Background and introduction

Unlike most animals, humans engage in many practices that are culturally transmitted (Richerson and Boyd, 2005). Importantly, cultural learning is not limited to easily observable behavioral traits, but it also extends to the more subtle aspects of how we think, reason, and communicate (Heyes, 2018). In this paper, we aim to specifically examine the cultural evolution of rhetoricFootnote 1, i.e., the ways in which people frame their arguments to persuade others. While previous research has highlighted the genetically evolved component of human argumentation, particularly in the context of reasoning (Mercier and Sperber, 2011), the abundant cultural variation in argumentation styles suggests a significant culturally transmitted element in how we persuade others (Lloyd, 2013; Roccli, 2006). The significance of cultural input in argumentation is particularly clear in the Western intellectual tradition: effective argumentation techniques have to be acquired via substantial learning since Greco-Roman times (Clark, 1957), and rhetoric classically constituted a central part of the traditional educational curriculum, along with logic and grammar (Harbsmeier, 1999).

For cognitively-minded researchers, the study of rhetoric is important in two ways. Firstly, as a key aspect of human intellectual life, rhetoric provides a unique lens that enables us not only to examine overt argumentation but also to discern concealed patterns in thinking and reasoning in specific social, cultural, and historical contexts (Conley, 1994). Secondly, an investigation into the variety of rhetorical strategies individuals employ could provide valuable insights into the cognitive mechanisms of knowledge acquisition. The second point is less evident and requires some further clarification. While rhetoric has traditionally been studied as a strategy of effective persuasion, there is an analogy to be made between changing other people’s opinions and updating one’s own beliefs. This is because the methods with which we change another person’s opinion—highlighting existing information or presenting new evidence of various sorts—closely resemble how we learn from our natural and social environment. This idea is not entirely new; Scott (1967) famously suggested that “rhetoric is epistemic”, a proposal which spawned a series of scholarly work (Brummett, 1976; Cherwitz and Kneupper, 1984; Railsback, 1983). However, Scott’s analysis of rhetoric as epistemic, along with the subsequent work, remains largely philosophical and concerns the nature of truth, whereas our focus here is on the cognitive aspect. There exist myriad ways one can present their arguments to persuade others, such as enumerating examples, making analogies, or appealing to authority. These methods notably mirror aspects of human learning: we learn by induction (Tenenbaum et al. 2006), by analogical reasoning (Beach et al. 2021), and by deferring to prestigious individuals (Henrich and Gil-White, 2001). Theoretically, an effective rhetorical strategy should bear structural similarities to a widely accepted epistemic method in a given society. Therefore, by studying rhetoric, we may also gain insights into the epistemic practices and norms within a societyFootnote 2.

The term “rhetoric” originates from the Greek word ρήτωρ, meaning “orator” (Condor et al. 2013). While its application extends to all aspects of daily life, the most notable manifestation of rhetoric is on the political stage (Gill and Whedbee, 1997). Classically, being a proficient rhetorician is synonymous with being an effective public speaker on political matters. Indeed, considerable research has delved into the cultural, social, and economic implications of political rhetoric in democratic societies (Chávez, 2018; Eaton, 2010; Endres and Senda-Cook, 2011; Griggs and Howarth, 2004). In this study, we specifically look at the cultural evolution of rhetorical strategies (broadly defined) in political settings using ancient China as a case study. Despite obvious differences between Western rhetoric that emphasizes the public aspect of persuasion and the Chinese rhetorical tradition where persuasion typically occurs in more private settings (Yakobson 2012; Lloyd 1996, p. 74–92), there is substantial reason to consider the propensity and ability to argue as universal (which leaves ample room for cultural variation in successful persuasion) (Hacking, 2013; Mercier, 2013) as it likely confers adaptive benefits (Parrish, 2013), and the Chinese undoubtedly engaged in communications aimed at persuasion (Mercier, 2011).

