The impact of alcohol on affiliative verbal behavior: A systematic review and meta-analysis

Abstract

Background:

Language is a fundamental human social behavior that is linked to many rewarding social experiences, such as social bonding. Potential effects of alcohol on affiliative language may therefore be an essential feature of alcohol reward and may elucidate pathways through which alcohol is linked to social facilitation. Examinations of alcohol’s impact on language content, however, are sparse. Accordingly, this investigation represents the first systematic review and meta-analysis of alcohol’s effects on affiliative language. We aimed to test the hypothesis that alcohol increases affiliative verbal approach behaviors and to discuss future directions.

Methods:

PsychInfo and Web of Science were systematically searched in March of 2023 according to our preregistered plan (OSF: https://doi.org/10.17605/OSF.IO/AY9R8). Eligible studies included social alcohol administration experiments in which affiliative verbal language was assessed. We present a random-effects meta-analysis examining the effect of alcohol compared to control on measures of affiliative verbal behavior.

Results:

Our search identified 16 distinct investigations (961 participants) examining the effect of alcohol on affiliative verbal behavior. Studies varied greatly in methods and measures. Meta-analytic results demonstrated that alcohol is modestly associated with increases in affiliative verbal behavior (Hedges’ g = 0.164, 95% CI [0.027, .301], p = 0.019). Study quality was rated based on an adapted version of the Quality Assessment Tool for Quantitative Studies, and did not significantly moderate alcohol’s effects.

Conclusions:

This study provides preliminary evidence that alcohol can increase affiliative verbal behaviors. This effect may be an important feature of alcohol reward. Given heterogeneity in study features, low study quality ratings, and limited reporting of effect size data, results simultaneously highlight the promise of this research area and the need for more work. Advances in language processing methodologies that will allow future work to systematically expand upon this finding are discussed.

Introduction

Alcohol has held a significant role in human social life since the dawn of history (Slingerland, 2021) and every documented society with the raw materials required for alcohol production has consumed it, often in social contexts (Dunbar & Hockings, 2019). Today, some 2.35 billion people drink alcohol worldwide (WHO, 2018) and alcohol continues to be used most frequently in social settings (Fairbairn & Sayette, 2014), often for its socially enhancing effects (Cooper et al., 1992). While many enjoy its effects, alcohol consumption is also linked to considerable harm. Alcohol use disorder (AUD) is one of the most common mental health disorders (Kranzler & Soyka, 2018) and intoxication is linked to violence, risky sexual behavior, and road injuries (WHO et al., 2018). Why has alcohol, a toxin that ought to be evolutionarily disadvantageous, been so pervasively used and culturally accepted? Why, despite evident harms, do humans drink?

Generations of lab scientists have sought to answer this question by identifying the reinforcing effects of alcohol in isolated laboratory settings well-suited to experimental control and causal inference. Notwithstanding a near century of effort, laboratory research has not demonstrated consistent rewarding impacts of alcohol on emotional experience (Goodwin & Sayette, 2022). Strikingly, despite the historical and cultural evidence suggesting that alcohol is a social drug, few laboratory alcohol studies have incorporated social contexts (Goodwin & Sayette, 2022). This oversight stands out when one considers that social experiences are powerful motivators. For example, collective emotions tend to be stronger and longer lasting than their individual counterparts (Goldenberg et al., 2020) and research indicates that both feelings of shared reality with others and social bonding are fundamental human needs (Baumeister & Leary, 1995; Echterhoff et al., 2009). Experiments conducted in isolated settings are likely to struggle with generalizability and to overlook many of alcohol’s most rewarding properties linked to social connection and bonding.

Recently, an emerging literature has begun to examine alcohol’s effects in social settings. In contrast to isolated laboratory studies, this research has found alcohol to enhance emotional (e.g., Bowdring & Sayette, 2021; Fairbairn et al., 2018) and social experience (e.g., Gurrieri et al., 2021; Kirkpatrick & De Wit, 2013; Sayette et al., 2012). Such socially enhancing effects may motivate drinking. Indeed, alcohol expectancy research suggests that individuals expect alcohol to be socially facilitating (e.g., make social interactions easier, increase the pleasure derived from social interactions, increase friendliness; Brown et al., 1980) and that such expectancies predict actual alcohol use (Smith et al., 1995). Work is now needed to understand how this occurs. In particular, there is an absence of systematic understanding of alcohol’s impact on language: the words we say to each other. Language is an essential human social behavior and consideration of alcohol’s effects on language may be an important lacuna in understanding drinking motivation.

Why Language?

Language is a unique form of communication that provides critical information about social interactions. While nonverbal behaviors (e.g., smiles) express current emotions, language conveys complexity and describes the present, past, future, and hypothetical. These faculties have led anthropologists to suggest that language developed to bond large hominid social groups (Dunbar, 1993). Eventually, language became so fundamental that it is believed to have played a more important role in the recent evolution of our species than genes (Pagel, 2017).

Given the essential role of language in human social and cultural evolution, it is unsurprising that language data have demonstrated great utility in psychological research. Verbal content reveals features of personality (Koutsoumpis et al., 2022), cognitive styles (Wang et al., 2016), and even gender differences in friendship (Aries & Johnson, 1983). Moreover, language is a pathway for social reward and bonding; humor, constructive problem solving, gossip, and sharing positive news, have been linked to increases in social functioning/relationship quality (Boydstun et al., 2023; Brondino et al., 2017; Hall, 2017; Kriegelewicz, 2006; Peters et al., 2018; Weaver & Bosson, 2011). Moreover, self-disclosure can increase feelings of interpersonal closeness (Aron et al., 1997) and liking (Collins & Miller, 1994), and even activate the brain’s reward systems (Tamir & Mitchell, 2012). In sum, language is a rich source of behavioral data that captures the complexity of social experience and may play a causal role in social reward and bonding— fundamental human needs. Studying alcohol’s impact on language could therefore illuminate the rewarding effects of alcohol and unveil mechanisms underlying such reward.

Despite its promise, there has been limited systematic investigation into the impact of alcohol on verbal communication. Some studies have examined content-free features of speech production to identify markers of intoxication (e.g., Hollien et al., 2001), or describe bonding processes (e.g., Sayette et al., 2012). Other work indicates that alcohol increases verbal aggression (e.g., Stappenbeck & Fromme, 2014). Informative though they are, these topics offer an incomplete picture of alcohol social enhancement and research examining alcohol’s impact on affiliative verbal content is required to fully unpack alcohol reward. Unfortunately, only a small literature has examined affiliative verbal content, and no review has organized this work.

How Might Alcohol Impact Affiliative Verbal Content?

Social interactions involve an inherent conflict between approach and avoidance (Gable, 2006; Murray et al., 2006). We are driven to approach positive social outcomes, such as acceptance and intimacy, and to avoid negative social outcomes, such as rejection or exclusion. These drives are powerfully reinforced: socially rewarding experiences can activate the pleasure centers of the brain (Gable & Berkman, 2008) and have been described as fundamental human needs (Baumeister & Leary, 1995), whereas social threats are associated with deleterious consequences and activate the same regions of the brain as physical pain (Baumeister & Leary, 1995; Gable & Berkman, 2008). This conflict between approach and avoidance drives social goals and behavior, and consequently, verbal communication. For instance, self-disclosure is thought to be influenced by the balance between perceptions of possible social rewards (e.g., intimacy, approval) and risks (e.g., rejection, betrayal) (Omarzu, 2000).

