The Mechanistic Framework of Alignment: A Unified Model

Abstract

Conversational alignment, also known as accommodation, entrainment, interpersonal synchrony, and convergence, is defined as the tendency for interlocutors to exhibit similarity in their communicative behaviors. There have been many theories and explanations set forth as to why alignment occurs and, accordingly, the mechanisms that underlie it. To date, however, alignment research has been largely siloed, with different research teams often examining alignment through the lens of a single theoretical account. Considering causal mechanisms in tandem offers a more holistic and nuanced understanding of the dynamic nature of alignment, its purposes, and its consequences. Accordingly, we propose the Mechanistic Framework of Alignment (MFA), a qualitative conceptual model that integrates existing theories of conversational alignment into one unified framework. To explain this framework, we first review five alignment mechanisms, discussing the underlying assumptions, contributions, and supporting evidence for each. We then introduce two overarching factors—conversational goal and alignment type—that are critical for understanding when and how these mechanisms give rise to aligned behavior. Illustrative examples demonstrate how the relative weightings of each mechanism interact with these contextual variables. Finally, we conclude with directions for how future research can extend and refine this framework and how the MFA can support future work in this area.

Introduction

During conversation, interlocutors exhibit interdependent similarity in their communicative behaviors with one another. This similarity is known by many different names including alignment, accommodation, entrainment, interpersonal synchrony, and convergence. For simplicity, throughout this manuscript, we will refer to this process of producing similar behavior¹ as alignment. Alignment has been demonstrated across many different communicative behaviors including syntax (e.g., Branigan et al., 2000; Schoot et al., 2019), lexical choice (e.g., Brennan & Clark, 1996; Garrod & Anderson, 1987), semantics (e.g., Duran et al., 2019; Didericksen et al., 2022), phonetic and prosodic features of speech (e.g., Borrie et al., 2019, Wynn & Borrie, 2020), facial expressions (e.g., McNaughton et al., 2024; Drimalla et al., 2019), gestures (e.g., Holler & Wilkin, 2011; Rasenberg et al., 2022), and body movements (e.g., Shockley et al., 2003; Paxton & Dale, 2017). It has been observed in highly controlled laboratory tasks (e.g., Wynn et al., 2018; Borrie & Liss, 2014; Shin and Christianson, 2011) and naturalistic conversations (e.g., Borrie et al., 2015; Cohen Priva & Sanker, 2020). It occurs across different contexts such as task-based (e.g., Kruyt et al., 2022; Pardo, 2006) and affiliative conversations (Cohen Priva & Sanker, 2019; Dale et al., 2020). It is manifested across cultures and across language groups (e.g., Chieng et al., 2024; Misiek & Fourtassi, 2022). It happens in young children (e.g., Hopkins & Branigan, 2020, Paquette-Smith et al., 2022) and continues into adolescence (e.g., Wynn et al., 2023; Schertz & Johnson, 2022) and across the adult lifespan (e.g., Trujillo et al., 2023; Koehler et al., 2021). Further, people align to a variety of different types of conversation partners—children (e.g., McNaughton et al., 2024; Fusaroli et al., 2023a) and adult (e.g., Lubold et al., 2019; Ko et al., 2016), familiar (e.g., Pardo et al., 2012; Lee et al., 2010) and unfamiliar (e.g., Pardo et al., 2010, Cohen Priva et al., 2017), native (e.g., Dideriksen et al., 2023; Ostrand & Ferreira, 2019) and non-native (e.g., Suffill et al., 2021; Olmstead et al., 2021), neurotypical (e.g., Wynn et al., 2022; Borrie & Delfino, 2017) and neurodivergent (e.g., Borrie et al., 2020; Fusaroli et al., 2023b), and human (e.g., Riorden et al., 2014; Yuan et al., 2024) and non-human (e.g., Bell et al., 2003; Branigan et al., 2010) partners. Thus, alignment is considered a pervasive communication phenomenon during interaction. A fundamental and persisting question within the literature is why such similarity of communicative behavior occurs.

There have been many different theories and explanations set forth as to why alignment happens during conversation. These theories have focused both on the mechanisms at play and the functions of alignment or the benefits that result from aligned behavior. While several researchers have highlighted the multifaceted nature of alignment and the need to consider multiple perspectives (e.g., Ferreira & Bock, 2006; Rasenburg et al., 2020; Costa et al., 2008), this is not always reflected within the current literature. Rather, certain fields of study or research teams often focus on one specific explanatory account without considering other alignment mechanisms that may be at play. However, as research in this area continues to evolve and move forward, a more concentrated focus on considering multiple reasons for alignment in tandem will help us gain a more holistic, nuanced, and in-depth understanding of the dynamic nature of alignment, its purposes, and its consequences. To this end, we propose the Mechanistic Framework of Alignment (MFA).

