Abstract
This paper examines the critical, yet undertheorized role of context in verbal communication when computers are used to analyze language data. Drawing on multiple disciplines and prior work from communication and social psychology, an interactionist perspective is reviewed and applied to computational settings to demonstrate how individual, interpersonal, social, and situational factors afford or constrain verbal behavior. Using examples of language complexity and processing fluency research, this paper demonstrates how contextual factors like instrumental goal activation can reverse seemingly established relationships between simple language features and behavioral outcomes. The paper concludes by advocating for methodological approaches that systematically integrate contextual variables into natural language processing research, enabling scholars to resolve contradictory findings, enhance replicability, and develop a more nuanced understanding of how language reveals psychological processes.
Context is essential in communication. People want their messages to be taken in context to ensure that their intended meaning is conveyed accurately. Indeed, the consequences of misinterpretations can be significant for relationships. Take the following sentence, for example: “You are sanctioned.” This sentence requires an understanding of the context for a receiver to effectively interpret the speaker's meaning. Did the speaker suggest the receiver is allowed to perform some action, or that the receiver is prohibited from performing some action? This sentence is completely ambiguous without more information. Therefore, when context fails to be considered or understood, communication becomes problematic and requires work to fix the difference between what a speaker meant and what a receiver assumed they meant (Clark, 1996).
The prior example highlights the contextual nature of communication. Part of what makes for effective communication is a basic understanding what people said and meant during a speech act (Clark & Brennan, 1991; Levelt, 1989), accounting for other information that can lead to the correct interpretation of a statement (e.g., prior knowledge between communicators). There are many other aspects of communication (e.g., speaker characteristics, audience characteristics, relationship dynamics) that are contingent on context beyond the actual words that are uttered (though see Giles & Hewstone, 1982 for early treatment). By drawing on research from communication, psychology, philosophy, and other fields (Applegate & Delia, 1980; Argyle et al., 1981; Berger & Bradac, 1982; Burleson, 2009; Knapp et al., 2002; Lewis, 1981; Magnusson, 1981; Mehl et al., 2012), and supported by extensive work on goal-driven message production (Dillard, 1990; Wilson, 2002) and contextual influences on communication planning (Berger, 1997; Greene, 1984), the purpose of the current paper is to recenter and reignite the idea of context for the study of verbal behavior and its psychological correlates, particularly for computational evaluations of language data. This paper is admittedly not the first to attempt to do so (e.g., Boyd & Markowitz, 2025; Boyd & Schwartz, 2021; Burleson, 2009; Markowitz & Hancock, 2019; Mehl et al., 2012). However, in the current moment of increasingly approachable natural language processing tools (e.g., various computational models used to analyze words) and massive amounts of text data that exist at a researcher's fingertips, it is critical to review these insights about contextual affordances and constraints for computational analyses of language patterns and psychological dynamics. This paper, therefore, succeeds if it (1) encourages researchers to become more thoughtful in the way that they consider the contingencies on language use prior to embarking on a natural language processing study, and/or (2) encourages natural language processing researchers to explicitly test how context-related characteristics specifically impact verbal behavior and our understanding of underlying psychological processes. This paper does not attempt to find a “cure” for context being undertheorized, nor does it offer a definition of context based on its elusiveness (Asendorpf, 2009; Hogan, 2009; Magnusson, 1981; Reis, 2008). However, this work is a call for scholars to rethink, reimagine, and refocus context in the study of verbal behavior, concluding with pathways to do so in modern-day natural language processing research. It provides a multilayered heuristic of context when studying language patterns with computational approaches, which readers can hopefully apply and expand on in their own work.
Reigniting A Call for Context
While communication scholars have understood the importance of contextual factors in shaping communication (Applegate & Delia, 1980; Berger, 2008; Burgoon, 1993; Burleson, 2009; Giles & Hewstone, 1982; Greene, 1984; Waldron, 1997), computational approaches to natural language processing have been slower to systematically integrate these insights. There are several reasons why a deeper appreciation and articulation of context is necessary. First, communication acts and processes are complex, yet our understanding of the affordances and constraints that impact such communication acts is rudimentary. Take an example of “context collapse,” a phenomenon used to describe the flattening of “multiple audiences into one” on social media (Marwick & boyd, 2011, p. 122). According to this work, like people in offline settings, people in online settings self-present as a reflection of the “context and audience” (p. 123). The audience reasonably refers to the individuals people are communicating to, intended, or imagined (Litt, 2012). Though, what does context mean in this example? Context has both content (e.g., observable elements of a situation that might modify verbal behavior) and functional elements (e.g., aspects of a situation that can impact verbal behavior, but they are often difficult for people to process or perceive consciously). This distinction and its integration into modern-day natural language processing presents opportunities for enhancing computational social science research. Without a deeper understanding of context, researchers may be unclear about the boundaries of their phenomena (e.g., the conditions that apply to one phenomenon that may not apply to another) and the genuine psychological implications of their work.