Therefore, by examining the Chinese historical records spanning the past two millennia, we can gain valuable insights into how argumentation has evolved over time and the driving cultural evolutionary forces behind this transformation. We’re particularly interested in the potential for cumulative payoff-biased cultural evolution, i.e., whether rhetorical strategies with more documented persuasion success have increased temporally in relative frequency and/or if strategies linked to higher persuasion failures have decreased in frequency. To this end, we assembled a dataset of political persuasion in traditional China using keyword searches in the fully digitized Chinese official dynastic histories. In our dataset, we grouped all persuasion episodes into different themes (e.g., military, ritual, economy, etc.), classified the rhetorical strategies used in each persuasion episode into various types (argument by analogy, argument by historical example, argument by authoritative texts, etc.), and recorded whether the persuasion was successful when such information is available. To preview the results, we identify a rich variety of rhetorical strategies used by political actors in traditional China and show a decrease in the frequencies of specific rhetorical strategies associated with persuasion failures. At the conclusion of our paper, we highlight the significance and limitations of the present study and propose potential directions for future work.

Dataset description

Data source

The dataset of political rhetoric in traditional China was constructed using keyword searches in the fully digitized official Chinese dynastic histories, the Twenty-Four Histories (二十四史), covering from mythic times (~2600 BCE) to the end of the Ming Dynasty (1644 CE). These dynastic histories are official in the sense that later dynasties would designate court historians to write comprehensive books that are mostly biographies of important people such as kings, emperors, generals, and administrators, as well as important historical events (Zinin and Xu, 2020). The keyword used in the search is “jian yue 谏曰”, which roughly translates as “to remonstrate…that…” (Schaberg 1997). In classical Chinese, jian originally means individuals in an inferior position offering a set of alternatives for their superior to choose from, but was frequently used to refer to inferiors trying to dissuade their superiors from doing something that they would like to do, or to persuade them to do something that they are unwilling to do. In other words, to “jian” is to attempt to change the mind of the superior. In the political context of traditional China (especially imperial China) where the political authority of the superior was often absolute, jian was one of the main ways in which a minister could seek to modify his ruler’s conduct (Smart 2008).

Time assignment of dynastic periods

For persuasion episodes in each dynastic period, we took the mid-point of the period as the “dynastic time” variable, with a total of 24 values. Notice that the first Dynastic History document, Shiji, was broken into two time periods, “Pre-Han” and “Western Han,” based on sinological convention. In presenting descriptive statistics, some shorter historical periods were binned together for better visualization purposes by ensuring the overall time coverage of each period was relatively comparable (e.g., dynastic histories between 420 to 589 CE were collapsed into a single “Southern and Northern Dynasties”). For more information, see Supplemental Information.

Results

Overall descriptive statistics

We extracted 763 persuasion episodes (using the keyword jian yue 諫曰; see Methods section for more details) in total and further categorized them by themes (one persuasion episode may contain multiple themes) as depicted in Fig. 1. These episodes encompass a wide array of topics, with military topics being the most frequent, followed by personal conduct and personnel appointment related issues. Military-themed persuasions typically involve the persuader(s) offering specific, pragmatic advice to their superiors in the context of military decision-making. For instance, during the Three Kingdom period, General Lu Kang of the Wu received the following advice from his advisors during a battle in 272 CE: “Now, with all our military strength, if we launch a full attack on the city of Buchan, we will have already captured the city by the time the enemy’s reinforcements arrive. Why bother surrounding the city and building all these defensive structures, exhausting our soldiers?”Footnote 3 Notably, military persuasions often present straightforward cost-benefit analyses with minimal moral considerations. In contrast, persuasions concerning personal conduct primarily include moral pedagogical statements regarding the superior’s behavior. For example, one such statement reads, “I hope [our] wise emperor will eschew the luxuries of horses and carriages, disregard the alchemists’ vague words, and focus on the art of [being a good] emperor. Then, our nation’s prosperity will soon be achieved.”Footnote 4 These moral persuasions related to the superior’s personal conduct typically adhere to Confucian ethics, often advocating restraint from material and bodily pleasures. Personnel-themed persuasions might pertain to either general policies or specific instances of personnel appointments or removals. More comprehensive descriptions of persuasion themes can be found in the Supplemental Information.