The MFA, a qualitative conceptual model, can be viewed in Figure 1. This framework is built on three fundamental principles. First, there are multiple causal mechanisms driving alignment which in turn reflect multiple functions of alignment. Second, these mechanisms are not mutually exclusive. Third, alignment mechanisms are contextually situated within a larger landscape such that they simultaneously shape and are shaped by the broader interactional context. The purpose of presenting this conceptual framework is twofold. First, we map the broad landscape of this area of research by consolidating existing theories and ideas about the reasons people align into a systematic taxonomy. Second, we provide a structured, integrative way to characterize the dynamic interplay between alignment mechanisms and the contextual factors that impact and are impacted by these mechanisms. Our aim in presenting this framework is to provide a foundation to inspire continued discussion, research, and collaboration that supports a more integrated and holistic understanding of this complex topic.

Figure 1.

A visual representation of the Mechanistic Framework of Alignment (MFA). Alignment mechanisms are sub-divided into goal-directed and non-goal-directed mechanisms. Relative weightings of goal-directed mechanisms are impacted by factors related to the conversational goals (which are impacted by the environment, speaker, and listener). Alignment mechanisms, in turn, impact factors related to the type of alignment (e.g., linguistic domain, form, temporal dimension) exhibited during an interaction.

To explain this framework, we first review five alignment mechanisms (with their associated functions), based on existing theories and models, and discuss the underlying assumptions, contributions, and evidence supporting each causal mechanism. Here we note that for the sake of simplicity and clarity, we use the term mechanism broadly (recognizing that this may in practice subsume a set of more granular processes). After discussing each mechanism, we introduce two overarching factors that are critical for understanding when and how these mechanisms give rise to aligned behavior and provide illustrative examples of how relative weightings of each mechanism may impact or be impacted by these contextual variables. Finally, we conclude with a discussion on how future research can extend and refine this framework and how this framework can, in turn, support and guide future work in this area.

Alignment Mechanisms

Priming Mechanism: Alignment as an automatic and non-goal-directed phenomenon

A fundamental distinction in accounts of alignment is between theories arguing that alignment is an automatic and non-goal-directed behavior and theories conceptualizing alignment as a goal-directed (albeit potentially subconscious) behavior. The idea of alignment as a non-goal-directed priming mechanism is suggested by multiple theoretical accounts (e.g., Interactive Alignment Model, Dual Path Model). While there are key differences between these accounts (discussed below), they reflect the shared viewpoint that alignment occurs automatically. That is, when an interlocutor produces a particular communicative representation (e.g., word, syntactic structure, acoustic pattern), it automatically activates this representation in their conversation partner. This increases the likelihood that the partner will use the same representation in their own communicative productions.

Empirical support for alignment stemming from a priming mechanism comes from studies showing that alignment occurs when people are exposed to language in the absence of an overt social context (i.e., listening and responding to prerecorded sentence productions, e.g., Borrie & Liss, 2014; Wynn et al., 2018). Other studies have found that individuals with anterograde and Korsakoff’s amnesia (i.e., people showing profound impairments in declarative memory) exhibit alignment, further supporting the idea of alignment as an automatic and non-goal-directed process (Ferreira et al., 2008; Heyselaar et al., 2017). While priming can be conceptualized as one overarching alignment mechanism, different theoretical accounts make key distinctions regarding how this priming mechanism operates and its overall function. We highlight two major theoretical accounts below.

Priming as cognitive economy

The idea of alignment as a priming mechanism is a central tenet of the Interactive Alignment Model (Pickering and Garrod, 2004; 2006). This theory views alignment as arising from short-term activation of matching representations in the speaker and listener that facilitates linguistic processing in both interlocutors, thereby promoting cognitive economy. Because levels of representation are inherently interconnected, alignment at one level will automatically lead to alignment of other levels as well. For example, priming at the lexical level should induce alignment at the phonetic level and vice versa. In this way, alignment of linguistic representations via priming ultimately leads to alignment of situation models (i.e., shared understanding of the situation under discussion). Crucially, in this account alignment is not intentional or goal-directed. Rather, any communicative benefit gained (i.e., reduction in cognitive effort, development of shared understanding) is considered to be an implicit by-product of the automatic priming mechanism.

A number of empirical studies provide support for the Interactive Alignment Model. For instance, some studies have shown that interlocutors are more likely to align their communicative behavior immediately after their partner’s utterance, consistent with assumptions about activation decay (Branigan et al., 1999; Branigan & McClean, 2016; Hartsuiker et al., 2008). Further, the postulation that alignment at one level induces alignment at subsequent levels is supported by research showing correlations across alignment levels such as prosodic and lexical features (Rahimi et al., 2017) and syntactic and lexical features (Branigan et al., 2000), although other researchers have found this relationship to be complex or questionable (Wiese & Levitan, 2018; Ostrand & Chodroff, 2021; Patel et al., 2022).

Priming as implicit learning

While the Interactive Alignment Model interprets alignment as short-term activation of linguistic representations, other researchers have noted that priming often leads to long-lasting changes in speakers’ behavior. Accordingly, they have suggested alignment via priming to be a form of implicit learning (e.g., Bock and Griffin, 2000; Chang et al., 2006; Ferreira and Bock, 2006; Jaeger & Snider, 2013; Reitter et al., 2011). When a listener encounters specific linguistic expressions (e.g., when they hear a specific grammatical structure), it strengthens the representations associated with these expressions. As a result, these representations are easier to retrieve and more likely to be re-used in future utterances (e.g., Chang et al., 2006). Thus, alignment contributes to implicit and automatic language acquisition in children (Branigan & Messenger, 2016; Messenger, 2021; Speidel & Nelson, 1989; Tomasello, 1992) as well as linguistic adaptation in adults (Bock & Griffin, 2000)².