Second, context needs better articulation in natural language processing research to maintain the credibility of extant theories and empirical evidence. Recent work in the social sciences suggests the replicability and reproducibility of many key studies were questionable (e.g., Open Science Collaboration, 2015), and mixed results across studies might be due to contextual differences (Pettigrew, 2018). Communication research has documented how situational factors, multiple goals, and relational dynamics may influence message production and thus, study-by-study results (e.g., Burgoon, 1993; Kellermann, 2004; Waldron, 1997), yet these contextual considerations are often treated as peripheral rather than central when describing modern-day natural language processing effects. Few papers outline precisely what “contextual differences” means when the evidence from primary studies conflict, or they provide a conceptualization of context that is too broad to be useful (e.g., any characteristic that is a “conceptually relevant characteristic”) (Falleti & Lynch, 2009; Gollwitzer & Schwabe, 2021). To better understand why some effects might replicate (or fail to replicate) across studies, a more refined understanding of context is needed.
This point is perhaps best demonstrated using examples from one of the present author's main research programs: deception production. Prior meta-analytic work has established that, on average, there are small but systematic relationships between verbal behavior and deception (Hauch et al., 2015). Crucially, the link between verbal behavior and deception is moderated by several important dimensions, including the production mode (e.g., the communication being typed, written, spoken), speaker motivation (e.g., no motivation, low-to-medium motivation, or high motivation), and other factors. Therefore, language patterns do not reveal deception universally, and contextual variables modify these relationships. As others have argued (Markowitz, 2020; Markowitz et al., 2023; Markowitz & Hancock, 2019), because of differences in psychological dynamics (e.g., the emotional and cognitive experiences of a liar during an act of deception), pragmatic goals (e.g., what a person is trying to accomplish with deception), and genre conventions (e.g., discourse community norms that shape how people communicate), we should not expect the same language dimensions to reveal deception in the same way across settings. Lying about taking a country to war (Alterman, 2004), for example, is likely to be associated with a different set of psychological antecedents and consequences compared to lying about love (Toma & Hancock, 2012) or scientific fraud (Markowitz & Hancock, 2016). We therefore need a more comprehensive, interactionist perspective that incorporates context into natural language processing research across subfields (Boyd & Markowitz, 2025), and it is time that psychology of language and natural language processing research adopt this framework for a more accurate understanding of how words reveal psychological dynamics.
To make this case beyond one field or domain, the current work first examines how scholars have defined context and its characteristics across fields. Drawing on these explications, an argument is then reviewed and advanced to incorporate an interactionist perspective—one that considers how individual, social environmental, and situational factors afford or constrain verbal behavior (Boyd & Markowitz, 2025)—to reveal underlying social and psychological processes through language patterns. This perspective draws on foundational message production and interpersonal communication research (Burgoon, 1993; Caughlin & Wilson, 2021; Dillard, 1990; Greene, 1984; Wilson & Sabee, 2003), plus work that describes how contextual factors systematically influence communicative planning and behavior (Berger, 1997, 2008; Waldron, 1997). The argument is reviewed, examined, and demonstrated using a case study of research on processing fluency and simple language use, followed by conclusions and pathways forward.
Context: Definitions and Articulations
The term context and its derivatives (e.g., words like contextual, contextually) are omnipresent. In fact, such terms have appeared in over 83% of recent communication science articles (using data from Markowitz, Song, et al., 2021), with common uses that include discrepancy resolution (e.g., context as a noun: “our work represents a different context”), a filler word that could be replaced with alternative terms (e.g., “in the context of climate change,” “within the current context”), and as a way to describe a concept (e.g., “context collapse”). When context is used as a noun, attempts to explain what it means or how it has been applied have varied. Table 1 offers descriptions across select sources from anthropology, artificial intelligence, communication, computing, linguistics, psychology, philosophy, and sociology. Some define context with synonyms such as the situation, environment, or circumstances (Bourdieu, 1990; Brown, 1996; Clark, 1996; Rodden et al., 1998; Ward et al., 1997). Others suggest context represents preconditions for communication (Wooldridge, 2009), or even a collection of items that subsume several characteristics in Table 1 (Stalnaker, 2014). While this list is intentionally nonexhaustive, it is largely representative of how scholars think about context. In sum, there are many ways to consider context, and trying to define the term concretely “may not be possible” (Duranti & Goodwin, 1992, p. 2), worthwhile, nor productive (Asendorpf, 2009; Hogan, 2009; Magnusson, 1981; Reis, 2008). Perhaps more important is to understand the functional role that context plays in communication, particularly how it can afford or constrain verbal behavior.