Fig. 1: Themes of persuasion episodes.
figure 1

See SI for a more detailed description of the criteria for categorization.

Full size image

In all the data combined, the relative proportions of persuasion successes and failures are comparable (56.1% vs. 43.9%), with success rates ranging from ~20% to ~75% across both historical periods and themes (see Fig. 2). The presence of similar proportions of successful and failed persuasions is significant. Not only does it suggest that there are unlikely to be strong, systematic biases in the reporting of persuasion outcomes compared to other types of historical documentation Hong, 2022a; Hong et al. 2024), but it also enhances our ability to more effectively examine the predictors of successful persuasion from a statistical analysis perspective.

Fig. 2: Persuasion success rate by historical periods and themes.
figure 2

Left panel shows the success rate from the pre-Han period (Han dynasty and before) to the Ming dynasty. Right panel shows the breakdown of overall persuasion success rate by themes.

Full size image

We observe a temporal increase in persuasion success rate, with the exception of the Ming Dynasty. It is worth noting that from the Tang dynasty (618–907 CE) onward, the government began to appoint a special role of “remonstrance ministers” yanguan 言官 or jianguan 諫官 whose primary responsibility was to offer corrective advice to the emperor, functioning in theory as an institutional check against potential misconduct (Chen, 2001). However, during the late Ming dynasty, the relationship between the emperor and the ministers soured (Andrew and LaFleur, 2014; Chen, 2001). Emperors often resented the incessant advice from ministers, and ministers would frequently pride themselves on offering rather harsh critiques to the emperor, accepting the subsequent punishment as a badge of honor (Cai, 2009; Li, 2013; M. Zhang, 2014). This friction resulted in a large number of failed persuasions, and the peculiar circumstances of the Ming Dynasty warrant further investigation. However, for the purposes of our analysis, we focus on the overall trend indicating an increase in the persuasion success rate over time, which suggests the influence of potential cultural evolutionary forces. As for persuasion success by theme, most themes show comparable success rates that do not deviate significantly from 50%, indicating an absence of obvious theme-related reporting bias.

Cultural evolution of rhetorical strategies

To gain a clearer understanding of how the methods of argumentation have evolved over time, we coded the rhetorical strategies used in persuasion episodes. These strategies fall into two broad groups: substantive and stylistic. We calculated both their total counts (Fig. 3) and their relative frequencies in each historical period (Fig. 4). Substantive strategies involve the presentation of substantial information, such as quoting an ancient authority or source (“I have learned from the I Ching: ‘In safety, do not forget danger; in preservation, do not forget ruin. Thus, one remains secure and the state can be preserved’”Footnote 5), while stylistic strategies cater to the non-rational aspects of argumentation such as flattery (“your Majesty, young and already celebrated for valor, is respected both far and near”Footnote 6) and self-deprecation (“I humbly defy your majesty of Heaven, and I beg for your compassionate forgiveness”Footnote 7). The classification criteria were developed by the authors in consultation with experts of Chinese history at Peking University, and a random sample of 100 persuasion episodes was independently coded by two coders who are proficient in classical Chinese and blind to our hypotheses, resulting in substantial intercoder reliability (see Supplemental Information for details of coding scheme and intercoder reliability analysis).

Fig. 3: Total counts of different rhetorical strategies used.
figure 3

Top panel shows the total counts of stylistic strategies and bottom panel shows the total counts of substantive strategies in our dataset.

Full size image
Fig. 4: The relative frequencies (counts per 1000 characters) of rhetorical strategies in different historical periods.
figure 4

Values are calculated by dividing the number of occurrences of rhetorical strategies in each historical period by the total number of characters in the dynastic records of the period.

