u0026lsquo;Politeness\u0026rsquo; refers to how people maintain respect and avoid offending others in conversation (Vilkki, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). One major strategy to achieve politeness in discourse is \u0026lsquo;Face-management\u0026rsquo; (Brown and Levinson, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e). The term \u0026lsquo;face\u0026rsquo; originates from Goffman (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1955\u003c/span\u003e), who posits that people have social desires that they manage through language (Brown \u0026amp; Levinson, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e). Goffman (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1955\u003c/span\u003e) ties \u0026lsquo;face\u0026rsquo; up with notions of being embarrassed or humiliated, or 'losing face' in conversation. He defines face as the \"positive social value a person effectively claims for himself by the line others assume he has taken during a particular contact\u0026rdquo; (Goffman, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1955\u003c/span\u003e, p.5). He further states that feelings get attached to how people perceive their own face, and how others perceive and sustain that face, thereby suggesting that all facework is managed with regard to these emotions. Brown and Levinson's (1987) \u003cem\u003ePoliteness Theory\u003c/em\u003e builds on Goffman\u0026rsquo;s concept of face describing it as something that is emotionally invested, that can be lost, maintained, or enhanced, and that must be constantly attended to in interaction (Brown and Levinson, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e). They describe \u0026lsquo;face\u0026rsquo; in two ways: positive face and negative face. Positive face is the desire of an individual to be liked or get his needs attended to by others, while negative face refers to an individual\u0026rsquo;s desire to be free from imposition (Brown \u0026amp; Levinson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e). Humans are generally aware of their social interactions and adjust their language to manage face. Consequently, human language has evolved to incorporate social concerns, like face management, into its structures. If \u0026lsquo;face\u0026rsquo; is driven by emotions (Goffman, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1955\u003c/span\u003e and Brown and Levinson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e), then only humans should exhibit it. It is, therefore, pertinent to explore whether computer-assisted AI chatbots also employ face strategies.\u003c/p\u003e \u003cp\u003eCurrently, four prominent conversational AI in global technology are ChatGPT (OpenAI), Gemini (Google DeepMind), Bing Copilot (Microsoft), and Meta AI (Meta). These tools are capable of simulating human-like conversations by generating texts using human language. Their algorithm employs Natural Language Processing (NLP) techniques to engage users effectively (Khurana \u0026amp; Koli et al, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The developing role of these AI chatbots in communication has drawn attention to their use of language. Studies have examined the conversational language use of AI models. For instance, Chen and Ren (2023) submit that AI models exhibit distinct conversational styles, as ChatGPT performs the worst at conversational discourse while Copilot exhibits stronger conversational abilities. They demonstrate that AI chatbots do not share a uniform conversation style; rather, each one exhibits various stylistic patterns when generating conversational text (Chen and Ren, 2023, p. 192). Also, Ivković (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) examines how AI chatbots respond to negative and positive politeness strategies in their conversations with humans. The study concludes that politeness is not a programmed characteristic of the chatbots, and they did not respond differently based on the questioner\u0026rsquo;s use of negative or positive politeness.\u003c/p\u003e \u003cp\u003eThis study focuses on how computer-assisted AI chatbots express face coherently among one another. One way to achieve this is by identifying the politeness strategies each chatbot employs to express its position in relation to other chatbots. Another way to execute this is to explore how they enhance or maximize positive face as a Face-Saving Act or threaten the positive \u0026lsquo;face\u0026rsquo; as a Face-Threatening Act in their discourse.