1. Introduction
Language in human interaction can be described as a process of individuals committing to the content of their speech, forging a sense of responsibility and enabling coordination between speakers (Fernández-Castro & Pacherie, Reference Fernández-Castro and Pacherie2023; Geurts, Reference Geurts2019; Marsili, Reference Marsili2024). These linguistic commitments are not limited to mere information exchange; they can also influence how people perceive one another and navigate social norms (e.g., Boogaart et al., Reference Boogaart, Jansen and van Leeuwen2021; Hall & Mazzarella, Reference Hall and Mazzarella2023; Mazzarella et al., Reference Mazzarella, Reinecke, Noveck and Mercier2018; Reins & Wiegmann, Reference Reins and Wiegmann2021; Weissman, Reference Weissman2024). Often, speech content extends beyond the explicit statement, incorporating implied meanings that require listeners to make inferences to fully grasp the intended message. Consider, for instance, the following interaction between Dan and Josh:
(1)
Dan: ‘Are we going to the gym?’
Josh: ‘My ankle is hurting’
The question in (1) requires a direct ‘yes’ or ‘no’ response, yet such a response is notably missing in the answer. Nonetheless, it is reasonable to assume that Dan interprets Josh’s reply as a rejection of his suggestion, even in the absence of an explicit ‘no’. This type of inference, known as relevance inference or relevance implicature, is rooted in the relevance maxim of Grice’s influential theory (Reference Grice1975). Grice posited that speakers strive to be truthful, informative, concise and relevant. In the example provided, the omission of a clear ‘yes’ or ‘no’ can appear irrelevant. To determine its relevance, it is necessary to analyze the relationship between the initial question (regarding gym attendance) and the subsequent response (referencing ankle pain). The inference of the implicit meaning is derived from this further examination, implying that Josh’s ankle discomfort impedes his ability to attend the gym. In his work, Grice distinguished between two critical aspects of an implicature: ‘what is said’, the explicit meaning (or the verbatim), encompassing the closely conventional meaning of the uttered words, and ‘what is meant’, pertaining to the implicit meaning (or the ‘gist’) inferred by the listener (Grice, Reference Grice1957). That is, in our example, ‘what is said’ concerns Josh’s ankle pain and ‘what is meant’ relates to his gym attendance. We note that Grice conventionally defined meaning as the content that the speaker intended to convey. Thus, ‘what is meant’ is only so if the speaker intended it. This crucially complicates conducting empirical research, where the speaker’s intention is difficult or even impossible to decipher (see Terkourafi, Reference Terkourafi2021). In light of this, a prevailing approach involves viewing meaning as a socially grounded construct of the speaker’s intent, where speakers can be accountable for what they say and mean, sometimes regardless of intention (Assimakopoulos, Reference Assimakopoulos2022; Haugh, Reference Haugh2013; Terkourafi, Reference Terkourafi2021).
The socially grounded construct of the speaker’s intent gives rise to the notion of ‘commitment’, which is intricately linked to accountability. Commitment is defined as the assertion to the truth of the message. Importantly, in this view, communication is seen as a negotiation of commitments (Geurts, Reference Geurts2019; Marsili, Reference Marsili2024; Moeschler, Reference Moeschler2013), where speakers can be seen accountable for their explicit words, their intended messages and even the ultimate interpretation by the listener (Bonalumi et al., Reference Bonalumi, Scott-Phillips, Tacha and Heintz2020, Reference Bonalumi, Michael and Heintz2022, Reference Bonalumi, Bumin, Scott-Phillips and Heintz2023a; Geurts, Reference Geurts2019; Lyu & Yuan, Reference Lyu and Yuan2023; Reins & Wiegmann, Reference Reins and Wiegmann2021; Weissman, Reference Weissman2024; Yuan & Lyu, Reference Yuan and Lyu2022).