Full size image

Though it is challenging to discern definitive trends, we do observe several notable patterns that align with common sense and the existing literature in Chinese literary analysis and communication studies (Garrett, 1983; Smart, 2008). For example, cost/benefit analysis and the provision of useful information greatly outnumber other substantive strategies, as they are the most straightforward ways to change one’s mind. A case in point occurred during the Spring-Autumn period (770 BCE–476 BCE). Minister Ye tried to discourage the King of Chu from calling back General Sheng from abroad: “Sheng is a violent and aggressive person, and he is secretly recruiting desperados; he probably has his own plans.” Here, Ye clearly laid out the (presumed) factual information (Sheng is violent and aggressive and recruiting desperados) and, though not completely explicit, indicated the potential danger of inviting Sheng back (Sheng probably has his own plans, i.e., rebellion).

The use of historical examples is also substantial, and it has been suggested that argument from examples is likely to be a universal type of argument across human societies (Aristotle famously thinks that it is a form of induction in Topics). Indeed, learning from past experiences (examples) is such a fundamental biological imperative that it is hard to imagine it not being an important mode of argumentation. In our analysis, the examples used by those trying to persuade are almost always historical ones, usually from transmitted historical records. We further divided these into two categories: historical examples not specifying outcomes (enumerating the actions of historical figures and indirectly implying what is to be done/avoided) and historical examples with outcomes (enumerating actions and also pointing out the outcome of such actions, therefore directly specifying what is to be done/avoided). Argument from authority in the form of citing authoritative texts also features prominently. A more detailed analysis regarding the frequency of various texts cited can be found in Supplemental Information, and Table 1 offers a list of illustrative examples of all substantive rhetorical strategies to give readers a better sense of how they are employed in specific persuasion episodes.

Table 1 List of illustrative examples of substantive rhetorical strategies used in persuasion episodes in dynastic histories.
Full size table

Among stylistic strategies, constructive criticism, which entails criticism paired with actionable suggestions, was used most frequently in comparison to other strategy types, which makes sense given that the persuaders typically want their superiors to perform specific actions, and thus criticizing their current plans is often not sufficient. It is worth noting that certain stylistic features (e.g., flattery and self-deprecation) are well established linguistic conventions signaling politeness and modesty (Kádár, 2007), and as such, convey little useful information that may aid the superior’s decision-making.

While corroborating many findings from the existing literature, our results also cast doubt on some received wisdom of early Chinese argumentation. Chain reasoning (if A, then B; if B, then C…), for example, described as a common form of traditional Chinese argumentation (Garrett, 1983), is absent from our dataset. Arguments from analogy, another frequently cited characteristic of Chinese rhetorical strategy, were only prominent during the early dynasties (more on this later). It is worth noting, however, that our dataset was compiled solely from dynastic histories and may have genre-specific features. It is possible, for example, that chain reasoning and/or argument from analogy occur more often in philosophical writings or solitary reflections. Nevertheless, our results provide the first quantitative evidence against some existing views in this important literary genre, indicating directions for further investigations.

In addition to descriptive statistics, we are also interested in whether certain rhetorical strategies are more likely to lead to persuasion success. Considering that there are a total of 18 rhetorical strategies and a single persuasion episode may employ multiple strategies, we used the StepAIC method in R—an algorithm that performs stepwise regression iteratively to select the more informative variables based on the Akaike Information Criterion (Zhang, 2016)—to identify important predictors of persuasion success (we also included argument length as measured by the number of characters and dynastic time during which the argument was made as control variables), and then performed a multiple regression analysis with the final set of rhetorical features listed in Table 2. In addition to dynastic time (with a positive coefficient, indicating that the persuasion success rate increased over time), there are four substantive rhetorical strategies significantly associated with persuasion success: auspicious/inauspicious signs, analogy, historical example with the outcome, and conventional wisdom. Among these, conventional wisdom is the only strategy that is positively associated with persuasion success.