One’s immediate environment and cognitive state influences this dynamic between social approach and avoidance (Gable & Berkman, 2008). Alcohol’s acute effects may, therefore, skew this balance. Indeed, several influential theories converge on the idea that alcohol may increase social approach behaviors by diminishing inhibitions. For instance, alcohol may reduce concerns over rejection/social threat by decreasing self-awareness (Hull, 1981) or attenuating stress reactions to uncertain (Bradford et al., 2022; Hefner & Curtin, 2012) or complex (Sayette, 1993) stressors, such as social threat. As a result, alcohol may allow for less constrained behavior and expression (Hull, 1981). Further, Steele and Josephs’ (1990) suggest that alcohol may cause more extreme behavior when that behavior is under an inhibition conflict: both provoked by a strong immediate cue and inhibited by a strong but more complex/less accessible cue. Consider, as an example, running into a long-time crush at a bar. Here, the provoking cue of your crush’s presence may drive you to approach and express your excitement. However, there are many more distal cues which, with some consideration, may reign in this urge: your concern that this crush could reject you, or perhaps awareness of subtle nonverbal behaviors that suggest they are not interested in further interaction. In this situation, the provoking cue to approach and affiliate with another may be inhibited by equally strong cues that require additional processing/foresight to comprehend. According to Steele and Josephs’ model, in such social situations, alcohol disrupts the processing of these more complex/distal cues, making it more likely that one would behave in accord with their initial impulse: to approach and express their feelings.

Applicability to Typical Drinking Contexts

In sum, considerable research suggests that alcohol may increase social and verbal approach behaviors by decreasing inhibitions and perception of social risk. Importantly, however, alcohol is not predicted to universally fuel affiliation, rather, it is expected to broadly disinhibit behavior. The nature of this disinhibition is contextually bound. In situations where affiliation is a salient motive and inhibiting forces are uncertain/distal, alcohol should increase affiliative verbal approach behaviors. On the other hand, if instigating forces are not affiliative, alcohol’s disinhibiting effects may not be prosocial (e.g., Taylor et al., 1977). Indeed, meta-analyses reveal that alcohol increases aggressive/disinhibited behaviors (Bushman & Cooper, 1990; Ito et al., 1996; Steele & Southwick, 1985). Accordingly, the current review focuses on alcohol’s effects on affiliative verbal tendencies, with the goal of better understanding alcohol reward and motivation.

Alcohol is commonly consumed when social affiliation is a salient motive and inhibiting cues are uncertain. Social facilitation is the most strongly endorsed reason for drinking (Cooper et al., 1992) and most drinking contexts involve low-stakes social interactions where social approach goals are presumably present (Kairouz & Greenfield, 2007). Moreover, multiple domains of research suggest that, in such settings, inhibiting cues are complex, indicating that alcohol should disrupt their processing. Steele and Southwick (1985) posit that recognition of inhibiting forces, generally, requires complex processing (e.g., conceptualization of consequences/awareness of standards of conduct). Specifically, social avoidance cues are often subtle and ambiguous (e.g., pause durations, eye contact; Kerr & Levine, 2008; Wesselmann et al., 2016), especially in low-stakes social interactions (Banki, 2012). Moreover, alcohol disrupts functioning in brain areas responsible for the detection and appraisal of social threat (Gorka et al., 2018). Accordingly, in many typical drinking contexts, alcohol may reduce avoidance motivation and consequently increase verbal behaviors aimed at social reward/affiliation.

The Present Study

Although generations of studies have struggled to observe consistent alcohol reward, recent social paradigms have revealed enhancing effects of alcohol on social and emotional experience. We aim to build on these findings by reviewing a disparate literature assessing the effects of alcohol on affiliative verbal behavior in social settings. This work is grounded in research suggesting that in many typical drinking settings, alcohol may increase affiliative/approach verbal behaviors by decreasing inhibitions. These effects may offer a critical pathway for alcohol reward through their effects on social facilitation and bonding. Specifically, we aim to: (1) systematically examine the effect of alcohol on affiliative verbal behavior, hypothesizing that alcohol will increase verbal approach behaviors; and (2) draw upon a burgeoning speech processing literature to inform future research directions.

Method

We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA; Moher et al., 2009) guidelines, and preregistered our review through the Open Science Foundation (https://doi.org/10.17605/OSF.IO/AY9R8).

Study Retrieval

APA PsycInfo and Web of Science Core Collection were searched in March of 2023 using a combination of terms relating to alcohol AND social interaction AND verbal content (see Supporting Information, Section 1). Search results were uploaded to Covidence, an internet-based software for the management of systematic reviews, and duplicates were automatically removed. References from two recent reviews (i.e., Fairbairn, 2017; Fairbairn & Sayette, 2014) were also uploaded. Studies were then screened for eligibility. Finally, references and reverse citations found in the Web of Science Core Collection for each eligible study were also screened. Search and eligibility determinations were completed by the first author.

Study Eligibility Criteria

Eligible studies included social alcohol administration experiments (in which alcohol condition is explicitly manipulated) that assessed affiliative verbal content. The following eligibility criteria were preregistered before the search: 1. Participant population: there were no limitations by gender, sex, race, ethnicity, or alcohol use status. In social paradigms including couples, samples/subsamples that were recruited explicitly for histories of violence toward their partner were excluded because verbal approach behaviors in these intense conflict interactions are less obviously linked to prosocial/affiliative behavior (e.g., Leonard & Roberts, 1998). 2. Manipulation: alcohol had to be experimentally administered at a level that would produce intoxicating effects (i.e., ≥ 0.04 % BAC). 3. Comparator: there was a comparison condition where placebo or no-alcohol is administered. 4. Setting: subjects were exposed to a social situation (i.e., in a room with at least one other participant/confederate and allowed to interact in some manner). 5. Outcomes: subjects’ affiliative verbal content was assessed during the social exchange in which the alcohol condition was manipulated. 6. Other study features: the study was published as of March 2023 and was printed in English.

Study Outcomes

Outcomes of interest include measures of affiliative verbal content that reflected a tendency towards social approach. Because this is a small and unsystematic literature, identification of eligible outcomes required study-specific decisions. To limit bias, the following guidelines were preregistered prior to the search: eligible measures must represent a social approach tendency (e.g., putting yourself out there more/acting less inhibited, self-disclosing more, acting more friendly/positive, engaging in more prosocial problem-solving behaviors) and be based largely on verbal content (e.g., could be approximated with information from a transcript and didn’t rely primarily on content-free/non-verbal features). Measures based on first-person pronouns/self-references were excluded, as these have been used as measures of self-awareness (Hull et al., 1983). To further limit bias, all social alcohol experiments identified in the search that included a measure of verbal content that was not explicitly a measure of anxiety, self-awareness, or aggression were reviewed by an independent researcher blind to study outcomes. Only measures that this researcher coded as affiliative were included (see Supporting Information, Section 2, for details).

Data Collection and Management

Alcohol effect sizes for each eligible measure were extracted by the first author. For those studies without sufficient data, authors were contacted via email when possible. Authors were given a month to respond and were then prompted. Studies for which authors were unable to provide information about, at a minimum, the presence/absence of an effect were excluded. For this initial test, we were most interested in examining the effect of alcohol irrespective of the specific measure of verbal behavior or participant characteristics. Thus, several preregistered steps were taken to compute a single effect size per study. First, when studies reported data for separate subsamples of participants, results were collapsed across groups assuming independence. Second, when multiple eligible measures were reported, effect sizes for each outcome were averaged. For studies that included multiple comparison conditions (e.g., placebo and no-alcohol control), results were combined across comparisons (per the Cochrane Handbook, Higgins et al., 2011). Similarly, for studies that included multiple timepoints, we combined across timepointsⁱ (with the exception of only examining interactions between adult participants and excluding timepoints where parents interacted with children). See Supporting Information, Section 3 for more details on data extraction. In addition to effect size information, the first author collected data on the study task, alcohol comparison condition, BAC range, and participant gender for each study. Study quality was also coded by the first author using a modified version of the Quality Assessment Tool for Quantitative Studies (QATQS) (de Groot et al., 2019) reported in Kumar et al. (2022) (see Supporting Information, Section 4).