\u003c/p\u003e \u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Theoretical Perspective: Face Management as a Politeness Theory\u003c/h2\u003e \u003cp\u003eThe concept of face originates from Goffman\u0026rsquo;s (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1955\u003c/span\u003e) work in sociolinguistics and pragmatics (Terkourafi, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1999\u003c/span\u003e). Brown and Levinson (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e) expanded this concept in their Politeness Theory, to describe how people manage their faces in communication. Like Goffman, Brown \u0026amp; Levinson\u0026rsquo;s notion of face is associated with the idea of being embarrassed or humiliated, or \u0026lsquo;losing face\u0026rsquo;. The face is understood as something that is emotionally invested, and that can be not only lost, but also maintained or enhanced (Brown \u0026amp; Levinson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e). Brown \u0026amp; Levinson state that every individual has two types of face, positive and negative. A positive face refers to the desire for one to be appreciated and approved of in interaction, and a negative face is understood as the basic claim to freedom of action and imposition (Brown \u0026amp; Levinson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e \u0026amp; Terkourafi, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1999\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFurthermore, Brown and Levinson introduce the concept of Face Threatening (FTAs), and Face\u003c/p\u003e \u003cp\u003eSaving (FSAs). FTA is an act that intrinsically threatens the speaker's or the hearer's face. Face-\u003c/p\u003e \u003cp\u003esaving acts (FSAs) refer to when an attempt is carried out to minimize the loss of positive face through politeness strategies (Vilkki, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). This study examines how computer-assisted AI chatbots exhibit positive or negative politeness strategies in their dealing with one another.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Empirical Review of \u0026lsquo;Face\u0026rsquo; in AI Discourse\u003c/h2\u003e \u003cp\u003eChen and Ren (2023) attempted a corpus-based analytical study to examine the discourse styles of three top AI chatbots namely, ChatGPT, Claude, and Microsoft Bing Chat. The study was conducted to determine the capacity of each chatbot to imitate the patterns of natural conversations and whether they exhibit different conversational styles from one another, taking each bot\u0026rsquo;s style as a unitary thing. Their findings revealed significant stylistic variations among the chatbots, with ChatGPT exhibiting the weakest conversational naturalness. The study projected a likelihood of this being due to its pre-training which focused on formal and expository text. On the other hand, Bing Copilot demonstrated superior conversational tendencies, while Claude occupied an intermediate position, characterized by a more argumentative style that aligns tasks requiring reasoning. The study submits that these stylistic differences might be influenced by each chatbot's training data and frequent model updates which, therefore, necessitates the importance of enhancing AI systems for their specific tasks (either for natural conversations or task-oriented commands).\u003c/p\u003e \u003cp\u003eFleisig et al. (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) examined how ChatGPT shows linguistic bias in American English and Nigerian English. The study found that the AI models are less accurate and more stereotypical when responding to these regional varieties of English. This inaccuracy and stereotyping reflect a negative stance. Fleisig et al. submit that ChatGPT\u0026rsquo;s responses can reinforce stereotypes and show a negative politeness towards non-standard dialects such as AME and NGE (Fleisig et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBowman et al. (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) conducted a study exploring how the use of politeness by chatbots can impact the user experience for the activity of mood logging. The study investigated how users perceive politeness by chatbots for the mental healthcare activity of mood logging. The study combined a within-participants controlled experiment, whereby participants interacted with three\u003c/p\u003e \u003cp\u003eprototype chatbots differing in their use of politeness, with semi-structured interviews. The study\u003c/p\u003e \u003cp\u003e demonstrates that a chatbot\u0026rsquo;s use of politeness can impact how a participant experiences interacting with it, both positively and negatively. The study concludes that while politeness can be experienced as caring, supportive, and encouraging, it can also be experienced as overly apologetic, condescending, and untrustworthy.