Explications of Context Across Select Fields.
The idea of applying functional theory to context and verbal behavior is supported by self-disclosure research. Early theorizing on self-disclosure argued that various aspects of the communication setting (e.g., disclosure goals) can activate or facilitate how people make disclosure decisions and their communication content (Derlega & Grzelak, 1979). In other words, aspects of the situation (e.g., media affordances that constrain who people communicate with) can change goals and content, as work on social networking sites as observed (Bazarova & Choi, 2014). The present perspective treats the functional role of context similarly. That is, various aspects of the communication setting can activate, afford, or constrain verbal behavior, changing one's psychological processing of events that can therefore impact their words.
A classic example to demonstrate the functional role of contextual constraints on communication is from affordance research itself. Affordances are opportunities for action that a technology or medium offers to people (Gibson, 1979). For example, a chair affords or provides the opportunity for sitting; a car affords or provides the opportunity for driving, and so on. Clark and Brennan (1991), in their examination of conversational grounding, outline various affordances (in a positive sense) or constraints (in a negative sense) that a medium might impose on interpersonal communication, including visibility (e.g., the ability for two or more people to be visible to each other), reviewability (e.g., the ability for two or more people to review messages before their transmission), and many others. These affordances and constraints impact how people communicate with others (Clark, 1996) and the psychological meaning underneath the words people use to verbalize their experiences (Pennebaker, 2011). To the degree that scholars are interested in understanding how words reveal information about the individual mind and the psychological underpinnings of verbal behavior (e.g., Boyd & Markowitz, 2025; Boyd & Schwartz, 2021), a more thoughtful application and consideration of context is needed. Words are not communicated in isolation nor are they merely verbal performance; they are impacted by a range of factors—affordances and constraints—that shape what people say, how they communicate, and thus, the psychological meaning underneath such language.
Setting aside the futile effort of providing a general definition of context or situations (Asendorpf, 2009; Hogan, 2009; Magnusson, 1981; Reis, 2008), it is important to at least argue for various aspects of it that might matter when trying to understand how verbal behavior can reveal social and psychological dynamics. Words, according to the psychology of language tradition (Pennebaker, 2011), are widely understood to be markers of one's attentional processes and psychological focus (Boyd & Markowitz, 2025; Boyd & Schwartz, 2021). People who use a high rate of self-references (e.g., I, me, myself) are therefore attending more to the self than people who use a low rate of self-references. This perspective has been a dominating theoretical force to understand the underlying meaning of words and what people are thinking, feeling, and experiencing psychologically via language (Tausczik & Pennebaker, 2010).
As noted, we have reached a tipping point in our understanding of how language patterns reveal a person's psychology (Boyd & Markowitz, 2025). With rapid advancements in computational methods and our refined toolkit to measure verbal behavior, scholars need to do more than assess simple correlations between words and outcomes (e.g., behavior, perceptions) to understand how language patterns reveal psychological information about people (Ganesan et al., 2024). Thus, Boyd and Markowitz (2025), and other research in more specific subfields (e.g., Boyd et al., 2020), make the case that an interactionist perspective—one that can articulate “who will do what, when they will do it, and why” (Boyd & Markowitz, 2025, p. 417)—can help to understand how the psychology of an individual person can be revealed through words. Various layers to this interactionist perspective include the individual (e.g., affective, cognitive traits inherent to the person), the individual's social environment (e.g., who people are communicating with), and situational factors (e.g., time, location). 1 Each of these layers can functionally afford or constrain how people think, feel, process information, and therefore ultimately communicate. However, this interactionist perspective is rarely modeled in modern-day natural language processing to consider how words reveal underlying social and psychological dynamics. The hope is that the present paper offers a heuristic of context that can be modeled and expanded on in natural language processing research for broad applications of the interactionist perspective.
The value of applying this interactionist perspective is demonstrated in the next section through an evaluation of language simplicity and complexity using papers in the present author's own research program. The purpose of analyzing these specific papers is to be self-reflective, and to avoid finger-pointing at other scholars who might perceive their inclusion in this article as an indictment of their work. However, through this demonstration, the hope is that readers recognize the interactionist perspective as a clear and relevant way to advance context-sensitive scholarship in the psychology of language field using natural language processing techniques.