Data Analyses

Quantitative synthesis was conducted using Comprehensive Meta-Analysis Version 4 (Borenstein et al., 2022). For each identified study, a standardized mean difference (Hedge’s g) was computed for the comparison between groups that received alcohol and those that did not (Hedges, 1981). Where necessary, effect sizes were computed to account for issues of interdependent data stemming from nested study designs (see Supporting Information, Section 5). A random-effects meta-analysis was then conducted to determine the overall effect of alcohol. We employed a random-effects model because it can account for both within and between-study variance and, as a result, is a recommended approach in meta-analyses that involve data from studies that have been performed independently (Borenstein, 2009; Dettori et al., 2022). In addition to reporting the average effect across studies, in meta-analyses it is important to assess the extent to which the “true effect size” varies across studies. Here, the Q and I² statistics were used to assess such heterogeneity. To test for study quality moderation, we conducted a subgroup analysis to determine whether the effect of alcohol differed by study quality, where study quality was a categorical variable with three levels (weak, moderate, and strong) and a common among-study variance component was assumed across subgroups. To examine the robustness of our analyses against potential outliers, studies with effect sizes that were more than three standard deviations from the overall effect were removed in a secondary analysis (Lipsey & Wilson, 2001). Further sensitivity analyses involved a series of ‘leave-one-out meta-analyses’, in which we recalculated the results of our meta-analysis 16 times on a subset of 15 studies, each time excluding one of the 16 studies. This approach allowed us to identify any studies that were particularly influential on our findings. Finally, publication bias was evaluated using common metrics (i.e., funnel plot, Begg’s rank correlation test, Egger’s test, and trim and fill analyses).

Results

A total of 5,468 (722 duplicates) studies were screened and 24 were identified as potentially eligible. An independent researcher coded study outcomes for eligibility in each of these examinations and two studies were excluded on the basis that all reported outcomes were not measures of affiliative verbal behavior (Monahan & Samp, 2007; Smith et al., 1975 a). An additional study was excluded because the authors were unable to provide necessary information (Abbey et al., 2000). Thus, the final search resulted in 21 eligible studies. Six of these studies reported on data drawn from a single large database (Jacob et al., 1989). To avoid redundancies, data from two of these studies (Haber & Jacob, 1997; Jacob et al., 2001) were combined as independent subgroups to produce a single effect estimate representative of this database (see Supporting Information, Section 6). Two publications reported separate analyses of the same sample; effect sizes from these studies were aggregated and treated as a single study (Frankenstein, et al., 1985 a, b). Finally, one article reported on two separate samples that completed different experimental procedures; each was included as a separate study in the review (Caudill et al., 1987), from here on out these are referred to as the Caudill et al. (1987 Study 1 and Study 3) studies (Study 2 was not included in the meta-analysis because there was no alcohol administered). Thus, the present review focused on results from 16 separate studies of alcohol’s impact on affiliative verbal behavior (for a breakdown of search steps, see Figure 1, for specific information on these studies see Table 1).

Figure 1. Flow Diagram of Study Search and Selection.

Note. This figure was derived from the Covidence software for systematic reviews.

Table 1.

Summary of Studies in Review

Study		N	Sex	Interaction	Cntrl	BAC Range (before to after)	Outcome	Partner Intoxication	Effect	Quality
Abbey et al. (2005)		51a	M	speak to a responsive female confederate about neutral topics provided by researchers	P/C	0.072-0.076%	-dating availability index	partner is a sober confederate	Alcohol significantly increased rates of hints about dating. b	strong
Babor et al. (1983)		8 dyads	M	develop a humorous story about an image together	C	0.108-0.083%	-disinhibited humor content: -sexual themes; -aggressive themes	dyads had matched conditions	Alcohol did not impact frequency of sexual content. Alcohol significantly increased aggressive themes at various points on the BAC curve.	weak
Caudill et al. (1987 Study 1)		48	M	make favorable impression on minimally responsive female confederate	P	low group-0.04-0.03% high group-0.09-0.08%	-per-statement depth of self-disclosure -global depth of self-disclosure	partner is a sober confederate	The low dose of alcohol significantly increased per-statement depth of self-disclosure. There was no effect in the high dose condition and no effects for the global measure.	weak
Caudill et al. (1987 Study 3)		32	F	make favorable impression on minimally responsive male confederate	P/C	0.04-0.036%	-per-statement depth of self-disclosure -global depth of self-disclosure	partner is a sober confederate	Alcohol increased the per-statement depth of self-disclosure compared to the placebo condition but decreased it compared to the control. There were no clear effects for the global measure.	weak
de Boer et al. (1994)		72	F	self-disclosure interview with minimally responsive confederate	P /Cc	0.047%	-amount of self-disclosure -depth of self-disclosure	partner is a sober confederate	No significant effects.	weak
Frankenstein, Hay, and Nathan (1985) and Frankenstein, Nathan, et al. (1985) d		8a	M/F	conflict resolution discussion with spouse	C	0.099-0.078%	-positive verbal behavior -problem solving -problem describing statements -questions	partner is sober	Alcohol significantly increased question asking. No significant effects on positive verbal behavior (though it did increase significantly across the marital dyad) or problem solving/describing.	strong
Jacob et al. (1981)		8a	M	conflict resolution discussion with spouse	C	0.08%	-positive verbal behavior -problem solving -agreements	partner given the opportunity to drink but did not reach threshold BAC	No significant effects.	moderate
Jacob et al. (1989) databased	Haber and Jacob (1997)	31a	F	conflict resolution discussion with spouse	C	0.07%	-positivity -problem solving	partner given the opportunity to drink but did not reach threshold BAC in most cases e	No significant effects.	moderate
	Jacob et al. (2001)	90a	M			0.063%			No significant effects.
Leonard and Roberts (1998)		75a	M	conflict resolution discussion with spouse	P/C	0.102-0.087%	-positivity -problem solving	partner is sober	Alcohol significantly increased problem solving. No effect on positivity. b	strong
Lindfors and Lindman (1987)		16 in groups of 4	M/F	discuss a film in groups of four	C	0.105%	-codes reflecting passive affiliation -codes reflecting active affiliation	condition is matched across the group	Alcohol significantly increased some but not all measures of active and passive affiliation.	weak
Monahan and Lannutti (2000)		50	F	get-to-know-you interaction + flirting cue with responsive male confederate	C	0.083%	-self-disclosure -positive acknowledgement of flirting	partner is a sober confederate	Alcohol significantly increased self-disclosure after the flirting clue in women with low social self-esteem. No other significant effects.	weak
Murdoch and Pihl (1985)		44 in groups of 4	M	cartoon captioning task in groups of six (four participants and two confederates)	P/Cc	0.105-0.110%	-positive comments	mixed within groups: one participant in each arm of balanced placebo and two sober confederates	Alcohol significantly increased positive comments.	weak
Rohrberg and Sousa-Poza (1976)		16 dyads	M	alternate speaking freely on five disclosure topics with partner	P	0.058%	-depth of self-disclosure	dyads had matched conditions	Alcohol significantly increased the depth of self-disclosure.	weak
Schippers et al. (1997)		48a	M/F	self-disclosure interview with minimally responsive confederate	P/C	0.05%	-depth of self-disclosure	partner is a sober confederate	Alcohol had no significant effects.	weak
Smith et al. (1975 b)		18 dyads	M/F	talk about anything important to you	P	0.055-0.065%	-acknowledgment	dyads had matched conditions	Alcohol tended to decrease acknowledgement.	strong
Testa et al. (2014)		152 dyads	M/F	conflict resolution discussion with spouse	C	0.071-0.076%	-positivity -problem solving	half of couples had concordant conditions, other half were discordant	Alcohol significantly increased rates of positivity. Nonsignificant increase in problem solving.	strong

Note.