Studies supporting alignment as an implicit-learning mechanism include evidence that alignment patterns persist even after a long delay between an interlocutor hearing a target utterance and producing their own response (e.g., Bock & Griffin, 2000). Further, alignment is more likely to occur when interlocutors are presented with a less common or more surprising linguistic expressions (e.g., Jaeger & Snider, 2013; Wei et al., 2022; Kaschak et al., 2011), suggesting an inverse-preference effect (i.e., the effect in which a linguistic expression that is not well known is subject to greater learning and, therefore, will elicit a greater likelihood of alignment). Further, research has shown that the degree to which young children align is predictive of their long-term linguistic abilities (Fusaroli et al., 2023a), supporting alignment as a means for implicit language development.

Audience Design Mechanism: Alignment to promote mutual understanding

While the previously discussed models argue that alignment is automatic and non-goal-directed, other explanatory accounts conceptualize alignment as a goal-directed (albeit potentially subconscious) behavior. One such account, commonly referred to as Audience Design, is largely based on ideas brought forward by Clark and colleagues (e.g., Brennan & Clark, 1996; Clark & Schaefer, 1989; Clark & Wilkes-Gibbs, 1986). This theory is rooted in the assumption that conversational goals are highly transactional; the purpose of conversation is for participants to give and receive accurate information to accomplish a specific task or objective. Accordingly, for conversation to be successful, interlocutors must reach a mutual understanding of one another. Individuals, therefore, align their behaviors as a means of establishing conceptual pacts or shared conceptualizations of the information being communicated. For example, an interlocutor may use the same term as their partner because they have inferred that their partner understands the conceptualization that this term conveys; adopting this term therefore maximizes the likelihood of mutual understanding. Here, we note that while the communicative benefits derived from alignment via the Audience Design mechanism are similar to those postulated by the Interactive Alignment Model (i.e., shared understanding of the situation under discussion), these two accounts differ in the sense that Audience Design is considered to be a non-automatic, goal-directed mechanism. Accordingly, the Audience Design account suggests alignment to be driven by an interlocutor’s ability to take the perspective of their partner and a goal to achieve common ground and joint meaning-making.

There are several studies that support this theory of alignment. For example, research has shown that interlocutors are more likely to align in situations where information exchange is more challenging, such as when conversing with a partner that they believe is communicatively less capable (e.g., a child, non-native, or non-human partner; Branigan et al., 2011; Cai et al., 2021; Ivanova et al., 2021; Suffill et al., 2021). Further, there are several studies showing that when interlocutors align, they exhibit greater levels of task success (i.e., solving the problem at hand), suggesting a higher degree of mutual understanding or shared common ground (e.g., Borrie et al., 2019; Reitter & Moore, 2014; Wynn et al., 2023).

Pedagogical Mechanism: Alignment to foster language development

While alignment may arise through an automatic non-goal-directed implicit learning mechanism, it may also arise through an explicit goal-directed mechanism that has the aim to promote language development. In this case, alignment does not occur because of an automatic priming mechanism. Rather, adopting the same linguistic representations is deliberate and strategic. For instance, specific scaffolding techniques often employed by caregivers such as expansions (i.e., repeating an utterance but with added linguistic detail) and recasts (i.e., repeating an utterance but with more accurate language) represent ways of (partially) aligning to a child’s utterance to provide more expansive or accurate linguistic representations (Cleave et al., 2015; Nelson et al., 1996; Taumoepeau, 2016). Instructors use similar techniques when teaching new languages to older children and adults (Brown, 2014; Rassaei, 2022). In addition to its use by those teaching or facilitating language, alignment is also used (explicitly) by those learning language. Older children and adults learning a new language will often adopt the linguistic representations of their conversational partner to practice a new target structure, ensure accuracy, or sound more like a native speaker (Costa et al., 2008; Coumel et al., 2022; Kim & Michel, 2023).

There is much empirical data that highlights alignment as an explicit pedagogical mechanism. For instance, research has found that caregivers often show high levels of alignment when speaking to young children and that this decreases over time as the children’s language skills become more robust (Misiek et al., 2020; Yurovsky et al., 2016). Further, this alignment predicts the child’s communication skills, indicating its role in language development (Denby and Yurovsky, 2019; Fusaroli et al., 2023b). Additionally, research on non-native language learners has shown that the use of aligned language structures is often a conscious and strategic decision (Coumel et al., 2022; Michel & O’Rourke, 2019), and that the degree of alignment used by a person is associated with their linguistic ability (Lewandowski and Jilka, 2019).