Table 2 Multiple regression analysis of rhetorical strategies (selected by StepAIC) that predict persuasion success.
Full size table

Next, we asked the more interesting question of whether the frequencies of rhetorical strategies that significantly correlate with persuasion success changed over time, i.e., whether there is a temporal increase in frequency for more successful rhetorical strategies and/or a decrease in frequency for less successful ones. To tackle this question, we conducted regression analyses to examine the temporal change in frequency of each rhetorical strategy more rigorously, with the coefficient estimates shown in Fig. 5. Given the caveat of multiple testing of running many simple regressions, we also computed adjusted p values after applying the Bonferroni correction (Armstrong, 2014).

Fig. 5: Coefficient plot of Simple regressions for the frequency of individual rhetorical strategies.
figure 5

The independent variable is “time”, which is proxied by the mid-year of each dynastic period (See SI for more details). Solid lines denote 95% confidence intervals of the coefficient estimates and red ones denote estimates of strategies that significantly varies with time after the Bonferroni correction.

Full size image

Most rhetorical strategies show a decrease in frequency over time, as indicated by the sign of the coefficient estimates. Three strategies—authoritative text, auspicious/inauspicious sign, and analogy—exhibit a significant decline after the Bonferroni correction. These findings resonate with prior qualitative research on Chinese history and culture: the use of auspicious/inauspicious signsFootnote 8 in political argumentation and decision-making was prominent during the early dynasties (Xia, Shang, and Zhou dynasties, approximately until 221 BCE), but its popularity declined as people adopted more “rational” methods for making political decisions (Chen, 2001; Zhao, 2021). Similarly, most research on argument by analogy in traditional China does not explicitly identify a temporal trend, but almost all published studies focus on cases before the Han dynasty (202 BCE) (Kirkpatrick, 1995; Lu and Frank, 1993; Reding, 1986; Xie, 2019). On the use of authoritative texts in argumentation, there is also evidence that direct quotations of ancient authorities in literary texts appear more often in early dynasties (Shu et al. 2021).

Granted, the match between our results and previous qualitative studies should be viewed as suggestive due to the lack of other systematic, quantitative studies of Chinese rhetoric that explicitly focus on the temporal dimension. Our results do, however, raise the intriguing possibility that our evolved transmission biases may have played a role in the cultural evolution of rhetorical strategies in China. Notice that two (auspicious/inauspicious sign and analogy) of the three strategies that show a decline in usage frequency are also associated with persuasion failures. These data are thus compatible with the hypothesis that there is a decrease in frequency for rhetorical strategies that are less successful, indicative of payoff-biased transmission where individuals preferentially adopt cultural variants with higher payoffs (Kendal et al. 2009). Such a cultural evolutionary interpretation presupposes that the persuaders are aiming for successful persuasion, which may not always be the case (as in the Ming dynasty, where much persuasion in the imperial court became a form of virtue signaling). Overall, however, it is reasonable to assume that most persuaders intended their advice to be taken. This is not only because it was typically in the persuaders’ direct interest to change their superiors’ minds but also because unsuccessful persuasions often carried a genuine personal risk (Graziani, 2012). In our dataset, failed persuasions often led to death, demotion, or estrangement of the persuader (61.1%), whereas most successful persuasions were associated with reward and promotion (81.5%). Therefore, it is possible that individuals may have observed the outcomes of persuasion with different rhetorical strategies and (possibly subconsciously) adopted those leading to more persuasion successes and avoided those resulting in more failures. Note that, unlike small-scale, illiterate societies, the existence of writing in China means that literate individuals (most actors in the persuasion episodes were literate) often had access to cumulative records of argumentation for hundreds of years and could learn from them accordingly. Indeed, familiarity with written history was a definitive sign of intellectual prestige and “learning from history” was held as an important way of acquiring knowledge in the Chinese tradition (Jiang, 2007). In such a context, learning occurs not only through observing one’s contemporaries but also by reading the past from transmitted texts.