^a.Represents the subsample of participants who were eligible for this meta-analysis (e.g., based on intoxication level, balanced placebo arm, or history of violence).

^b.Results were reported as a main effect that did not vary across subgroups of eligible and ineligible participants, and were imputed in the meta-analysis just for the eligible subgroup.

^c.Balanced placebo design where we were unable to separate expectancy conditions, this comparison therefore represents the effect of alcohol irrelevant of expectancy (e.g., TNA-A + TA-A vs. TNA-NA, TA-NA).

^d.Combines multiple publications on the same dataset.

^e.15 of the participants examined from the Haber and Jacob (1997) study were female alcoholics whose partner was also a male alcoholic. For these 15 individuals, their partner likely reached threshold BAC. We only included data from the females in the Haber and Jacob (1997) effect size estimate because male participants in this databased were included in the Jacob et al. (2001) effect size estimate.

A Picture of The Field

Studies were heterogenous in methods and measures. There were a range of measures of affiliative verbal behavior, many of which appeared in only one investigation. Most outcomes either reflected disinhibited expression (e.g., self-disclosure) or responsiveness to an interaction partner (e.g., acknowledgment), though specific measures within each of these categories varied widely. Other outcomes were summary codes of positivity and constructive problem solving, assessed with versions of the Marital Interaction Coding System (MICS) (e.g., Heyman et al., 1995). Table 2 includes more details on each category and describes the general pattern of alcohol’s effects within each. Studies also differed in the kinds of interaction tasks administered. Experimental tasks fell into one of four categories: free discussions, collaborative work, one-sided interactions with confederates, and conflict resolution paradigms. These task categories, along with the pattern of results in each, are detailed in Table 3.

Table 2.

Summary of Outcome Measures in Review

Task Type	Description	# Studies	Description of Results	Subjective Conclusion
Disinhibited Expression	Measures that reflect a tendency to be less constrained in speech and more willing to reveal features of the self.	9	Self-disclosure was the most common measure of disinhibited expression: six studies included measures of either depth or amount of self-disclosure. Three of these investigations evidenced significant effects of alcohol on at least one comparison (Caudill et al., 1987 Study 1; Monahan & Lannutti, 2000; Rohrberg & Sousa-Poza, 1976). Two studies involving minimally responsive opposite-gender confederates reported no significant effects (de Boer et al., 1994b; Schippers et al., 1997). Caudill et al. (1987 Study 3), reported an interaction where alcohol increased self-disclosure compared to placebo but decreased it relative to control. Importantly, the study with the largest overall effect of alcohol on self-disclosure was the only study that involved a free discussion between two same-gender participants, as opposed to opposite-gender interactions with a confederate (Rohrberg & Sousa-Poza, 1976). Several additional outcome measures also reflected disinhibited expression including providing hints about wanting to date an interaction partner (Abbey et al., 2005), taboo humor content (Babor et al., 1983), and a summary code of active affiliative behavior (e.g., sharing opinions, expressing feelings, Lindfors & Lindman, 1987). Alcohol significantly increased at least one measure of each of these behaviors, and notably, none of these behaviors were assessed during one-sided interactions with a confederate.	Preliminary evidence for a possible small effect of alcohol on measures of disinhibited expression, particularly in interactions with responsive partners.
Responsiveness to an Interaction Partner	Measures that reflect attention and responsiveness to an interaction partner.	5	Two studies included measures of acknowledgment to an interaction partner’s preceding communication, either any communication (Smith et al., 1975b) or specifically to flirting attempts (Monahan & Lannutti, 2000). Neither examination reported increases in acknowledgment under alcohol and results from Smith et al. (1975b) suggest that alcohol tends to decrease acknowledgment. Other measures reflecting possible responsiveness to an interaction partner included question-asking (Frankenstein, Nathan et al., 1985), agreement (Jacob et al., 1981), and passive affiliation (e.g., giving suggestions, repeating, clarifying, Lindfors & Lindman, 1987). Alcohol significantly increased question-asking and increased one measure of passive affiliation but did not impact agreement.	Mixed results.
MICS Positivity	Specific to couple’s conflict paradigms. A summary code for positive behaviors (e.g., agreement, reflection, humor, self-disclosure).	5	Only one of five investigations reported a significant positive effect of alcohol on positivity (Testa et al., 2014). Importantly, positivity codes were generally inclusive of a number of relationships enhancing behaviors, including providing approval, humor, reflections, and self-disclosure. It is not possible to tease apart alcohol’s effects on any of these specific behaviors at this time.	Little support for an effect of alcohol on positivity in couple’s conflict paradigms.
MICS Problem Solving	Specific to couple’s conflict paradigms. A summary code of constructive problem-solving behaviors (e.g., describing the problem, presenting solutions, asking questions).	5	The five conflict-resolution paradigms included in the present review all assessed problem solving behavior. Though only one study reported a significant increase in problem solving under alcohol (Leonard & Roberts, 1998), three others reported effects in the positive direction (Frankenstein, Hay, & Nathan, 1985; Jacob et al., 1989 database; Testa et al., 2014). The final examination did not report an effect direction (Jacob et al., 1981). Again, because problem solving was a summary code, it is hard to know which elements of this behavior drive results.	Inconclusive but alcohol may exhibit small effects on constructive problem solving.

Table 3.