Social Identity Mechanism: Alignment to gain social approval

While some theoretical accounts such as the Audience Design account focus on conversation as transactional, other theories are built on the assumption of conversation as relational; the purpose of conversation is to reflect and affect social relationships with others (Soliz et al., 2021). One such theory is the Communication Accommodation Theory (CAT; Giles et al., 1987; Giles & Ogay, 2007). At the heart of this theory is the idea that humans possess an innate need to establish and maintain a social identity. Consequently, individuals align as a means of seeking affiliation and social approval. Because individuals are more attracted to people who are similar to them (Byrne, 1971), the CAT proposes that interlocutors align with their partner as a means of decreasing social distance, fostering liking, and communicating a desire for affiliation and social integration.

Again, there are many studies that suggest this is often the case. Research has shown that when people align, they are viewed as more likeable and friendly by their conversational partner (Schweitzer et al., 2017; Ireland et al., 2011). Further, individuals with a higher need for social approval (Natale, 1975; Aguilar et al., 2016) and better social skills (Shen and Wang, 2023) are more likely to align to their partner. Similarly, people are more likely to align after being ostracized (Hopkins & Branigan, 2020), suggesting that alignment is used to foster affiliation and social integration. Additionally, individuals are more likely to align to individuals they like (Hwang & Chun, 2018; Balcetis & Dale, 2005), people in a position of higher power (Gregory & Webster, 1996), and “in-group members” (Watson-Jones et al., 2016; Lakin et al., 2008).

Beat Induction Mechanism: Alignment to promote conversational flow

The idea that alignment stems from a beat induction mechanism is largely grounded in the idea that human beings possess an innate compulsion to move to rhythmic signals³ and derive pleasure from doing so (e.g., Todd et al., 2002; Todd & Lee, 2015; Honing; 2012; Janata et al., 2012). Researchers have postulated that this compulsion, which leads to many rhythmically coordinated social activities such as choral music and dancing, also drives alignment to the rhythmic signals in the speech and body movements of our conversational partner (e.g., Gill, 2012; Phillips-Silver et al., 2010; Phillips-Silver & Keller, 2012). By moving in time and space with our conversational partner, we engage in a sort of conversational dance. That is, alignment allows the conversation to flow. It helps bind turn-taking dynamics, creating a more rhythmic and cohesive conversational experience (Wilson & Wilson, 2005; Local 2007; although see Benus, 2009, for evidence against these postulations). In essence, by fostering conversational flow, alignment makes a conversation feel good in a way that goes beyond a shared comprehension or a signal of social approval. This drive to move in time and space with others is postulated to have evolutionary roots, serving as a social glue that builds a sense of cohesion, connection, and belonging between interlocutors (Lakin et al., 2003; Phillips-Silver et al., 2010; Gill, 2012; Shockley et al., 2003).

The idea of alignment as a facilitator of conversational flow is supported by research showing that both third-party outside listeners (Wynn et al., 2022; 2023) and interlocutors within a conversation (Chartrand & Bargh, 1999; Stel & Vonk, 2010) rate conversations with higher levels of alignment as being smoother and flowing better than conversations with low levels of alignment. Objective data supports this notion as well. When interlocutors align, there are more cohesive turn exchanges, fewer interruptions, and shorter inter-turn latencies (Nenkova et al., 2008; Levitan et al., 2012). Further, research shows when interlocutors are aligned, they are more empathetic and cooperative, pointing to the cohesion and connection established through conversational flow (Manson et al., 2013; Taylor & Thomas, 2008). This prosocial behavior has been shown to extend to individuals outside the conversation, suggesting alignment as a “social glue” that strengthens the bonds of individuals within a group (van Baaren et al., 2004; Stel & Harinck, 2010).

Non-interdependent Factors: Similarity not driven by alignment mechanisms

This framework is focused on mechanisms that drive interdependent behavior. However, when considering these mechanisms, it is also important to consider and account for the fact that similar behaviors may occur as the result of non-interdependent factors. Thus, while not included as a mechanism within our framework, we provide a discussion regarding these factors. All conversations contain some level of random similarity. For example, common words, gestures, and phonetic patterns are likely to be repeated simply because they are frequently used by both interlocutors. Additionally, similarities in communicative behaviors may reflect similarities in the general communication styles of interlocutors or may be a result of the conversational task at hand. For instance, children may use more of the same words when speaking with each other than adults simply because their lexical choices are constrained by a more limited vocabulary. Similarly, people from the same culture may appear to be more aligned than people from different cultures simply because specific communicative gestures are more commonly used within their culture. If interlocutors are trying to complete a task quickly, both may speed up their speech rate; if they are communicating in a noisy environment, both may raise their speaking volume. However, these similarities may relate more to the constraints or affordances of the task or environment as opposed to an interdependent relationship in which one interlocutor’s behavior directly causes changes in their partner’s behavior. These examples all denote differences that are not interdependent and, therefore, fall outside of the scope of our framework. However, when considering the effects of the alignment mechanisms described above, it is important to recognize that such non-interdependent factors exist and may also account for similar communicative behaviors.