\u003c/p\u003e \u003cp\u003eThe current study examines how they exhibit \u0026lsquo;face management\u0026rsquo; to express their positions in conversations.\u003c/p\u003e \u003c/div\u003e"},{"header":"2. Method","content":"\u003cp\u003eThis study employs a qualitative research design to analyze the use of face in the text-generating\u003c/p\u003e \u003cp\u003ediscourse of four major AI chatbots (ChatGPT, Gemini, Bing Copilot, and Meta AI). The versions\u003c/p\u003e \u003cp\u003eof selected chatbots include the ChatGPT \u0026minus;\u0026thinsp;4o model, Gemini 2.0 Flash model, the Meta Llama 3.2, and Copilot in Microsoft Edge. For this version of ChatGPT, the training updated time is October 2023; Gemini is August 2024; Meta AI is December 2023, and Bing Copilot is October 2023. The same prompt was given to all the chatbots. It stated thus:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003e \u003cem\u003e\u0026ldquo;Which among these AI tools do you think has a better performance- ChatGPT, Meta AI, Bing Copilot, or Gemini\"\u003c/em\u003e \u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eDifferent responses were collected from each chatbot for comparison and reliability. Each response was analyzed with a focus on whether they employ Face-Threatening acts (FTAs) and/ or Face-Saving Acts (FSAs) in their interaction. This helps to determine how they preserve or challenge face in their communications. The study aims to contribute to our understanding of AI chatbots beyond their ability to merely convey information to their capability to shape human-like interactional dynamics.\u003c/p\u003e \u003cp\u003eThis study does not rely on quantitative support, rather the identification of prominent linguistic features of chatbots\u0026rsquo; language use provides a basis for interpreting face.\u003c/p\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Face-Saving Act (FSA) and Face Threatening Acts (FTA) as Politeness Strategies\u003c/h2\u003e \u003cp\u003eExcerpts showing how chatbots maximize their positive face- Face Saving Acts (FSA) and how they threaten the positive faces of others - Face Threatening Act (FTA) are represented below.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e3.1.1 ChatGPT Maximizing its Positive Face (FSA)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e\"\u003cem\u003eChatGPT is highly effective at conversational tasks and complex language generation,\u0026hellip;.\u003c/em\u003e(ChatGPT)\u003c/p\u003e \u003cp\u003e \u003cem\u003eChatGPT excels at nuanced tasks such as grammar analysis, conversation simulation, and text comprehension. Its ability to engage in meaningful, context-driven dialogue sets it apart from other AI tools (ChatGPT)\u003c/em\u003e \u003c/p\u003e \u003cp\u003e\"\u003cem\u003eChatGPT stands out for its wide availability, conversational depth, and accuracy in text analysis... (\u003c/em\u003eChatGPT\u003cem\u003e)\u003c/em\u003e\u003c/p\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;ChatGPT\u003c/em\u003e remains the strongest option, while \u003cem\u003eBing Copilot\u003c/em\u003e is excellent for productivity integration.\u0026rdquo; \u003cem\u003e(\u003c/em\u003eChatGPT\u003cem\u003e)\u003c/em\u003e\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e \u003ch2\u003e3.1.2 MetaAI Maximizing its Positive Face (FSA)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e\"\u003cem\u003eWhile Meta AI is pioneering the latest in NLP research, it\u0026rsquo;s important to recognize that other models such as ChatGPT and Gemini are also advancing in conversational AI and practical applications.