Summary of Interaction Tasks in Review

Task Type	Description	# Studies	Description of Results	Subjective Conclusion
Free Discussions	Participants engaged in free discussion periods in which they had back-and-forth conversation with one or more responsive partners (i.e., an unidentified confederate or fellow participant).	5	Four investigations evidenced significant effects of alcohol on at least one eligible measure. Specifically, alcohol increased measures of passive (e.g., giving suggestions, direction) and active (giving opinions, expressing feelings) affiliation in mixed-gender groups discussing a film (Lindfors & Lindman, 1987), self-disclosure in male dyads discussing personal topics (Rohrberg & Sousa-Poza, 1976), as well as in women with low social self-esteem following flirting cues from a male confederate (Monahan & Lannutti, 2000).a Alcohol also significantly increased hints about dating availability in men speaking with a female confederate (Abbey et al., 2005).b The final study in this category reported a trend toward alcohol decreasing measures of acknowledgment in male-female partners (Smith et al., 1975b). Notably, this was unique in that the verbal behavior assessed required attentional control and focus.	Inconclusive but alcohol’s effects may be observable in free discussion tasks.
Collaborative Humor Tasks	Participants worked together on creative humor tasks (e.g., deciding on captions for a cartoon).	2	Both investigations revealed significant positive effects of alcohol on at least one measure and no effects in the negative direction. Specifically, alcohol increased rates of positive comments (Murdoch & Pihl, 1985) and unrestrained humor (Babor et al., 1983).c	Inclusive but alcohol’s effects may be observable in collaborative humor tasks.
One-Sided Interactions with Confederates	Participants spoke to a minimally responsive confederate with the aim of either making a positive impression or answering questions about themselves.	4	One study reported that a low, but not high, dose of alcohol significantly increased the depth of self-disclosure per subject statement (Caudill et al., 1987 Study 1). Another investigation showed a significant effect of alcohol on per-statement self-disclosure, but only when compared to a placebo group; the effect was significant and negative compared to a control group (Caudill et al., 1987 Study 3). The final two investigations reported no significant effects (de Boer et al., 1994; Schippers et al., 1997).	Little support for an effect of alcohol in one-sided confederate interactions.
Conflict Resolution Paradigms	Couples generated lists of current disagreements and then engaged in a discussion with the aim of resolving the problem.	5	Mixed findings. Frankenstein, Hay, and Nathan (1985) and Frankenstein, Nathan, et al. (1985) found that alcohol significantly increased question-asking and had non-significant positive effects on problem-solving and positivity. Leonard and Roberts (1998) also observed a significant positive effect of alcohol on problem-solving d but not positivity. Testa et al. (2014) reported significant positive effects of alcohol on positivity and non-significant effects in the positive direction on problem-solving. The final two studies reported no significant effects (Jacob et al., 1981, Jacob et al., 1989 database). Notably, these two studies were unique in that the alcohol condition was manipulated within-subjects by providing alcoholic beverages and allowing participants to drink as much or as little as they wanted. Although subjects reached an average blood alcohol concentration (BAC) in line with other studies in this database (~0.07%) there was likely much more variability compared to studies with a fixed dose.	Mixed results.

Note.

^a.Results in the Monahan & Lannutti (2000) study were qualified by an interaction, whereby alcohol only increased self-disclosure in women with low social self-esteem following a flirting cue from a male confederate. This may still be viewed as supportive of the hypothesis presented here as it is logical that alcohol’s disinhibiting effects would be most evident for subjects who are more inhibited (e.g., low self-esteem) and when a social approach motive is salient (a flirting cue).

^b.Importantly, this was only reported as a main effect that did not differ across groups of men who either were or were not previous perpetrators of sexual assault (Abbey et al., 2005).

^c.Alcohol increased rates of aggressive humor content. This was coded as the least clearly affiliative behavior in the current review, though this measure was included as it reflects disinhibited expression in humor and shared humor has been linked to affiliative outcomes (Curry & Dunbar, 2013).

^d.This was reported as a main effect that did not vary between groups of maritally aggressive and non-aggressive men.

Beyond this heterogeneity, this literature is marked by a lack of programmatic or current research. Most authors appear to have only conducted a single investigation on this topic, and all but two investigations were conducted over 20 years ago. Not surprisingly, nine of these studies did not include complete effect size data for every eligible outcome. Further, study quality ratings were generally low, largely due to possible issues of selection bias (e.g., unrepresentative samples), limited reporting on or control for possible confounding variables, and limited reporting of participant exclusions.

In sum, research examining alcohol’s effects on verbal content is heterogenous, scattered, and has not generated a corpus that lends itself to precise quantitative estimates of effect size. Meta-analyses can struggle to describe heterogeneous research with great variability in study designs, experimental interventions, or outcomes (Tarsilla, 2010). Consequently, we present a brief quantitative synthesis of results to provide an initial test of the overall effect of alcohol. Rather than providing a definitive effect size, our goal is to investigate whether a signal emerges of an effect of alcohol, despite the noise. We believe that such a finding would stimulate more systematic research moving forward.

Quantitative Synthesis

On the basis of reported results and correspondence with study authors, precise effect size data were available for 7 of 16 studies (43.75%). For all remaining studies, effect sizes for at least one outcome had to be imputed. Specifically, per our preregistered plan, when authors reported that an effect was not-significant but did not provide additional information required to compute an effect size, the effect was included in the metanalysis assuming zero difference between groups (i.e., d = 0.00). Similarly, when authors reported a significant effect without additional information, the smallest effect possible that would result in a significant result was estimated (i.e., based on sample size and a p-value of 0.05). Importantly, this is a recommended and conservative approach to managing incomplete study data that has been used in previous meta-analyses in the field (Bowdring & Sayette, 2018; Fairbairn & Sayette, 2014; Lipsey & Wilson, 2001). Finally, two studies reported aggregate results across subsamples of participants, some of whom were ineligible (e.g., half the participants met predefined exclusion criteria for this meta-analysis because of prior relationship aggression), but specified that the effect of alcohol did not differ between subgroups. For these two studies, effect sizes were imputed for the eligible participants based on the available data reported across the sample.

Quantitative Synthesis Overall Effect

A random-effects meta-analysis of all 16 studies revealed a small significant effect of alcohol on measures of affiliative verbal content (Hedges’ g = 0.164, 95% CI [ 0.027, .301], p = 0.019, see Figure 2) with minimal heterogeneity across studies (Q (15) = 18.187, p = 0.253, I² = 17.525, 95% CI [ 0, 54.034]). The effect of alcohol remained significant after removing outlier studies (n = 6) suggesting that these findings are robust to outliers (Hedges’ g = 0.130, 95% CI [0.01, 0.249], p = 0.033). ⁱⁱ Additionally, ‘leave-one-out analyses’ demonstrated the robustness of our effect; all 16 of the ‘leave-one-out analyses’ showed a significant effect, with Hedges’ g ranging from g = 0.128 to g = 0.199. A subgroup analysis further suggested that there were no significant differences in the effect of alcohol across, weak, moderate, and strong studies (p = 0.602). Importantly, however, this result is likely driven by low power to detect subgroup effects in this database and should not be interpreted as definitive evidence of a lack of a study quality effect.ⁱⁱⁱ A follow-up analysis examined the effect of alcohol only for outcomes with precise effect size data. The effect of alcohol remained significant (Hedges’ g = 0.228, 95% CI [0.024, 0.431], p = 0.028, see Figure 3), and heterogeneity again fell below the threshold for statistical significance, though it was moderate in magnitude (Q(10) = 18.091, p = 0.053, I² = 44.725, 95% CI [ 0, 72.593]). Together, these analyses suggest that alcohol has a small positive effect on affiliative verbal content and that this effect is relatively consistent across studies, though some study features contribute to variability. Given the small sample size, scattered reporting of effect size data, and varying methodological and measurement features, statistical comparison of subgroups could be misleading. Accordingly, with the exception of study quality (noted above), potential features contributing to heterogeneity were not examined quantitatively.

Figure 2. Overall Meta-Analysis of Alcohol’s Effect on Affiliative Verbal Behavior.

Note. Study weight is indicated by the relative size of central icon and study variability is indicated by length of associated line. Figure was generated with the Comprehensive Meta-Analysis software, version 4.

Figure 3. Meta-Analysis of Alcohol’s Effect on Affiliative Verbal Behavior: Complete Effect Size Data.

Note. Study weight is indicated by the relative size of central icon and study variability is indicated by length of associated line. This analysis includes only outcomes for which complete effect size data were available. Figure was generated with the Comprehensive Meta-Analysis software, version

Assessment of Bias.

Visual inspection of funnel plots (see Figure 4), Begg’s rank correlation test (Tau = 0.233, p = 0.207), and Egger’s test (t = 1.507, p = 0.154) showed no evidence of publication bias. Trim and fill analyses, however, resulted in a smaller point estimate for the effect of alcohol (Hedges’ g = 0.082, 95% CI [−0.079, 0.243]), suggesting that results may have been inflated by biased publication of significant effects.