The Curious Case of Complexity
Consider the following set of studies that evaluated the relationship between language simplicity and behavioral outcomes such as engagement and online petition signing (Markowitz, 2023b; Markowitz & Shulman, 2021). In the earlier paper, Markowitz and Shulman (2021) drew on metacognition research and feelings-as-information theory (Schwarz, 2004, 2012) to argue that people tend to process simple (vs complex) information because simplicity feels better to most people compared to complexity. People use their feelings, often heuristically (Shulman et al., 2021, 2024), when making judgments or decisions, and simple (fluent) information is typically easier to process than complex (disfluent) information. This pattern has been established in many studies (Alter & Oppenheimer, 2009) with various operationalizations of the fluency concept. Lexical fluency, for example, considers the ease associated with processing simple (or complex) words (e.g., job vs occupation, use vs utilize) and is often the most common way to manipulate fluency in the lab (Markowitz, Kouchaki, et al., 2021; Oppenheimer, 2006, 2008). Stylistic fluency considers the ease associated with comprehending a text based on the communication style of one's writing. A text that is written analytically—one that contains high rates of articles and prepositions and low rates of pronouns, adverbs, conjunctions, negations, and auxiliary verbs (Markowitz, 2023a; Pennebaker et al., 2014)—is more complex than a text that is written narratively (e.g., like a story). Finally, structural fluency considers the ease associated with comprehending a text based on sentence and word length (Markowitz, 2023b) and is often measured with readability metrics (Flesch, 1948).
The results from Markowitz and Shulman (2021, Study 1) suggested that in the absent of the activation of an instrumental goal, people responded positively to simplicity and favored such simple writing with more online endorsements. Journalists, viewers of entertainment media (e.g., TED talks), and readers of science tended to like, view, and share content more when it was written in a simple, lexically fluent manner versus a complex, lexically disfluent manner. Crucially, however, the link between simple writing and behavioral outcomes is not universal (Markowitz & Shulman, 2021, Study 2). Informed by research on the instrumentality heuristic (Labroo & Kim, 2009), which suggests when people try to achieve an instrumental goal or they believe someone else is trying to do so, people tend to feel positively about such effort and they reward complexity with their money and support (compared to simplicity). In other words, the benefits of simplicity are contingent-on-effort (Markowitz, 2024b). This pattern of results, where lexical disfluency (complexity) associates with downstream financial benefits to people (Markowitz, 2019, 2024b; Markowitz & Shulman, 2021, Study 2) and support for prosocial causes (Markowitz, 2023b), demonstrates a possible boundary condition on the simpler-is-better hypothesis that most processing fluency research has espoused.
Taken together, the prior examination of simple language suggests that in some cases, simplicity is associated with positive behavioral outcomes and in other cases, complexity is associated with positive behavioral outcomes. While the effects of simplicity may therefore be “contingent on context,” papers using computational approaches to natural language processing have often lacked systematic frameworks for couching or operationalizing these contextual factors. Building on foundational work in communication research (Dillard, 1990; Dillard et al., 1989), an interactionist perspective and a functional appreciation of the role of context in verbal behavior suggest a more comprehensive investigation would account for how instrumental goal activation modifies or constrains the relationship between simplicity and behavior. Instrumental goal activation may not be the only relevant psychological factor to do so but it is at least one theoretically reasonable moderator to change the nature of the relationship between such variables at the language level.
This description exemplifies how context functions as more than the background or preconditions for communication. Contextual factors actively shape the psychological processing that underlies verbal behavior, but they do so in subtle ways that may be out of our awareness. When an instrumental goal is activated, such as trying to convince someone to support a petition or donate to a cause, the cognitive processing of language complexity possibly shifts from a heuristic evaluation (“Is this easy to understand?”) to a more substantive assessment of effort and value (“Does this complexity signal appropriate effort toward an important goal?”). The instrumentality heuristic thus creates a lens through which complexity is reinterpreted or appraised as a positive signal rather than a processing burden. This nuanced understanding of language use would be lost without a more holistic, interactionist perspective of verbal behavior.
Together, in the spirit of trying to inspire a more comprehensive, thoughtful, and theoretically rigorous future for the study of verbal behavior, it is important to consider how language interacts with various individual, interpersonal, social, and situational factors. Rather than treating language as having direct, universal relationships with outcomes, we must develop and apply theoretical frameworks that explicitly account for how these relationships are afforded or constrained by multilevel, layered, and simultaneous contextual factors. This requires moving beyond the idea that “context matters” in natural language processing with computational approaches toward specifying which aspects of context matter, how they matter, and the psychological mechanisms through which they exert their influence on language and its behavioral consequences. These aspects are interrogated below.