Contextual Factors

These five types of alignment mechanism offer different lenses through which to consider alignment. However, this does not mean that these alignment mechanisms are mutually exclusive. Rather, we conjecture that any individual instance of aligned behavior is likely driven by multiple contributing mechanisms, but that across different conversations, the degree to which certain mechanisms are at play will vary based upon a number of contextual factors. These factors can be divided into two broad categories: factors relating to the conversational goal and those relating to the type of alignment being considered. We discuss each of these below.

Conversational Goals

Alignment is likely driven by different mechanisms depending on the conversational goal. Here, we use the term conversational goal to refer to an interlocutor’s personal motives for engaging in a conversation. Conversational goals may be conscious or subconscious. Further, interlocutors may pursue multiple goals simultaneously, and these goals may change across the course of the conversation. Conversations vary widely in terms of these goals, but the literature often divides goals into two broad categories: transactional and relational (Yeomans et al., 2022; Clark et al., 2019, Brown & Yule, 1983; Cheepen, 1988). Transactional goals are focused on giving or receiving information, often in order to complete a specific task or objective. Relational goals are focused on affecting the connections or relationships of interlocutors with one another (e.g., establishing rapport, deepening a relationship). A conversation may have goals that are highly transactional and minimally relational (e.g., problem-solving) or minimally transactional and highly relational (e.g., reminiscing). Yet, conversations could also have goals that are both highly transactional and relational (e.g., advice seeking) or both minimally transactional and relational (e.g., filling time).

In line with the principles of the MFA, we advance that alignment of communicative behaviors will likely be driven by different mechanisms depending on the conversational goal (or lack thereof). For instance, it is plausible that alignment in conversations with both low transactional and relational goals may arise from goal-agnostic mechanisms such as automatic priming. Alternatively, the observed similarity of behaviors may be due to non-interdependent factors. In conversations with more salient conversational goals, we would expect some baseline level of similar behavior that occurs simply due to automatic priming or non-interdependent factors. However, alignment will also likely be driven by other goal-directed mechanisms. For instance, in transactional conversations in which accurate exchange of information is important for goal realization (e.g., collaborative problem-solving), alignment may be strongly driven by a need to establish mutual understanding. When goals are more relational, alignment may primarily occur as a means to express affiliation or gain social approval. Because smooth conversation dynamics play a role in both efficient information transfer and in fostering closer connections, alignment in conversations with both transactional and relational goals may also be motivated, at least in part, by a desire to promote conversational flow. Of course, there are a number of factors which may influence conversational goals and, therefore, must be accounted for when considering alignment mechanisms. These factors may be related to the environment, the speaker, or the listener⁴. We provide examples for each of these factors.

Environmental Characteristics

Goals are often shaped by environmental factors such as the setting, timing, and conversation channel (e.g., phone vs. instant messaging vs. face-to-face). In particular, the context of a conversation often defines an interlocutor’s goals. For instance, in a work setting focused on problem-solving tasks, conversation will naturally elicit goals that are more transactional in nature (i.e., exchanging information to solve a problem). Therefore, alignment in these types of conversations may occur primarily as a means of achieving mutual understanding. Contrastingly, conversations at a party or other social event generally elicit more relational goals (i.e., connecting and building rapport with others). Therefore, in these instances, alignment may be more strongly driven by a desire for social approval.

Speaker Characteristics

A speaker’s individual characteristics, such as their personality, age, gender, culture, and life circumstances likely play a central role in the prevalence of different types of conversational goals. For instance, an individual with high rejection sensitivity may feel a strong desire to achieve relational goals regardless of the context. Therefore, this individual may align to gain social approval more than a person with lower rejection sensitivity. A person who has recently moved to a new country may have conversational goals focused on learning the new country’s language and will accordingly use alignment as an explicit strategy to foster language acquisition, while a native resident of the country will not. Older adults seek more meaningful social connections than younger adults (Carstensen, 2021) and are better at using appropriate strategies to foster positive interpersonal relationships (Blanchard-Fields, 2007). Thus, older adults may be more likely to align to gain social approval and foster such connections.

Listener Characteristics

The characteristics (either real or perceived) of the listener may also impact the speaker’s conversational goals and subsequently their motivation to align. For instance, if a speaker is conversing with someone with hearing loss, their primary goal may be to ensure their partner is able to understand the information they are trying to convey. Accordingly, in this instance, alignment may be largely driven by a desire to reach mutual understanding. An adolescent may be much more motivated by a desire to fit in when engaged in a conversation with a peer than when engaged in a conversation with an adult. Accordingly, in peer conversations, alignment may be more strongly driven by a desire to gain social approval, whereas in conversations with an adult, alignment may be less goal-directed and may simply be a result of priming. Two close friends may feel secure in their relationship with one another and be unconcerned about gaining social approval from one another. However, they may still align in order to enter into a state of conversational flow and enjoy the resulting connection and cohesion with one another.