Figure 4. Funnel Plot of Overall Meta-Analysis.

Note. This figure depicts a scatterplot of the treatment effect in each study against the standard error from that study. Each dot represents a study in the overall meta-analysis. Figure was generated with the Comprehensive Meta-Analysis software, version 4.

Narrative Review of Studies

Results of our quantitative synthesis are qualified by the quality of data available and should be viewed as preliminary. Importantly, though, a supplementary narrative review of results also suggests that alcohol may have a small positive effect on affiliative verbal behavior. Of the 16 studies reviewed, 10 reported significant positive effects of alcohol on at least one measure of affiliative verbal content, whereas only two reported significant negative effects on any measures. Combining across measures, 13 of the 16 studies showed effects in the positive direction, and only two studies showed effects in the negative direction.^iv A one-sample non-parametric sign test comparing the number of positive and negative effect sizes across studies indicated that the effect size of alcohol is greater than zero (p = 0.004).^v In sum, general trends across studies support the results of the overall meta-analysis, indicating that alcohol may have small positive impacts on measures of affiliative verbal behavior.

Discussion

Language is a core feature of human communication that underlies many rewarding social experiences. Accordingly, alcohol’s effects on affiliative language may be integral to alcohol’s socially rewarding properties. Examinations of language content, however, are scattered and have received limited attention in recent decades. This review aimed to organize this disparate set of studies and to test the hypothesis that alcohol increases affiliative verbal approach behaviors.

Our meta-analysis revealed a significant, albeit small, positive effect of alcohol, and results from a narrative review reinforced this finding. Both quantitative and qualitative synthesis supported the hypothesis that alcohol produces modest increases in affiliative verbalizations. Importantly, this signal was observed despite the noise introduced by a small number of studies, varied methods and outcomes, limited data reporting, and conservative imputation of missing effect sizes. The weak magnitude of this effect, although surprising in light of considerable theoretical and anecdotal evidence, may thus be partially explained by the substantial noise in available data.

This finding is of broad importance to the alcohol literature. Affiliative verbal behaviors have reliably been linked to rewarding personal and interpersonal experiences, including social bonding—a fundamental human need. Alcohol’s impact on affiliative verbal behavior may therefore reflect alcohol reward and serve as a mechanism that reinforces alcohol use. This is of etiological and clinical relevance. Etiologically, these results point to populations that may be at risk for problematic alcohol use. Specifically, alcohol ought to be particularly rewarding for individuals who otherwise struggle with social approach, self-expression, and affiliation. Of note, epidemiological data reinforce this idea, as individuals with social anxiety disorder are at increased risk for AUD (Schneier et al., 2010). Clinically, findings suggest potential treatment targets: If alcohol reinforcement partially reflects its impact on affiliative verbal behavior and subsequent social connection, quitting may decrease opportunities for such reward and interventions might further develop ways to enhance self-expression/social approach without drinking.

Limitations and Future Directions

Review Process Limitations

While this meta-analysis had some strengths (e.g., preregistering clear inclusion/exclusion criteria and outcome measure features, having a researcher blind to study results code the eligibility of study outcomes in all possibly relevant studies), it also had several limitations. We did not include unpublished studies, which could inflate results. Indeed, though most publication bias metrics did not show evidence of bias, trim and fill analyses suggested possible inflation of the effect size of alcohol. Other limitations involved the inability to examine mediational and moderating processes due to limited data. Below we detail these limitations and address ways future work may move forward.

Causal Processes

A major inference of this work is that the observed increases in affiliative verbal behavior under alcohol contribute to experiences of social reward and thereby reinforce drinking. Research is needed, however, to test this pathway. Such research is vital, as both theory and data indicate that by the very same processes that increase social approach, alcohol can produce antisocial forms of disinhibition, such as aggression (e.g., Bushman & Cooper, 1990). Indeed, while not a focus here, several studies observed increases in verbal negativity in addition to affiliative verbal behavior (e.g., Babor et al., 1983; Leonard & Roberts, 1998). A critical next step will be to assess the mediational impact of alcohol-related increases in affiliative verbal behavior on subsequent reward and bonding.

Effect Moderation

Given the limited number of studies, missing effect size data, and little replication of methods or measures, a major limitation of the extant literature is the inability to quantify when and for whom alcohol’s effects on affiliative verbal behavior are most evident. Nonetheless, there are several future directions for moderation studies.

Task and Context.

First, studies differed in terms of interaction contexts. Though data were tentative, there was some indication that the effects of alcohol varied between these tasks, such that there was stronger evidence for an effect in free discussion paradigms and collaborative tasks than in one-sided confederate interactions. Data were also mixed for the conflict resolution paradigms, perhaps due to the lack of a fixed dose of alcohol in some studies. Notably, this variability across interaction contexts accords with evidence that alcohol’s effects vary by context (e.g., Corbin et al., 2015) and with a meta-analysis reporting selective mood-enhancing effects of alcohol in unscripted experimental paradigms, as compared to interactions with confederates (Fairbairn & Sayette, 2014). Perhaps because the confederate’s behavior is controlled and predictable, alcohol is less effective at disrupting the processing of rejection cues in these interactions (Fairbairn & Sayette, 2014). Alternatively, one-sided interactions might reduce the salience of approach motivations, as participants may be less interested in affiliating with an unresponsive partner. The less clear effects of alcohol observed in conflict paradigms might also be understood from this perspective; resolving a marital dispute likely made approach motives less salient and may have primed anti-social motives.^vi Future research should consider testing the bounds of alcohol’s effects on affiliative verbal behavior by manipulating the salience of approach motives and the certainty of avoidance cues.

Relationship Familiarity and Quality.

Familiarity among interactants in the current review ranged from strangers to married couples. Because relationship features differed largely between task categories (e.g., spouses were involved in conflict paradigms, and most free discussions involved strangers), it is challenging to distinguish the potential effects of familiarity from other task features. Nevertheless, alcohol reward appears to be more evident among interactions with less familiar individuals (Fairbairn, 2017; Fairbairn et al., 2018; Gurrieri et al., 2021). Future studies should consider the effects of alcohol on affiliative verbalizations among groups that vary in familiarity. If alcohol increases verbal approach, particularly in unfamiliar interactions, it would suggest that alcohol aids in initial formation more so than maintenance of social bonds.

Relatedly, research might contrast the effects of alcohol in romantic vs. platonic relationships. Interactions involving the potential for romantic connection are often marked by increased communication awkwardness and emotional volatility (Giordano et al., 2005). This heightened ambiguity and uncertainty suggests that alcohol may exert especially strong influences on verbal approach behavior in early romantic interactions. Moreover, because romantic relationships involve increased intimacy, passion, and commitment and are thought to have uniquely powerful effects on subjective well-being (Kansky, 2018), such effects of alcohol in initial romantic interactions may be uniquely reinforcing.

Alcohol Manipulations.

A major gap in current research is the lack of systematic consideration of alcohol manipulation effects (e.g., limb, dose, expectancy set). Though space precludes a thorough review, each of these topics represents an important future direction and is briefly addressed below.

Limb Effects.

Alcohol consumption generates biphasic effects, with more euphoric stimulating experiences as BAC rises to its peak and more sedative experiences as it declines (Holdstock & de Wit, 1998; Martin et al., 1993). Extant research on alcohol and affiliative verbal behavior has largely disregarded limb effects (i.e., only one study in the current review attempted to differentiate between rising and falling BAC). However, other research suggests that stimulant-like, but not sedative-like, effects of alcohol are related to increased behavioral disinhibition (Quinn & Fromme, 2016). Alcohol may therefore increase verbal approach behaviors particularly on the ascending limb. Because the current set of studies made analysis of limb difficult (i.e., there were relatively few studies, and these studies varied by dose, time for drinking, time for absorption, type and timing of social interaction task or tasks), we cannot yet address the hypothesis that affiliative behavior is most pronounced on the rising limb of the BAC curve. Studies aimed at carefully manipulating and describing BAC limb while also assessing verbal behavior are needed to inform this possibility.