We need to keep in mind that dynastic histories, like most historical documents, are biased regarding what actors and events are included as well as how they are depicted and interpreted (Hong, 2022a; Hong and Zinin, 2023). Typically, these dynastic histories were official records compiled by contemporary or later historians, and we might expect various incentives in the selection of records to include. For example, rainmaking rituals were often selectively recorded to justify the political legitimacy of the rainmaker (e.g., the emperor) and as a result, most rainmaking failures went unrecorded (Hong et al. 2024). An important feature of our dataset is that most of the persuasion episodes imply some sort of moral lesson: the historians who wrote these dynastic histories used these stories to either justify their own moral worldview or to fit into some grand narrative. Additionally, the fact that many arguments recorded in these dynastic histories had non-trivial consequences (e.g., punishment or promotion) suggests that more mundane argumentative conversations were less likely to be recorded. What matters for our purposes, however, is that there is no obvious reason to suspect any systematic bias in the type of rhetorical strategies used in argumentation episodes as recorded in dynastic histories. Indeed, unlike the Greek tradition, where rhetoric was an explicit subject of scholarly study, the Chinese, with a few exceptions (e.g., some early thinkers such as Xunzi were aware of the power and danger of flattery in persuasion, see Galvany, 2012) rarely treated rhetoric as an independent domain of intellectual inquiry (Oliver, 1971). Therefore, it is unlikely that historians of traditional China were consciously biased towards selectively including or excluding persuasion episodes with certain types of rhetorical strategies.

Discussion

Universal vs. cultural-historical specific rhetorical strategies

While there is little doubt that individuals in most human societies employ various forms of rhetorical strategies to argue with one another, contemporary studies of rhetoric have largely focused on the Greco-Roman tradition. Recent years have witnessed a blossom of rhetoric studies in other societies with a literary tradition, such as India (Sen, 2012) and China (Kirkpatrick, 1995), which offer important insights into cultural factors in argumentation, yet none of the studies provide systematic, quantitative analysis of individual rhetorical strategies. We contend that China serves as an ideal candidate for comparative analysis due to its rather different mode of argumentation from the West: there was barely any form of public collective decision-making, such as voting. In fact, argumentation in early China was suggested to be no more than a personal exercise in persuasion that resorts to a wide variety of rhetorical devices (Graziani, 2012). The present study, which focuses on traditional China with an explicit temporal dimension, contributes to our understanding of the universality of rhetorical strategies as well as the culturally and historically specific aspects. For instance, we show that as in the Greco-Roman tradition, argument by example was a very prevalent form of rhetoric throughout Chinese history; in contrast, referencing authoritative texts, though not unique to the ChineseFootnote 9, was likely utilized much more often than in other civilizations. Indeed, the habit of preserving, transmitting, and valuing authoritative texts was so strong in China (Gentz and Meyer, 2015; Kirkpatrick, 1995) that citing authority saw a revival during the Cultural Revolution in the 1960s and 1970s, when it became almost obligatory for people to buttress their arguments with Chairman Mao’s quotations (Cook, 2014). Our dataset’s extensive temporal coverage also enables us to reveal some interesting patterns that are specific to certain historical periods, such as the prominence of argument by analogy and the use of natural signs in early dynasties, all of which merit further theoretical and empirical investigations.

Due to the connection between rhetoric, epistemology, and reasoning, our work additionally informs the cross-cultural studies of information acquisition and processing in important ways. The popularity of argument from authority, for example, suggests that authoritative individuals may have been an important source of knowledge (Furedi, 2013). Similarly, the frequent use of argument by analogy in early China also mirrored the prevalence of analogical reasoning at the time (Zhang, 2000). Ultimately, effective rhetorical strategies in some cultural context should reflect its underlying epistemic standards regarding acceptable ways of acquiring knowledge, and our present study thus suggests that in addition to variations among contemporary societies in epistemic practices and norms (Machery et al. 2021; Nichols et al. 2003) there may be important historical variations worth exploring.