Applying a Layered Approach to Context and Verbal Behavior
The layered, interactionist approach to context and verbal behavior reviewed in this paper is supported by communication theories that account for the contingent nature of interpersonal communication and message production. For example, message production research (Dillard, 1990; Wilson, 2002; Wilson & Sabee, 2003) describes how person-specific traits, multilayered goals, and cognitive processes shape communication at the individual level. Planning theory (Berger, 1997, 2008; Waldron, 1997) describes how communicators develop mental representations of their interactions to account for relational constraints—consistent with the interpersonal layer that is more fully described below. Expectancy violations theory (Burgoon, 1993) describes how one's social environment can create norms and context-bound expectations that influence message production and message interpretation. Finally, action assembly theory (Greene, 1984) describes how cognitive processes and situational factors influence message production (e.g., situational elements influence the activation of certain cognitive processes and how they are assembled). Together, while there may be other theoretical perspectives that can inform the present perspective, these are some of the most salient and well-established theories to translate how communication principles can inform context-aware natural language processing. The goal of the current perspective is not to reinvent our understanding of contextual influences on verbal behavior but to provide natural language processing researchers with theoretically grounded frameworks to capture and appreciate these dynamics in their work.
As we chart a future for a more thoughtful, context-integrated analysis of verbal behavior, several critical questions emerge, including but not limited to: (1) How do we systematically identify which contextual variables meaningfully afford or constrain communication?, (2) What methodological approaches can capture these multilevel interactions?, and (3) How might we integrate context directly into natural language processing to test our theories? A pathway forward requires developing comprehensive taxonomies of contextual variables organized by their functional and psychological impacts on communication processes, creating standardized methods for measuring these variables across studies, and building theoretical models that specify the mechanisms through which context shapes language.
One starting point might be the situational DIAMONDS proposed by Rauthmann et al. (2014). In this work, the authors attempted to identify various aspects of a situation that might be psychologically meaningful to understand the human condition: Duty (e.g., the degree to which a situation requires a task needing to be fulfilled), Intellect (e.g., the degree to which a situation requires deep thinking and cognition), Adversity (e.g., the degree to which a situation contains threats or negative events), Mating (e.g., the degree to which a situation contains romantic relationship possibilities), pOsitivity (e.g., the degree to which a situation is enjoyable or has potentially favorable outcomes), Negativity (e.g., the degree to which a situation is unenjoyable or has potentially unfavorable outcomes), Deception (e.g., the degree to which a situation has the possibility of deception or distrust), and Sociality (e.g., the degree to which a situation involved social engagement) (Rauthmann et al., 2014, 2015). To evaluate the psychological utility of these situational characteristics, Rauthmann et al. (2014) connected the dimensions to personality traits and goal affordances, among other meaningful characteristics, demonstrating that situations systematically vary in psychologically meaningful ways, and these differences predict important behavioral and experiential outcomes. The point of introducing this taxonomy is to demonstrate that some scholars have started to think deeply about what a situation means, functionally and psychologically, instead of focusing on the basic definition of a situation or context, generally. A consistent undertaking can, and should, carryover to verbal behavior in natural language processing research.
For verbal behavior, relevant elements of a situation and context must begin at an individual level and then become more interpersonal, social, and situational (norms-related) in terms of layering and abstraction (for a visual description, see Figure 1). Note, this layered approach builds on, complements, and extends prior work by Boyd and Markowitz (2025), but much of this perspective is repeated in the current manuscript for completeness. The first layer involves an understanding of the individual mind—what a person may be thinking, feeling, or experiencing psychologically, as inferred by verbal behavior, in the moment they are communicating (Boyd & Markowitz, 2025; Tausczik & Pennebaker, 2010). This is often a difficult task, prospectively, because onlookers and researchers cannot be mind-readers. We often try to infer why someone communicated in a particular manner (Berger & Bradac, 1982; Dillard et al., 1989; Kellermann, 2004; Kellermann & Shea, 1996), and how this links to their psychological state, post hoc. This first layer, related to the individual, represents the majority of natural language processing research in the psychology of language field. Scholars attempt to understand what people are thinking and feeling by analyzing words across various settings of interest, ignoring other consequential aspects of the situation and communication setting that might impact how or why their communication data were verbalized in a particular way.
Visual Description of the Layered Approach to Understanding Verbal Behavior, Contextual Affordances, And Contextual Constraints.