Dosage Effects.

Studies included in the current review varied in alcohol dose from BAC ~ 0.04% to ~ 0.11 %. Significant effects of alcohol appeared along this continuum, providing little evidence of a dose effect.^vii However, this may partially reflect limited consideration of alcohol dose as a moderating factor within studies and heterogeneity in other methodological features across studies. Elsewhere, there is evidence that alcohol’s effects are dose-dependent (e.g., Peterson et al., 1990; Pohorecky & Brick, 1988), and higher alcohol doses (e.g., BAC > 0.06 %) are related to increased risk-taking (Lane et al., 2004) and more extreme social behavior (Steele & Southwick, 1985). One might, therefore, expect a dose-response relationship whereby verbal approach increases with increasing intoxication up to an inflection point, at which time the physical mechanics of speech come into play. Further work ought to test this possibility by comparing the effects of alcohol at various doses while holding other interactional features constant.

Dosage (Expectancy) Set.

People hold powerful expectations about the effects of alcohol (Goldman et al., 1999). Accordingly, placebo and control comparison conditions are often included in research to differentiate between expectancy and pharmacological effects of alcohol. We observed minimal evidence of a difference between these conditions (most studies contrasting these conditions showed no differences between placebo and control), indicating that alcohol’s impacts on affiliative verbal behavior are unlikely to be based on dosage set. Nevertheless, one small study reported that alcohol decreased self-disclosure compared to control but increased it compared to placebo (Caudill et al., 1987b). Because many studies in this review reported significant increases in measures of affiliative verbal behavior in alcohol compared to control conditions (e.g., Babor et al., 1983; Frankenstein, Nathan, et al., 1985; Lindfors & Lindman, 1987; Monahan & Lannutti, 2000; Testa et al., 2014), we caution against drawing conclusions from this one interaction. That said, it will be important for future studies to explore placebo effects that could explain this interaction, including compensatory effects (Testa et al., 2006, where the belief that one has consumed alcohol leads to hypervigilance and caution in an attempt to offset the expected impacts of intoxication) and negative alcohol expectancies. Studies could elaborate on these possibilities by examining placebo effects across contexts that prime varying levels of hypervigilance and conducting person-centered analyses that address individual alcohol expectancies (Kusurkar et al., 2021).

Type of Affiliative Verbal Behavior.

Research highlights the value of describing alcohol’s effect on both self- and other-focused forms of affiliative verbal behavior (e.g., disinhibited self-expression vs. responsiveness to an interaction partner). Specifically, though self-expression can be independently beneficial (e.g., Collins & Miller, 1994; Pennebaker & Beall, 1986), perceptions of partner responsiveness to such revelations are hugely influential on the degree of reward derived and on the continued process of intimacy-building (Krueger & Forest, 2022; Reis, 2017). Given the limited number of studies and variability across measures/methods in the current database, it is premature to draw conclusions about the differential effects of alcohol on self and other-focused affiliative verbal behavior. A study that simultaneously assesses the impact of alcohol on self-disclosure, partner responsiveness to self-disclosure, and perceptions of partner responsiveness would inform a deeper understanding of the processes through which alcohol is linked to social reward.

Study-Level Measurement and Methodological Limitations and Next Steps

The existing literature is marked by a lack of systematic data. We found 16 eligible studies, the vast majority of which were conducted decades ago. Many of the investigations reported incomplete effect size data, methods and measures were heterogenous, and study quality was generally weak. These limitations led to a conservative approach to the meta-analysis and results are therefore somewhat preliminary. The remainder of this discussion thus considers why research has been so sparse, describes methodological and measurement setbacks in the extant literature, and offers guidelines for future investigations.

Perhaps the most obvious explanation for the sparsity of research on alcohol and verbal behavior is the sheer amount of time and resources required to collect and analyze natural language data. To make full use of speech data, verbalizations should be transcribed. This is a laborious, multi-step process that requires coding manuals, reliability training and testing, speaker labeling, segmentation of speech turns, and careful typing of spoken words (e.g., Glenn et al., 2010). Once speech is transcribed (or if transcription is forgone altogether in favor of rating video/audio recordings), one must extract relevant information. For instance, many studies in this review used behavioral coding schemes to assess verbal content. Such schemes require the development of detailed manuals, reliability training and testing, and often many hours of manual coding (Black et al., 2013). Given the considerable effort to collect and make sense of language data, it is unsurprising that research has been slow-going. Moreover, the challenges associated with properly handling these data likely inform many current limitations in methods and measures.

Methodologically, many studies in this literature used small samples, assessed short interaction durations, or implemented tightly controlled laboratory paradigms.^viii These choices make the task of analyzing language data more manageable by reducing volume and limiting interactional complexity. The unfortunate byproduct, however, is reduced power to address complex questions and limited generalizability. For instance, small studies are underpowered to examine the many potential moderators of alcohol’s effects, and investigations that analyze only a few minutes of interactional data struggle to assess dynamic interactional processes over time (e.g., responsivity and intimacy building). Further, confining interactions to controlled laboratory environments damages the generalizability of results. Context matters when it comes to alcohol’s effects, and naturalistic interactions with increased uncertainty and more authentic social rewards may reveal the most benefits of alcohol.

Challenges associated with language data may also impact measurement decisions. Many investigations relied on established coding schemes or confined their assessments to idiosyncratic behaviors that could easily be placed into discrete categories (e.g., number of questions). Other studies implemented global assessments (e.g., summary codes across many forms of verbal positivity or Likert ratings across long durations). Such measures are economical and enhance reliability, yet they are limiting. They demand a focus on behaviors that have either been validated in previous research (and are not necessarily well-suited to the context at hand) or that are concrete and easily observable. Both constraints reduce the variety of questions that can be asked of text data. Alternatively, global codes can evaluate more abstract concepts but lack specificity. Moreover, many measures in the current database were based on coding of audio/video data, rather than transcriptions. This means verbal content is confounded with nonverbal features.

In summary, collecting and analyzing language data is a monumental task, which has limited the number of investigations examining alcohol’s impact on verbal behavior and has led to labor-saving, but suboptimal, methodological and measurement choices. Fortunately, emerging text-processing methodologies offer a promising direction. By efficiently collecting and assessing language data, these technologies can address many existing methodological and measurement challenges. Below, we outline how just some of these advancements may inform future work.

Collection of Language Data

Accurate transcripts of language data are required for most text-processing approaches. The arduous nature of human transcription, however, creates a bottleneck. Fortunately, novel technologies allow for automatic transcription of spoken language using artificial intelligence (Coto-Solano, 2022; Liao et al., 2023). Currently, these approaches have limitations, and they may struggle with noisy settings involving multiple speakers (Southwell et al., 2022). Manual transcription, therefore, remains best practice with natural conversations recorded by a single microphone. Nevertheless, future studies can design their data collection procedures to aid automated transcription: recording devices can now capture a single speaker’s utterances while blocking ambient noise (Southwell et al., 2022). Moreover, speech recognition research is a booming field and technologies will become even more effective over time.

Assessment of Language Data

Dictionary Methods.