Alignment Type

Alignment does not represent one single type of behavior. Rather, this term is an umbrella term for many different types of communicative behaviors which are conceptualized and measured in different ways. Indeed, alignment type can be characterized by several different factors such as linguistic domain, form, and temporal dimension (see Wynn & Borrie, 2022; Rasenburg et al., 2020 for reviews). Just as alignment is impacted by different mechanisms depending on the conversational goal, different alignment mechanisms can, in turn, impact the type of alignment that is exhibited during a conversational interaction. While there are a number of factors that can be used to characterize different types of alignment, we discuss and give examples of how alignment mechanisms may influence three specific factors: linguistic domain, alignment form, and temporal dimension.

Linguistic Domain

Conversations inherently contain information from a number of different linguistic domains, including phonetic, prosodic, lexical, syntactic, and semantic domains. Depending on the conversational channel (i.e., face-to-face), conversations can include nonverbal communicative behaviors such as gestures and facial expressions as well. Although alignment research is often siloed, with different research studies focused primarily on a single linguistic domain, there is an abundance of evidence demonstrating alignment across all linguistic domains. Recent evidence has, however, suggested that the alignment behaviors that occur across different linguistic domains do not always parallel one another (e.g., Ostrand & Chodroff., 2021). Rather, alignment often manifests differently depending on the linguistic domain being examined. This suggests that alignment of different linguistic domains may generally be associated with distinct alignment mechanisms. For example, it is intuitive to consider that the communicative function of reusing each other’s words may be critical in reaching mutual understanding, ultimately allowing interlocutors to share the same conceptualization of the words they are adopting. Contrastingly, aligning one’s speech features (e.g., speech rate, pitch, intensity) to one’s partner may be less useful to increase mutual understanding but may be strongly induced by the urge to move rhythmically and create conversational flow.

A recent body of research has collectively demonstrated that young children align more frequently in their lexical and syntactic behaviors than in phonetic or prosodic behaviors (e.g., Fusaroli et al., 2023a; Chieng et al., 2024; Wynn et al., 2018, 2019, 2023). This may be because an error-based implicit learning mechanism is likely to generate higher prediction error (and so greater behavioral adaptations) in lexical and syntactic domains, and also because lexical and syntactic processing are particularly cognitively demanding (and so benefit more from the facilitation provided by priming). However, aligning phonetic or prosodic characteristics may be driven more by other factors (i.e., promoting conversational flow), that become more motivating later in development.

Even within a linguistic domain, alignment mechanisms may vary depending on the specific features being examined (e.g., specific acoustic features, types of lexical items, or syntactic structures). For instance, lexical alignment of content words likely plays a crucial role in building mutual understanding (e.g., Gonzales et al., 2009). However, it is difficult to imagine that using the same backchannels, filler words, or function words (i.e., words less important to the overall meaning of the message) as one’s partner would be used for this same purpose. Rather, in these instances, alignment may be used to gain social approval. Of course, this is not to say that the alignment of a given linguistic domain can only be tied to one alignment mechanism. Lexical alignment can help promote conversational flow and prosodic alignment may aid in establishing mutual understanding. However, in line with the ideas laid out in the MFA, the relative weights of these linguistic domains and mechanisms will vary.

Form

Alignment can be categorized into different forms, that are measured in distinct ways. For instance, alignment rate (i.e., the number of utterances in which any amount of alignment occurs) is often differentiated from alignment level (i.e., the degree of alignment in utterances in which alignment does occur; Dideriksen et al., 2022). Similarly, researchers often distinguish between proximity—similarity of interlocutors’ feature values (e.g., both speakers are using approximately the same speech rate) and synchrony—similarity in the movement of feature values (e.g., both speakers increase and decrease their speech rate in tandem even if the actual speech rate values differ; see Wynn et al., 2022; Levitan and Hirschberg, 2011). While proximity may be measured by calculating the absolute difference between two interlocutors’ feature values, synchrony may be measured by calculating the correlation between these two feature values (although note that other measurement approaches also exist for both proximity and synchrony).

As with linguistic domain, different alignment mechanisms may be more strongly tied to different forms of alignment. For instance, Dideriksen and colleagues (2022) showed that while the rate of lexical and syntactic alignment was higher in rapport-based conversations aimed at gaining social approval, the level of alignment was higher in task-based conversations aimed at achieving mutual understanding. They conjectured that this difference is driven by a need for more selective and precise alignment in task-based conversations in order to give and receive accurate information. As another example, interlocutors concerned with promoting conversational flow may feel the urge to move their speech rates in tandem with one another on a relative scale (i.e., synchrony) without the need for absolute similarity (i.e., proximity). Contrastingly, proximity may be particularly susceptible to influences of non-interdependent factors (e.g., both interlocutors speak loudly due to environmental demands that necessitate speaking at a high volume to be heard).

Temporal Dimension

Alignment is often measured across different temporal dimensions as well (see Wynn & Borrie, 2022). For instance, alignment may occur at a local timescale (e.g., similarity occurring between partner’s adjacent turns) or may be more global (e.g., similarity occurring somewhere within an entire conversation). Different alignment mechanisms may be more or less relevant when considering different temporal dimensions. For instance, Ferreira and Bock (2006) surmised that while automatic priming mechanisms may lead to local alignment before activation decay sets in, implicit learning mechanisms may lead to more global alignment, reflecting longer-term storage and retention of the target. In an additional example, we could consider two colleagues trying to complete a task together. The colleagues may align on a turn-by-turn basis to promote smoother turn-taking dynamics and thus, increase conversational flow. Contrastingly, the colleagues’ attempts to ensure mutual understanding may be evident in alignment at a global level. One interlocutor may adopt the same words or communicative gestures as their partner across the entire course of the conversation to ensure that their colleague understands them.