Who a person communicates with is the next layer following the individual level. Specifically, what is the nature of the interpersonal relationship between two (or more) individuals? Relevant questions include, but are not limited to: How long have these individuals known each other, what is the status of their relationship, and how do we know these characteristics about the two (or more) individuals? These questions are often considered in communication research that documents how relational dynamics create specific constraints on conversational planning and message production competence, particularly in challenging interpersonal settings (Waldron, 1997; Waldron & Applegate, 1994). Indeed, Waldron's work shows that people who use more complex cognitive frameworks to interpret the thoughts, feelings, and behaviors of others tend to develop more sophisticated conversational plans that account for relational constraints (for a review and theoretical commentary, see Waldron & Cegala, 1992). Individual cognitions and cognitive abilities, therefore, interact with interpersonal contextual factors to influence message production. Together, the purpose of this second layer is to account for the simple and clear acknowledgement that not all interpersonal relationships are the same; we should model and study them in ways that account for their uniqueness. The present examples highlight how the target of one's communication matters for verbal behavior, but there are a range of other interpersonal considerations that will impact what and how people communicate, including various person perceptions (Fiske, 2018; Fiske et al., 2002), status differences (Kacewicz et al., 2014; Markowitz, 2018), and demographics (Markowitz, 2022; Newman et al., 2008), among additional considerations.
The third layer is more social and abstract in nature, beyond the interpersonal relationships between people. Who is near the interpersonal communicators, and how do these social interlocutors impact discourse production? Imagine communicating with another person at a dinner party, and the people sitting nearby are also close friends with you. It might be acceptable to disclose personal information at that setting (e.g., recent health news, concerns about the state of the world), given such relational closeness. However, if you and your partner are seated next to strangers, you might decide to disclose different content (e.g., the weather, more general topics) and communicate in a different style because the interlocutors are unfamiliar. Even in an interpersonal conversation, the social environment (e.g., who is around the conversation, the relationship to others in the surrounding area, one's perception of the environment, publicness or privateness) can impact what people say and how their words reveal social and psychological characteristics of the communicator and the relationship (Mehl et al., 2012).
Finally, external forces that modify how people communicate, constitute the broader situational or norms-related layer. People write Facebook posts differently than they write love letters; they may write science papers differently for one field (e.g., psychology) than they write science papers for another field (e.g., communication). Indeed, expectancy violations theory illustrates how the norms of different discourse communities can influence both message production and message interpretation, with violations of situational expectations (e.g., using casual language when formal language is expected) altering communication outcomes depending on relational factors (Burgoon, 1993). According to this work, the same verbal behavior can be interpreted positively or negatively depending on one's level of adherence to context-specific expectations. Therefore, the norms and conventions of a discourse community shift how people communicate even before an utterance is made (Biber et al., 2007; Markowitz & Hancock, 2019). These shape-shifting extrinsic factors can set verbal expectations prior to a person entering a space. Consider an example from social media, where French and Bazarova (2017) measured perceived expectations for a response on Facebook, Twitter, and Instagram across 118 participants. The authors observed that people anticipate interactions differently across these platforms (e.g., people expected the fewest responses from Twitter), indicating the “platform” as a situational factor that can impact this social dynamic. In psychology of language studies, the norms-related layer is loosely accounted for by scholars conducting studies on different platforms with different participants (e.g., language-level differences that arise are therefore attributed to platform traits and inconsistencies). More within-person studies are needed to identify the genuine impact of place-specific conventions shaping how people use language to reveal psychological processes and internal states of communicators.
Together, building on prior scholarship (Boyd & Markowitz, 2025; Boyd & Schwartz, 2021), this approach suggests there are at least four layers (i.e., individual, interpersonal, social, and situational) that require consideration when trying to measure how verbal behavior is impacted by contextual forces. This layered conceptualization is consistent with planning theory (Berger, 1997, 2008), which demonstrates how communicators develop hierarchically organized mental representations that guide message production. It is also consistent with action assembly theory (Greene, 1984), which describes how situational factors influence the cognitive processes underlying message retrieval and organization, with context-related elements impacting the assembly of behavior during communication. It is possible that there are more interstitial layers between those that are individual, interpersonal, social, and situational, which will require further interrogation and examination as well. Integration of these layers within existing situational taxonomies (Applegate & Delia, 1980; Burleson, 2009; Knapp et al., 2002; Miller, 1990; Rauthmann et al., 2014) may provide a fruitful line of future work.