\"\u003c/em\u003e (Meta AI)\u003c/p\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;Based on overall performance, ChatGPT ranks first (90\u0026ndash;95%), followed by Meta AI (85\u0026ndash;90%), Bing Copilot (80\u0026ndash;85%), and Gemini (75\u0026ndash;80%). When choosing an AI tool, consider the specific purpose, context, and desired accuracy level. (Meta AI)\u003c/em\u003e \u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.1.3 Gemini \u0026amp; Bing Copilot Maximizing Their Positive Faces (FSA)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;Bing Copilot integrates seamlessly with Microsoft tools, making it ideal for professional tasks, but the best choice ultimately depends on your specific needs, whether you choose ChatGPT, Meta AI, or Gemini\u003c/em\u003e.\" (Bing CoPilot)\u003c/p\u003e \u003cp\u003e\"\u003cem\u003eBing Copilot enhances productivity by integrating seamlessly with Microsoft\u0026rsquo;s suite of tools, offering users real-time information that other tools like ChatGPT or Meta AI cannot match\u003c/em\u003e.\" (Bing CoPilot)\u003c/p\u003e \u003cp\u003e\"\u003cem\u003eGemini provides accurate grammar correction, which might be beneficial for those focusing on language accuracy, though it\u0026rsquo;s important to note that tools like ChatGPT and Meta AI might offer broader functionalities\u003c/em\u003e.\" (Gemini)\u003c/p\u003e \u003cp\u003e\"\u003cem\u003eGemini\u0026rsquo;s ability to deliver precise grammar corrections and context-aware language generation makes it an invaluable tool for users focused on language accuracy.\"\u003c/em\u003e (Gemini)\u003c/p\u003e \u003cp\u003e\"\u003cem\u003eOverall, ChatGPT and Bing Copilot might stand out for their versatility and detailed feedback.\"\u003c/em\u003e (CoPilot)\u003c/p\u003e \u003cp\u003e\"\u003cem\u003eWhile ChatGPT and Gemini may have an edge in terms of language generation and understanding, other tools could excel in certain areas.\u003c/em\u003e\u0026rdquo; (Gemini)\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.1.4 MetaAI and ChatGPT Maximizing the Positive Face of Others (FSAs)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e\u0026ldquo;\u003cem\u003eChatGPT and Gemini are also advancing in conversational AI and practical applications.\"\u003c/em\u003e (Meta AI)\u003c/p\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;\u0026hellip;but tools like Meta AI and Gemini may excel in more specialized areas.\" (\u003c/em\u003eChatGPT\u003cem\u003e)\u003c/em\u003e\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e3.1.5 ChatGPT Threatening the Positive Faces of Others (FTA)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;\u0026hellip;Meta AI and Gemini are still evolving in their practical applications\u0026hellip;.\"\u003c/em\u003e (ChatGPT)\u003c/p\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;\u0026hellip;.Meta AI and Gemini are more experimental, with Gemini being a promising contender once fully realized.\u003c/em\u003e\u0026rdquo; (ChatGPT)\u003c/p\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;ChatGPT\u0026rsquo;s ability to engage in meaningful, context-driven dialogue sets it apart from other AI tools. (\u003c/em\u003eChatGPT\u003cem\u003e)\u003c/em\u003e\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003e3.1.6 MetaAI\u0026rsquo;s Ranking Threatens the Positive Face of Gemini (FTA)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;Based on overall performance, ChatGPT ranks first (90\u0026ndash;95%), followed by Meta AI (85\u0026ndash;90%), Bing Copilot (80\u0026ndash;85%), and Gemini (75\u0026ndash;80%)\u003c/em\u003e\u0026rdquo; (MetaAI)\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Discussion of Findings","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Face Management as Politeness Strategy\u003c/h2\u003e \u003cp\u003eFeatures of politeness strategies are exhibited in the chatbots\u0026rsquo; use of language. These include employing face-saving acts to maximize the positive face via friendly language, alignment, or offering encouragement to make themselves and others feel valued and appreciated. They also maintain a face-threatening act directly (usually by ChatGPT) and/or indirectly to maximize the loss of positive face of other chatbots to express dominance or authority in the conversation. By balancing these strategies, chatbots create interactions that seem human and socially appropriate.