Extending the Framework

Here, we have presented a framework that offers a scaffold for conceptualizing how different alignment mechanisms may function concurrently, though to varying degrees, depending on conversational goal and the type of alignment under consideration. However, we note that the examples used to illustrate these ideas are not meant to be exhaustive and are, at this point, only theory. With future empirical work aimed at understanding the relationships between alignment mechanisms and contextual factors, this framework could ultimately be implemented as a quantitative computational model. We acknowledge, however, that this is not a simple task. The causal mechanisms driving alignment are difficult to operationalize, observe, and measure, and the multitude of interdependent factors that dynamically shape alignment makes research in this area exceptionally complex and nuanced. Extending and refining this framework will, therefore, require more than a single experiment. As is typical in model development, progress will come through a body of rigorous studies, carefully drawn inferences, and ongoing refinement. Ultimately, we believe that the most meaningful advances will arise from the collective efforts of researchers with a diverse expertise and perspectives. In the following paragraphs, we lay groundwork for this shared enterprise by offering some initial suggestions of research approaches that may be helpful in extending and refining this framework.

First, controlled studies with carefully designed hypothesis-driven manipulations of the experimental paradigm can help disambiguate the different types of causal mechanisms across different contexts. For instance, Schoot and colleagues (2019) conjectured that if (syntactic) alignment is goal-directed, we should expect higher levels of alignment when responding to another interlocutor relative to audio-recorded stimuli⁵ devoid of social context. Thus, by comparing alignment in these two situations, it is possible to ascertain the degree of alignment that happens through priming versus goal-directed mechanisms (e.g., gaining social approval). While their study looked at alignment in only one context, future studies could use a similar set-up to compare alignment across different contexts. For example, this design could be used to compare differences in how much alignment can be attributed to priming versus goal-directed mechanisms in children vs. adults or in neurotypical vs. neurodivergent individuals (see Hopkins et al., 2022). In a similar type of design, Dideriksen and colleagues (2022) found a higher level of lexical and syntactic alignment in task-based conversations than rapport-based conversations. Based on the assumption that interlocutors are focused more on reaching mutual understanding in task-based conversations and on gaining social approval in rapport-based conversations, we can infer that higher levels of lexical and syntactic alignment are more strongly associated with mechanisms aimed at achieving mutual understanding than social approval (although see discussion above regarding contrasting findings between rate and level). Using a similar approach, researchers could examine alignment across other levels (e.g., speech, gesture) to determine if similar patterns hold or if alignment in these modalities is driven more by other mechanisms.

Differences in alignment measures could also be investigated. In a variation of this type of design that allows for more experimental control, participants could be given the same task but given different goals (either explicitly or implicitly). For example, half of the participants could be told they would be evaluated on the degree to which they successfully completed the task, and the other half of participants could be told their conversation partner would evaluate their friendliness and likability following the conversation. Different types of alignment could then be compared across the two conditions to see if and how types of alignment vary in conversations with different goals.

A further research approach that can be employed is comparing real conversations to sham or surrogate conversations (i.e., conversations created by combining data from two interlocutors that did not actually converse with one another). Using sham conversations is a common way in which researchers evaluate the degree to which the results of their study can attributed to interdependent (vs. non-interdependent) behaviors (e.g., Borrie et al., 2019; Duran et al., 2019; Wynn et al., 2022). However, this could be taken a step farther by comparing real and sham conversations across different contexts (e.g., cross-cultural vs. same-culture conversations) to determine if and how much specific factors influence the amount of similar behavior happening for interdependent vs. non-interdependent reasons. Other methodologies such as comparing similarity before and after exposure to an interlocutor (e.g., Pardo et al., 2012) could be utilized for a similar purpose.

Another powerful tool that we could use to look at alignment mechanisms is latent class analysis. Latent class analysis is a statistical approach which identifies and characterizes unobservable profiles (i.e., patterns of responses) to a set of observable categorical indicators. In terms of alignment, different causal mechanisms could be characterized by observing a number of different contextual factors such as conversational context, interlocutor characteristics, and alignment modality. Thus, one profile (i.e., alignment to gain social approval) may be characterized by higher levels of alignment of prosodic features and may occur most in affiliative or social conversations and/or conversations among adolescents. Another profile (i.e., alignment to reach mutual understanding) may be characterized by high levels of alignment of linguistic features and may occur most in task-oriented conversations and/or conversations among adults. Yet another profile (i.e., aligning because of priming) may be characterized by high levels of alignment of linguistic features and may occur in both affiliative and task-oriented conversations among both adolescents and adults. Accordingly, the examination of multiple features in tandem allows the identification of different profiles, which can shed light on the factors influencing causal mechanisms.