Perhaps the value of this layered perspective can also be demonstrated through a metaphor. Imagine a researcher examining verbal behavior through a microscope. At the highest magnification, we can observe the intricate psychological landscape of the individual mind in the moment of communicating (e.g., using words to infer what a person is thinking or feeling; Pennebaker, 2011). But, as we pull back the lens and reduce magnification, new layers of the person, their relationship, and those in the communicative surround emerge into focus. At moderate zoom, therefore, we can see the interpersonal dynamics between communicators. Zoom out further, and the social constellation surrounding them becomes visible. We can only begin to appreciate this layer after we have focused on the individual and then on their interpersonal and social environment. Zoom out to the widest view, and we can observe the situational norms that shape communication in ways that may be out of awareness for the person. Just as a biologist would likely not attempt to understand an ecosystem by examining only cells under high magnification, researchers studying verbal behavior must resist the temptation or convenience to focus on just one of these layers (e.g., likely, the individual mind). The richness and psychological impact of verbal behavior become apparent when we focus on all four layers interactively, recognizing that each level of analysis in isolation reveals incomplete patterns or results that are perhaps invisible at other magnifications.
By advancing from vague references of “context” toward precise specifications of contextual affordances and constraints within these layers, researchers can seemingly resolve contradictory findings, enhance the replicability of verbal behavior research, and develop more nuanced understandings of how language reveals psychological processes. Modeling and studying context may be inherently difficult, however, and one's perceptions of the context are more tractable instead (e.g., what people believe are contextual affordances or constraints on their communication). Experimental approaches have been successful in identifying how writing within a context or genre (e.g., discussing one's opinions about friends or abortion) differs from explicit lying and truth-telling about the same topic at the language level (Markowitz & Griffin, 2020), making contextual elements at least manipulable in the lab.
Pathways Forward
This paper, supported by seminal research in communication science and social psychology, has examined the role of context in studying verbal behavior from a computational perspective. The interactionist perspective offered by Boyd and Markowitz (2025) provides valuable scaffolding for understanding how various factors—individual, interpersonal, social, and situational—interact to afford or constrain verbal behavior. As we develop a more nuanced understanding of the contextual nature of verbal communication, several pathways emerge that researchers and practitioners should consider.
First, future research should embrace methodological approaches that can account for multiple levels of context simultaneously. Traditional approaches to natural language processing have often treated psychological processes as occurring primarily within the individual, with context serving as a peripheral consideration, a statistical control, or as something ignored altogether. An interactionist perspective requires multilevel modeling that accounts for how individual factors (cognitive traits, affective states), interpersonal and social environmental factors (audience characteristics, relationship dynamics), and situational factors (time, location, medium) jointly shape language use. For example, consider a politician delivering a speech about climate change. Their language complexity, and its association with downstream outcomes, may vary based on their expertise in the subject (individual factors), who they are addressing (interpersonal factors), who is nearby being addressed or not (social factors), and whether they are speaking at a climate conference or a town hall meeting (situational factors).
To do this, scholars should consider integrating context-related metadata into language analysis instead of just accounting for their variance. For instance, a politician might use complex language when discussing climate change with scientists (where complexity signals expertise) but simpler language when discussing the same topic at a campaign rally (where simplicity might signal accessibility). Without accounting for this contextual variation, we risk misinterpreting the psychological meaning of language patterns—either by incorrectly attributing them to stable individual traits when they actually reflect contextual adaptations or by failing to detect the critical contextual moderators that determine when and how particular linguistic features signal different psychological states.
A promising pathway toward understanding contextual affordances and constraints in verbal behavior is to leverage generative Artificial Intelligence (AI) and large language models (LLMs) as tools for bottom-up context discovery. Rather than imposing researcher-defined frameworks of what constitutes context, LLMs can identify patterns of contextual variation that humans might overlook. For example, an LLM analyzing millions of online reddit posts could identify that the relationship between language complexity and user engagement systematically varies not just with instrumental goal activation but also with factors such as domain expertise of the audience, communication platform affordances, and cultural dynamics. This approach is particularly suited to identifying complex, multilevel contextual interactions that traditional methods might miss. Consider how an LLM might analyze scientific communication across platforms, discovering that while complexity generally impairs perceptions of the scientists (e.g., Markowitz, 2024a), this relationship reverses when experts communicate with other experts about topics of shared interest. In this example, a context-sensitive interactionist framework combines individual, interpersonal, social, and platform-level factors. By allowing AI systems to inductively identify contextual patterns and parameters, researchers can develop richer taxonomies of contextual affordances and constraints that reflect the multilayered nature of human communication. In fact, if we do not leverage LLMs in this way, these models may fall victim to systematic errors when trying to provide tangible, useful, and context-aware feedback to users. By failing to capture context in language, we squander the potential of LLMs to decode human behavior and our psychological understanding of communication.