Cultural evolution of political rhetoric in China

According to extensive research on cultural transmission biases, humans have a tendency to preferentially adopt certain cultural variants based on specific contexts and content cues (Henrich and McElreath, 2003), such as the preference to conform to the majority (Henrich and Boyd, 1998), to learn from the more prestigious individuals (Henrich and Gil-White, 2001), to adopt the cultural practices with higher payoffs (Vale et al. 2017), or a combination of different learning mechanisms (Hong, 2022b; Laland, 2004). If we treat various rhetorical strategies as different “culture variants”, then individuals may exhibit context and/or content-dependent biases in selecting what rhetorical strategies to use. Note that although we have focused on payoff-biased transmission, in reality, people may take a number of factors into consideration when deciding how best to change people’s minds. Nevertheless, our analyses provide suggestive evidence of selective imitation based on persuasion outcome. Note that the present study does not exclude alternative cultural evolutionary forces: prestige bias, for example, could have played a role where individuals preferentially adopt the rhetorical strategies used by those perceived to be more prestigious. This possibility is not explored here due to the difficulty with which to assign prestige in our dataset (the concept of prestige itself and how it is measured empirically has been under some scholarly debate; see Chellappoo, 2021; Jiménez and Mesoudi, 2019). Conformist bias, on the other hand, was unlikely the dominating factor as it typically leads to the fixation of the most common variants, which is not what is observed in our dataset: even in later dynasties, there was a plethora of rhetorical strategies being used. Content biases (Henrich and McElreath, 2003) could also have played a significant role: certain rhetorical strategies may be intrinsically more attractive than others and, therefore, were maintained at high frequencies (e.g., argument by example).

The validity of categorizing rhetorical strategies: natural kinds or arbitrary decisions?

Throughout this paper we have taken the ontological status of various rhetorical strategies as unproblematic, which is obviously a rather naïve assumption. Strictly speaking, none of the rhetorical strategies that our experts identified are “natural kind” in a way that “carves nature at its joints”. The advance of natural language processing and unsupervised learning algorithms have made it possible to perform more objective machine-aided categorization, yet we believe our manual coding by experts provides a necessary and important first step. In deciding the categorization criteria, we also kept in mind the existing work in communication and literary studies; many of the rhetoric categories are well-known and discussed extensively in the literature, such as analogy (Reding, 1986; Xie, 2019), the use of historical examples (Kirkpatrick and Xu, 2012), and the invocation of proverbs (Hacking, 2013; Vatri, 2019) in argumentation. Our decision to treat historical examples and historical examples with outcomes as separate categories is because “historical example (without outcome)” merely states that so and so did X in the past with no consequences indicated, implicitly assuming that the listener knows why they should copy/or avoid the action, whereas “historical example with outcome” explicitly suggest a causality by making clear that doing X would lead to bad/good consequences, and it is likely that they trigger quite different psychological responses. We also attempted alternative binning methods, including combining “historical example” with “historical example with outcomes”, and our results did not qualitatively change (see Supplemental Information for more details).

Our analysis does not include emotion, an important rhetorical technique that Aristotle identifies as one of the three primary modes of persuasion (logos, ēthos, pathos). This exclusion is purposeful, as we rarely observed indications of emotional appeals (e.g., through the use of imperatives or references to deities, as noted by Serafim, 2019) in our dataset. Given that most persuasive efforts take place in private settings, there’s typically no audience to emotionally influence. Moreover, the unique dynamic between the persuader and the persuaded, often characterized by an inferior-superior relationship, implies that invoking emotion could pose unforeseen risks for the persuader.

At a deeper philosophical level, much more can be said about the naturalness of these rhetorical categories. Hacking (2016), for instance, suggests that arguments from examples and arguments by analogy are, in fact, both arguments from likeness, in which what’s being offered—analogies or examples—have to be enlightening to their audience, i.e., enabling them to view the focal case in new ways. Our categorization efforts as in most work in communication and literary studies, necessarily fall short of such philosophical rigor, and we hope that future work could combine insights from epistemology, formal reasoning, and cognitive science to offer better ways to study rhetoric and argumentation.