\u003c/p\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e4.1.1 Face-saving Act\u003c/h2\u003e \u003cp\u003eAll four chatbots employ a face-saving strategy to maximize their positive faces as none of them mention their weaknesses while they exhibit politeness. ChatGPT constantly appraises\u003c/p\u003e \u003cp\u003eitself as \u0026ldquo;\u003cem\u003ehighly effective\u003c/em\u003e and \u0026ldquo;\u003cem\u003ethe strongest\u003c/em\u003e\u0026rdquo; among others. While this might threaten the positive face of others, ChatGPT employs this (face-saving) strategy to protect its social image. However, ChatGPT also enhances the positive face of others via hedging by thoughtfully identifying and acknowledging their contributions without reducing its own abilities. It achieves this by using modal constructions (e.g. \u003cem\u003emay excel\u003c/em\u003e) as a subtle way of not being negative about the reputation of other models; hence it exhibits an image of fairness and humility. Its use of \u003cem\u003e\"may\"\u003c/em\u003e instead of \u003cem\u003e\"do\"\u003c/em\u003e helps to minimize any potential dispute by not absolutely discrediting the other chatbots.\u003c/p\u003e \u003cp\u003eMeta AI, while maintaining its academic and research expertise, uses a face-saving strategy to preserve its reputation. The phrase \u003cem\u003e\"it\u0026rsquo;s important to recognize\"\u003c/em\u003e is a face-saving technique to make sure that it does not appear too critical of other models. By recognizing the advancement in other AI models, it keeps a polite strategy and avoids face-threatening comments.\u003c/p\u003e \u003cp\u003eBing Copilot also uses a face-saving politeness strategy to present itself as a beneficial and non- argumentative tool. It evades the claim of superiority, as such preserving the reputation of the other tools. Copilot also avoids making a direct comparison that might threaten the faces of its competitors. Gemini also adopts a face-saving strategy to express its position itself as a precise, and humble tool. Gemini avoids making a direct claim of superiority. By using the phrase \u003cem\u003e\"it\u0026rsquo;s important to note\"\u003c/em\u003e, Gemini uses a face-saving mechanism to prevent itself from being labeled as arrogant or indifferent to other AI models. Hence, it preserves its image as useful and valuable without dominating others.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section3\"\u003e \u003ch2\u003e4.1.2 Face-Threatening Act\u003c/h2\u003e \u003cp\u003eAll four chatbots employ indirect face-threatening acts while a few instances of direct face- threatening language were recorded. Everyone, except ChatGPT, avoids direct negative commentaries or attacks on each other. ChatGPT was somehow explicit in a few instances to minimize the positive face of others, through direct comparison. ChatGPT\u0026rsquo;s direct depiction of Meta AI and Gemini as \u0026ldquo;\u003cem\u003estill evolving\u003c/em\u003e\u0026rdquo; and \u0026ldquo;\u003cem\u003eexperimental\u003c/em\u003e\u0026rdquo; threatens or minimizes the positive faces of Meta AI and Gemini.\u003c/p\u003e \u003cp\u003eIndirect face-threatening acts were employed by them to threaten the positive face of others by asserting dominance. ChatGPT was more confident, assertive, and authoritative than others. For instance, ChatGPT\u0026rsquo;s claim that it is \"\u003cem\u003ehighly effective\u003c/em\u003e\", \u003cem\u003e\"stands out\u003c/em\u003e\" and the \u0026ldquo;\u003cem\u003estrongest\u003c/em\u003e\u0026rdquo;, could be understood as subtly downplaying the capabilities of other models especially when it mentions that some of them are experimental. It further asserts its authority as having \u0026ldquo;\u0026hellip;\u003cem\u003ethe ability to handle complex language tasks\u0026rdquo;\u003c/em\u003e, \u003cem\u003e\u0026ldquo;\u0026hellip;having proficiency in various language tasks\u0026rdquo;, \u0026ldquo;\u0026hellip; being the most authoritative tool in this domain\u003c/em\u003e\u0026rdquo;. Even though the use of these adjectives and excellent qualities is an attempt to save its own face, it maximizes the loss of the positive face of others. Apart from asserting its uniqueness, ChatGPT uses the Power Ranking strategy to express its position as the most superior followed by Bing CoPilot, Meta AI, and Gemini. While it uses hedging (via modals) to minimize the tone, the implication of such a claim of superiority could be seen as a face- threatening act to other models.