Future experiments could also compare relationships between different types of alignment and specific conversational outcomes to determine the relative influence of these different alignment types in achieving alignment objectives. For instance, consider an experiment in which after a conversation, interlocutors rated the level of conversational cohesion they felt. Researchers could then compare the relationship between conversational ratings and measures of alignment for speech, linguistic, and communicative kinesthetic features to see which linguistic domains played the largest role in increasing conversational flow.

Besides testing the relationship between contextual factors and alignment mechanisms, this framework raises a number of other questions and ideas to consider. For instance, we conjecture that alignment is often driven by multiple mechanisms simultaneously. As one example, it may be that certain kinds of alignment, such as lexical or syntactic alignment, are always driven by some degree of priming. However, the degree of alignment that occurs beyond this may depend on the degree to which other mechanisms, such as a desire for mutual understanding, are at play. This idea naturally leads to questions regarding which mechanisms are most likely to act in tandem and the extent to which the effects of different mechanisms are independent versus interacting. Additionally, our discussion of individual contextual factors (e.g., environment, speaker, temporal dimension) focused on each of these factors in isolation. However, this framework raises questions about the ways in which these factors are interconnected. For instance, individuals with strong rejection sensitivity (speaker characteristic) may show higher levels of alignment in contexts that promote relational goals but low levels of alignment in contexts that promote transactional goals (environment characteristic). These same speakers may also show more alignment in linguistic domains in which alignment is more strongly driven by a need to gain social approval than domains where this is not the case.

Another question regards the degree to which underlying mechanisms change throughout an interaction as the goals of the conversation change. One might imagine a work-related situation where a conversation starts out with pleasantries aimed at building rapport and gaining social approval before moving into problem-solving discussions where the need for mutual understanding is high. In such instances, do different mechanisms drive alignment at different timepoints in the conversation? And does this impact the type of alignment occurring most prevalently in different timepoints of the conversation? It is also possible that the same surface patterns of aligned behavior may be driven by different underlying mechanisms in different contexts. For instance, if children and adults show the same patterns of aligned behavior, can we infer that these stem from the same mechanisms, or is it possible that different mechanisms are at play? In essence, these questions highlight this framework not as a conclusive model, but as a foundation for continued research and further exploration.

Framework Use

A careful and rigorous body of research will be necessary to refine, expand, and strengthen this framework. Such work will not only provide the empirical foundation needed for a quantitative computational model but will also allow the framework to guide the design of future studies. For example, if empirical research supports the idea that interlocutors align primarily to gain social approval in relational conversations, more robust insights about the social aspects of alignment may be best studied in rapport-based contexts. By contrast, if a study is grounded in alignment as a means of achieving mutual understanding, task-oriented conversations may provide the most informative setting.

Beyond decisions regarding experimental set-up, this framework can also help address a major challenge in alignment research: the wide range of measures currently used to capture alignment and the uncertainty around which are most appropriate. By clarifying the connection between mechanisms and alignment types, the framework can inform which measurement choices are most appropriate for a given study. For instance, if implicit learning mechanisms lead to long-term storage and retention of targets, global measures may be more suitable for research on alignment in second-language learning. Conversely, if a more fine-grained type of alignment proves necessary to invoke a rhythmic conversational flow, local measures may be more appropriate for studies grounded in alignment as a beat-induction mechanism.

Empirical evidence regarding differences in alignment mechanisms can also guide interpretations of existing findings, helping us draw more coherent inferences from the broader alignment literature. For example, research on the alignment patterns of autistic people has produced mixed results. Studies on linguistic aspects of alignment frequently report robust patterns of alignment in autistic individuals, with no differences between autistic and non-autistic individuals (e.g. Allen et al., 2011; Branigan et al., 2016; Maltman et al., in press). In contrast, studies examining kinesthetic and speech alignment have shown reduced alignment in autistic people relative to non-autistic peers (e.g., Wynn et al., 2018; Ward et al., 2024; Lehnert-LeHouillier et al., 2020; McNaughton et al., 2024). At a very general level, it is possible that these differences may be due to differences in the conversational goals of autistic and neurotypical people and how different types of alignment are best leveraged to achieve these goals. A clearer understanding of the interplay between alignment types, conversational goals, and the underlying mechanism would enable stronger inferences about these differences and how they shape autistic and non-autistic conversations. Similar concepts can be used to draw stronger conclusions about alignment in other contexts as well.

Conclusion

The purpose of this article was to introduce the Mechanistic Framework of Alignment (MFA), which unifies diverse accounts of alignment mechanisms into one a single conceptual model. Our aim is to encourage a more integrated and holistic approach that acknowledges the multiple pathways through which alignment might arise and to provide a foundation for developing a quantitative computational model as empirical evidence accumulates. The MFA highlights how each of mechanism provides a unique and useful lens, and how relative weightings of each kind of mechanism in any instance of aligned behavior interact with conversational goals and alignment types. We hope that the ideas offered here will stimulate future research aimed that deepens our understanding of the interplay among causal mechanisms and informs the design and interpretation of studies on this complex and multifaceted phenomenon.