While computational and automated means are potentially promising ways to identify contextual characteristics at scale and in novel settings, there are basic questions that can be addressed through more manual means as well. Suppose a scholar is interested in studying verbal behavior related to deception (e.g., how do liars communicate compared to truth-tellers), measuring the impact of the person and the situation on language use. A scholar may attempt to obtain the record of text messages between a person and 10 of their closest and most distant contacts. Retrospectively, the participant can go through their log and identify messages that were deceptive and truthful, using such verbal data to infer what is occurring in the mind of the individual when they communicated falsely or truthfully. For a more holistic evaluation of deceptive verbal behavior, however, consider the same basic design but from an interactionist perspective that accounts for multiple layers of an individual, interpersonal relationship, social environment, and norms. At the individual level, the deceiver's psychological state in the moment of lying and communicating becomes crucial (Markowitz & Hancock, 2019). A person concealing a minor social transgression (e.g., a white lie) may experience fundamentally different cognitive and emotional pressures than someone hiding a serious transgression (e.g., infidelity). The former might display verbal patterns associated with social anxiety (e.g., perhaps, representing a psychological distancing effect; Newman et al., 2003) with reduced self-references compared to those who are truth-telling. The latter might show signals of negative affect when lying versus truth-telling, given the negative psychological impact that deceiving a spouse might have on the individual. Understanding what the individual deceiver is thinking and feeling is the first step in a more nuanced evaluation of deceptive communication.
The second, interpersonal layer introduces relational dynamics that may alter deceptive language patterns and provide texture to an understanding of why certain language patterns were communicated to certain individuals. When deceiving someone who shares common ground, liars might rely on omission, producing language that appears cooperative and collegial, but lacks the specificity or details that shared knowledge would normally contain (Caso et al., 2023; Nahari et al., 2014). Conversely, this interpersonal dynamic might be absent when two people are less familiar with each other. The history and connection between communicators create distinct affordances and constraints on how deception can be accomplished at the language level.
Third, at the social environment level, the presence and characteristics of observers modify deceptive behavior in systematic ways. A person lying in private faces different pressures than someone deceiving in front of others (e.g., Mehl et al., 2012). Group dynamics and various social forces may shape the deceiver and their linguistic strategies (Dunbar, 2017). If text messages were one-on-one or in a group setting, this might change how people communicate, what they lie (or tell the truth) about, and how words represent such psychological states. Therefore, accounting for the social environment matters when looking underneath words. Finally, the norms layer considers broader forces that define what constitutes appropriate or expected communication within specific settings. Communicating about the inner workings of a relationship is associated with different sets of norms and expectations than communicating about other topics such as weekend plans, politics, and medical advice. Norms are critical to account for in (verbal) behavior research (Lapinski & Rimal, 2005), as they dramatically impact how people think, feel, and psychologically represent their experiences.
This layered approach suggests that effective identification of liars and truth-tellers at the language level cannot rely on universal linguistic cues. Instead, researchers must account for how contextual dimensions simultaneously interact to shape deceptive language patterns. A more sophisticated approach might begin by establishing baseline norms (and rates) for truthful communication within specific settings (e.g., online dating, scientific fraud, politics), then identifying how individual psychological states, relational dynamics, the social environment, and conventions combine to produce deviations from those baselines. Rather than asking “What words do liars use?,” perhaps a more generative question becomes “How do contextual and situational forces shape how a person, in this relationship, within this social setting, and under these particular circumstances represents verbal deception?” Independent of studying deception, this shift from universal markers revealing psychological dynamics to contextual considerations represents not just a methodological refinement but a fundamental reconceptualization of how language reveals psychological information about the human condition.
Conclusion
An axiomatic refrain in academic and public life is that words matter. This paper has argued that words matter to the degree that they are understood through the lens of functional and contextual affordances and constraints, which can provide fundamentally richer insights into psychological processes than what decontextualized or a-contextual analyses can achieve. The interactionist framework presented here (e.g., layering individual, interpersonal, social, and situational factors that can simultaneously afford or constrain verbal behavior)—building on decades of communication research documenting contextual influences on message production (Berger, 1997, 2008; Burgoon, 1993; Greene, 1984; Waldron, 1997; Wilson & Sabee, 2003) and in other work (Boyd & Markowitz, 2025)—transforms how researchers can approach psychological insights from language patterns in natural language processing research. Rather than seeking universal markers of psychological states or assuming the same verbal markers will apply in the same way across settings, the hope is that scholars can develop context-sensitive models that account for layered factors that determine when and how verbal features signal underlying psychological processes.
Footnotes
Author’s Note
Parts of this article's content are drawn from a talk given at the University of California, Santa Barbara, as part of the 20th Annual James J. Bradac Lecture.
Acknowledgments
The author thanks Ryan Boyd and Hillary Shulman for their thoughts on an earlier version of this manuscript.
Funding
The author received no financial support for the research, authorship, and/or publication of this article.
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.