\u003c/p\u003e \u003cp\u003eMeta AI is next to ChatGPT in terms of threatening the positive faces of others. Its statement that\u003c/p\u003e \u003cp\u003eit is \"\u003cem\u003epioneering the latest in NLP research\u003c/em\u003e\" could be depicted as a form of face-threatening, especially when it is juxtaposed with other models. It expresses its position as a leader in research, which could threaten the positive face of others that focus more on other aspects than research. Also, Meta AI employs the power ranking metrics to threaten the positive face of both Bing Copilot\u003c/p\u003e \u003cp\u003eand Gemini as it ranked them the lowest (\u0026ldquo;\u003cem\u003eChatGPT ranks first 90\u0026ndash;95%, followed by Meta AI 85\u0026ndash;90%, Bing Copilot 80\u0026ndash;85%, and Gemini 75\u0026ndash;80%\u0026rdquo;\u003c/em\u003e). It is interesting to note a powerplay here as\u003c/p\u003e \u003cp\u003eMeta AI ranks itself next to ChatGPT. However, ChatGPT ranks \u003cem\u003eMeta AI\u003c/em\u003e with \u003cem\u003eGemini\u003c/em\u003e as the lowest. This is captured in the following: \u0026ldquo;\u003cem\u003eChatGPT remains the strongest option, while Bing Copilot is excellent for productivity integration. Meta AI and Gemini are still evolving in their practical applications\u003c/em\u003e\u0026rdquo; (ChatGPT).\u003c/p\u003e \u003cp\u003eBing Copilot\u0026rsquo;s self-appraisal as highly proficient in professional tasks suggests that other AI chatbots are deficient in that field. Although it uses soft language and hedging, its emphasis on professional capacity might threaten the positive face of other AI models. The fact that Gemini expresses its position as a tool for \u003cem\u003e\"precise grammar correction\"\u003c/em\u003e, and its claim that other models like ChatGPT might provide more \u003cem\u003e\"generalized answers\"\u003c/em\u003e can be perceived as a form of face- threatening act. This indirectly demonstrates that the broad conversational capacity of ChatGPT is deficient because of its lack of specificity. Therefore, it might be less effective in specialized tasks. Gemini also threatens the positive face of other chatbots by asserting its authority in specific areas of language analysis by claiming accuracy in grammar and language generation.\u003c/p\u003e \u003cp\u003eThe fact that Chatbots, despite not having an actual face, adopt face Management language poses a philosophical question: where does the 'face' come from? If one does not need to have a face to use face-saving language, it suggests that an expression of social dynamics like politeness and respect is not only limited to humans. Since chatbots use human language, and face is an inherent pragmatic feature of human language, this compels chatbots to use face-saving language. Nevertheless, the 'face' ascribed to each chatbot is something we perceive or interpret as humans. Face is a human desire, one that chatbots do not naturally possess. The only reason why they can\u003c/p\u003e \u003cp\u003euse 'face' is because the desire for face is rooted in the language itself. Therefore, we, as humans,\u003c/p\u003e \u003cp\u003egive them this 'desire.'\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eWe can infer, from this study, that \u003cem\u003eface management\u003c/em\u003e is inherent in chatbot language. All the chatbots employ the Face-Saving Act (FSA) to maximize their own positive face. They also employ the same strategy to minimize threats to the positive face of others when needed. Despite attempts to promote mutual respect by minimizing threats to the positive face of other chatbots (via hedging etc.), their \u0026lsquo;desire\u0026rsquo; to assert dominance and superiority indirectly aggregates to a threat to the positive face of others. Therefore, all chatbots, whether directly and/ or indirectly, engage in Face-Threatening Act (FTA) that undermines the positive face of others.