Language dictionaries are a simple but highly effective tool for computerized text analysis. These databases contain predefined lists of words associated with category labels. Linguistic Inquiry and Word Count (LIWC; Boyd et al., 2022), for example, is a widely used dictionary that has been refined over the past 20 years. LIWC reads in text data and computes the number of words that fall into over 80 well-validated categories, including parts of speech, psychological processes, and social behaviors. LIWC can extract meaningful information from text data. It predicts perceptions of compassion and shared decision-making in healthcare settings (Zhou et al., 2023), as well as social rank and team performance in work settings (Sexton & Helmreich, 2000). It also identifies complex thought processes, such as mentalization and cognitive processing styles surrounding trauma (Pennebaker et al., 1997). LIWC categories also correspond to many forms of affiliative verbal behavior, including positive emotional tone, affiliative drives, and politeness (Boyd et al., 2022). LIWC can assess forms of verbal immediacy through counts of first-person plural pronouns (e.g., we, our), and an immediacy index, thought to reflect involvement and present focus (Lee et al., 2011). Further, content categories may allow researchers to understand how alcohol impacts the intimacy of conversation topics by examining rates of words related to family, friends, sexuality, or mental health.

Perhaps most promising, LIWC can catalog reciprocal group processes; linguistic style matching (LSM) is a LIWC-based index that measures the extent to which individuals in a conversation use language in the same way; it can be assessed on a turn-by-turn basis and can be analyzed for change over time. LSM is predictive of negotiation outcomes (Taylor & Thomas, 2008) in addition to relationship initiation and course (Ireland et al., 2011). It would be fascinating to examine the impact of alcohol on verbal coordination using the LSM metric: does intoxication pull us into the present (perhaps by limiting preoccupation with rejection) and thus allow us to reflect, coordinate, and engage with our partners more effectively? In sum, LIWC is a transparent, affordable, and well-validated tool that allows researchers to analyze text data instantaneously. It is thus somewhat striking that this software has not yet been employed in laboratory alcohol studies.

Natural Language Processing.

More recently, advances in natural language processing (NLP), a field of computer science that uses machine learning algorithms to process and understand human language, have vastly improved the efficiency with which researchers can extract complex psychological information from text data (Hagiwara, 2022). NLP is an incredibly diverse and rapidly growing field and comprehensively summarizing the myriad ways that it can inform psychological research is beyond the scope of this review (see Kennedy et al., 2021). Briefly, NLP methods are commonly based on complex statistical algorithms that use large, pre-labeled datasets to take in text data and over thousands of iterations optimize the weights assigned to various nodes of the model to learn to accurately predict the labeled outcome. Once trained, NLP algorithms are used to automatically classify and understand novel text. Recently, these algorithms have become robust enough that they can use “transfer learning” to apply knowledge gained from one training task to new related tasks (Ruder et al., 2019). This drastically reduces the burden of developing any single NLP algorithm, as it allows researchers to build onto pre-established models that have already acquired sophisticated base knowledge of language using billions of data points (e.g., BERT; Devlin et al., 2018). By collaborating with computer scientists, psychologists can leverage these emerging technologies to explore the impact of alcohol on affiliative verbal behavior with increasing complexity and efficiency. Already, for example, computer scientists have developed models that can identify humor in short texts with great accuracy (Annamoradnejad & Zoghi, 2020). Other algorithms have successfully classified self-disclosure (Mehdy et al., 2019) and even distinguished between categories of self-disclosure (e.g., informational vs. emotional: Akiti et al., 2020; Vedantham, 2021). Finally, other models have detected forms of verbal responsivity (e.g., therapeutic reflections, Can et al., 2016). As NLP research continues to progress, even more nuanced algorithms aimed at detecting a wide variety of verbal features will arise, broadening the kinds of questions we can ask while simultaneously drastically reducing the time required to do so.

Beyond assessing a variety of pre-trained forms of verbal positivity, NLP methods can also untether researchers from specific theoretical expectations by detecting latent patterns in text data. Topic modeling, for example, is a form of unsupervised machine learning that can identify themes in a document. Topic modeling approaches are conceptually similar to factor analyses; machine learning is used to identify words that tend to co-occur, and based on patterns of co-occurrence, words are probabilistically assigned to topic groupings (Berger & Packard, 2022). From there, texts can be scored based on how much of any given topic they contain. Researchers have already used this approach to better understand psychological phenomena, such as fears underlying social anxiety (Chiu et al., 2022). It is not a far leap to imagine how these methods could advance social alcohol research. It will be interesting, for instance, to discover whether alcohol impacts the themes that come up in conversation and if topics tend to be more intimate or risky. Further, more advanced unsupervised NLP methods allow for additional granularity. Word-embedding is a computational approach that accounts for both the co-occurrence and distance between words to assess semantic relationships (Berger & Packard, 2022). It can consequently provide insight into how people are relating ideas in their speech. This approach, for example, has revealed gender prejudices by examining the association between negatively valenced words and female words across gendered and non-gendered languages (DeFranza et al., 2020). Given enough data, similar techniques may reveal how alcohol affects how we talk about ourselves and others (e.g., do the word associations with various pronouns change when people are intoxicated?). Ultimately, by being data-driven, unsupervised approaches such as topic modeling and word embedding will open the door to exploratory work and novel hypotheses about alcohol’s effects on verbal behavior.

Summary of Next Steps

Text processing technologies are efficient tools that unlock new potential for alcohol research. Because these approaches allow for near-automatic assessment of verbal features, they can address many current challenges in methods and measurement. We can now collect large volumes of data without creating additional assessment burdens; well-powered examinations of moderation, mediation, or time-based effects are now feasible. These tools can also create standardization and replication, as laboratories across the world can use the same dictionary/algorithm without the need to train new coders. Additionally, as speech recognition technologies continue to improve, language samples may be recorded in increasingly naturalistic environments without significant detriments to transcription procedures. This will improve the generalizability of our studies and may even allow for data collection in the real world: recording devices can now randomly sample conversations in daily interactions (Mehl, 2017), this in combination with transdermal BAC monitoring (Fairbairn et al., 2020) presents a promising direction for Ecological Momentary Assessment (EMA) protocols. Even more, as NLP methods continue to progress, the scope of potential analyses will rapidly expand. We will be able to assess a wider variety of verbal behaviors with specificity and will no longer have to rely on pre-established behavioral coding schemes (which take years to develop), discrete codes, or global summaries. More broadly, researchers can now move beyond hypothesis testing, as novel exploratory methods can expand the questions we think to ask of our data. Alcohol researchers who can make use of the growing abundance of text-processing tools will develop a much richer understanding of how alcohol impacts affiliative behavior and bonding.

Conclusions

This review to our knowledge represents the first-ever systematic evaluation of the effects of alcohol on affiliative verbal behavior. Results provide evidence that alcohol intoxication creates significant, albeit modest, increases in affiliative verbalizations. Perhaps more importantly, the picture that emerged was one of a scattered literature that is ripe for continued investigation. More work is needed to solidify the findings reported here and to explore the effects of alcohol with increased nuance by examining mediational, moderating, and reciprocal processes. Alcohol researchers will benefit greatly if they can lean on emerging technologies drawn from a flourishing text-processing literature to address these questions. By collaborating with computer scientists, alcohol researchers can analyze speech data with more efficiency and specificity than ever before. Ultimately, this will allow the field to develop a better understanding of how alcohol impacts verbal communication. This may be an important piece of the puzzle when it comes to understanding why humans drink and will inform theory and practice alike.

Data Availability:

The datasets analyzed during the current study are available from the corresponding author upon reasonable request.