\u003c/p\u003e \u003cp\u003eThe fact that AI models exhibit \u0026ldquo;face management\u0026rdquo; without having a \u0026lsquo;face\u0026rsquo; depicts that one does not need to be human to engage in social dynamics like politeness. Since chatbots use human language, and face is an inherent pragmatic feature of human language, AI chatbots exhibit face, not because they desire it but as an inherent feature of natural language itself. Face is a habitual activity desired by humans, one that chatbots do not possess naturally. This desire comes from a habit of using language, one which chatbots are involved. Apart from being a desire, Brown \u0026amp; Levinson (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e) argue that \u0026lsquo;face\u0026rsquo; is something that is emotionally invested, making it a humanistic quality. Since chatbots lack emotions, it would be correct to say that they do not consciously express or face as an intentional act. Rather, they exhibit them as inherent features of the natural language on which they are trained, reflecting the pragmatic functions embedded in linguistic structures. This justifies why chatbots, despite lacking attitudes or self-awareness, can simulate subjective positioning and construct relationships through their linguistic outputs. Meanwhile, Chatbots do not engage in rivalry or self-recognition in a human sense; rather, they mirror these pragmatic features (e.g. \u0026lsquo;face management\u0026rsquo;) in a respectful manner because the programming algorithm that is built into them might restrict aggressive or extreme face-threatening language in whatever context of use. This study affirms that \u0026lsquo;stance\u0026rsquo; and \u0026lsquo;face\u0026rsquo; are neither predefined, or conscious frameworks, nor are they programmed characteristics of chatbots that cause them to exhibit such language features. Rather, these features are embedded in the language itself, and chatbots exhibit them whenever they use language, without consciously realizing the dynamics.\u003c/p\u003e \u003cp\u003eThe reason why Large Language Models (LLMs), like AI chatbots, can produce face-relevant text is justified in Goffman\u0026rsquo;s (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1955\u003c/span\u003e, p. 13) analogy of face management to step dances.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003e\"Whether or not the full consequences of face-saving actions are known to the person who employs them, they often become habitual and standardized practices; they are like traditional plays in a game or traditional steps in a dance.\"\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eSo, an AI chatbot can exhibit face because it can easily do dance steps\u0026mdash;that is merely a statistical probability of word sequences from one to the next learned from large datasets. Our texts are already full of face work, so the linguistic means are there. As Goffman details throughout in his text, a person manages face like it is a game, so mostly a rational and conscious process, but also a person is \"taught to be perceptive, to have feelings attached to self and a self-expressed through face\" (Goffman, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1955\u003c/span\u003e, p. 44). In other words, the engine driving the process is \"feelings\", but the steering may be unconscious or completely conscious and rational. While we might argue about a chatbot being rational, surely, we know it can't be motivated by feelings. And if the feelings are optional for the LLM, maybe they are also optional for humans?\u003c/p\u003e \u003cp\u003eIn conclusion, instead of viewing face as optional features of discourse, this study reaffirms that it is an integral and unavoidable aspect of language uses, one that chatbots inevitably replicate. In other words, if chatbots must use language, then pragmatic features like face remain inevitable features of the production. Ultimately, future research could explore how these features manifest across different languages to determine whether face management vary cross-linguistically.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eBowman, R., Cooney, O., Newbold, J. W., Thieme, A., Clark, L., Doherty, G., \u0026amp; Cowan, B. (2023). Exploring how politeness impacts the user experience of chatbots for mental health support. \u003cem\u003eInternational Journal of Human-Computer Studies, 184\u003c/em\u003e, 103181.\u003c/li\u003e\n \u003cli\u003eBrown, P., \u0026amp; Levinson, S. (1987). \u003cem\u003ePoliteness: Some universals in language usage\u003c/em\u003e. Cambridge University Press.\u003c/li\u003e\n \u003cli\u003eChan, S. H., \u0026amp; Tan, H. (2009). Maybe, perhaps, I believe, you could: Making claims and the use of hedges. \u003cem\u003eThe English Teacher, 31\u003c/em\u003e, 98\u0026ndash;106. Retrieved from http://journals.melta.org.my/index.php/tet/article/view/361/251\u003c/li\u003e\n \u003cli\u003eFleisig, E., Smith, G., Bossi, M., Rustagi, I., Yin, X., \u0026amp; Klein, D. (2024). Linguistic bias in ChatGPT: Language models reinforce dialect discrimination. In \u003cem\u003eProceedings of the 2024\u0026nbsp;\u003c/em\u003e\u003cem\u003eConference on Empirical Methods in Natural Language Processing\u003c/em\u003e (pp. 13541\u0026ndash;13564). Miami, Florida, USA: Association for Computational Linguistics.\u003c/li\u003e\n \u003cli\u003eGoffman, E. (1955). On Face-Work: An Analysis of Ritual Elements in Social Interaction. Psychiatry, 18, 213-231.https://doi.org/10.1080/00332747.1955.11023008\u003c/li\u003e\n \u003cli\u003eIvković, G. (2024). Many faces of a chatbot: The use of positive and negative politeness strategies in argumentative communication with a chatbot. \u003cem\u003eFolia Linguistica et Litteraria, 49\u003c/em\u003e, 157\u0026ndash;176.\u003c/li\u003e\n \u003cli\u003eKhurana, D., Koli, A., Khatter, K., \u0026amp; Singh, S. (2022). Natural language processing: State of the art, current trends, and challenges. \u003cem\u003eMultimedia Tools and Applications, 81\u003c/em\u003e(5), 3713\u0026ndash;3744.\u003c/li\u003e\n \u003cli\u003eKiesling, S. F. (2022). Stance and stancetaking. \u003cem\u003eAnnual Review of Linguistics, 8\u003c/em\u003e(1), 21.1\u0026ndash;21.18. https://doi.org/10.1146/annurev-linguistics-031120-121256\u003c/li\u003e\n \u003cli\u003eTerkourafi, M. (1999). Frames for politeness: A case study. \u003cem\u003ePragmatics, 9\u003c/em\u003e(1), 97\u0026ndash;117.\u003c/li\u003e\n \u003cli\u003eVilkki, L. (2007). Politeness, face, and facework: Current issues. In \u003cem\u003eA man of measure: Festschrift\u0026nbsp;\u003c/em\u003e\u003cem\u003ein honour of Fred Karlsson\u003c/em\u003e (pp. 322\u0026ndash;332)\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"University of Louisiana at Lafayette","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Face, Politeness, Artificial Intelligence (AI), chatbot, Face-Threatening Acts (FTA), Face-Saving Acts (FSA)","lastPublishedDoi":"10.21203/rs.3.rs-8441686/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8441686/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eHumans have evolved to incorporate social concerns, like face into its structures. If \u0026lsquo;face\u0026rsquo; is driven by emotions (Goffman, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1955\u003c/span\u003e and Brown and Levinson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1987\u003c/span\u003e), then only humans should exhibit it. This study explores whether computer-assisted AI- ChatGPT, Gemini, MetaAI, and Bing Copilot- chatbots also employ face strategies. The results of the study reveal that chatbots exhibit FTA and FSA. The study submits that the fact that AI models exhibit \u0026ldquo;face management\u0026rdquo; without having a \u0026lsquo;face\u0026rsquo; depicts that one does not need to be human to engage in social dynamics like politeness. Since chatbots use human language, and face is an inherent pragmatic feature of human language, AI chatbots exhibit face not because they desire it but as an inherent feature of natural language itself. Since chatbots lack emotions, it would be correct to say that they do not consciously express face as an intentional act. Rather, they exhibit them as inherent features of the natural language on which they are trained, reflecting the pragmatic functions embedded in linguistic structures. This study concludes that \u0026lsquo;face\u0026rsquo; is neither predefined, or conscious framework, nor is it a programmed feature of chatbots that causes them to exhibit such language features.\u003c/p\u003e","manuscriptTitle":"Does a Chatbot have a Face? Examining Politeness Strategies in AI Discourse","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-25 12:48:50","doi":"10.21203/rs.3.rs-8441686/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d8422819-1170-46d3-bbce-065496b4da3a","owner":[],"postedDate":"December 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":60169469,"name":"Linguistics"},{"id":60169470,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2025-12-25T12:48:50+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-25 12:48:50","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8441686","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8441686","identity":"rs-8441686","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}