The emphasis on commitment has shifted the focus toward the observers’ or listeners’ perspective, enabling empirical exploration of the distinction between the explicit and implicit meanings. One common way to examine commitment is by manipulating the revelation of the actual events (i.e., State of Affairs; SoA). More specifically, studies present an exchange that either elicits an implicit meaning or delivers the same meaning explicitly. Afterward, this meaning is contradicted by introducing a scenario in which a subsequent event opposes what was initially understood (Bonalumi et al., Reference Bonalumi, Scott-Phillips, Tacha and Heintz2020, Reference Bonalumi, Michael and Heintz2022, Reference Bonalumi, Bumin, Scott-Phillips and Heintz2023a, Reference Bonalumi, Mahr, Marie and Pouscoulous2023b; Boogaart et al., Reference Boogaart, Jansen and van Leeuwen2021; Boulat & Maillat, Reference Boulat and Maillat2023; Hall & Mazzarella, Reference Hall and Mazzarella2023; Lyu & Yuan, Reference Lyu and Yuan2023; Mazzarella et al., Reference Mazzarella, Reinecke, Noveck and Mercier2018; Reins & Wiegmann, Reference Reins and Wiegmann2021; Yuan & Lyu, Reference Yuan and Lyu2022). In the example above, such manipulation would reveal that Josh went to the gym, despite Dan’s (probable) inference. Then, researchers compare commitment under conditions in which the speaker used explicit and implicit responses. This line of research has yielded a complex picture: Several studies demonstrated greater commitment to explicit meaning over implicit meaning (Hall & Mazzarella, Reference Hall and Mazzarella2023; Mazzarella et al., Reference Mazzarella, Reinecke, Noveck and Mercier2018). However, other studies have shown that under certain conditions, the level of commitment to implicit meaning can be comparable to that of explicit meaning (Bonalumi et al., Reference Bonalumi, Scott-Phillips, Tacha and Heintz2020; Reins & Wiegmann, Reference Reins and Wiegmann2021; Yuan & Lyu, Reference Yuan and Lyu2022). Indeed, commitment is a complex construct influenced by more than just the explicitness of the content. Various factors can modulate commitment, including the strength of contextual assumptions (Boulat & Maillat, Reference Boulat and Maillat2023), the type of speech act (Yuan & Lyu, Reference Yuan and Lyu2022), the SoA (Braun et al., Reference Braun, Kuperwasser and Shetreet2024), alternative interpretations (Braun et al., Reference Braun, Kuperwasser and Shetreet2024) and speaker-related contexts (Bonalumi et al., Reference Bonalumi, Michael and Heintz2022; Lyu & Yuan, Reference Lyu and Yuan2023). When these factors are kept constant and only explicitness/implicitness is examined, the pattern of results suggests a general trend of higher commitment to an explicit message compared to its implicit counterpart.
Importantly, in previous studies, the dynamics of explicit versus implicit meanings have been compared using separate utterances that convey the same message, either explicitly or implicitly. To illustrate, in Mazzarella et al. (Reference Mazzarella, Reinecke, Noveck and Mercier2018), when asked why the projector was not set up, the responses were either explicit, such as ‘The projector is not here. Louis removed it. It is at the central office’, or implicit, such as ‘Louis gave a presentation at the central office yesterday’. That is, these experiments compared different forms of utterances. However, within a single utterance (e.g., ‘my ankle is hurting’), observers may process commitment toward two potentially distinct messages: (a) the explicit meaning, ‘Josh’s ankle is hurting’, and (b) the implicit meaning, such as ‘Josh cannot come to the gym’. Previous studies have not examined the commitment patterns between the explicit and implicit meaning that exist within a single response. Consequently, it remains unclear what elements of a response eventually affect perceived commitment.
The interplay between explicit and implicit meanings within a single utterance is particularly relevant in everyday interactions, especially when commitment is at stake. Consider, for example, the role of ‘white lies’ – those minor untruths intended to avoid conflict or spare someone’s feelings. Imagine a semi-close friend notifying you that they cannot make it to your party by claiming ‘I woke up really sick, probably the flu’. Even if you are skeptical about the truthfulness of their excuse, you would likely perceive the friend as fully committed to not attending the party. This subtle tension between explicit and implicit communication reflects an attempt to balance honesty with social harmony through ‘white lies’ (Bok, Reference Bok2011; Terkourafi, Reference Terkourafi and Meibauer2018). Conversely, one may deliberately manipulate the balance between explicit and implicit meanings as a form of deception, exploiting the implicit meaning of literal truths. For example, someone might use carefully worded statements that are technically accurate but designed to mislead by implying something untrue (Meibauer, Reference Meibauer2023; Reins & Wiegmann, Reference Reins and Wiegmann2021; Wiegmann et al., Reference Wiegmann, Willemsen and Meibauer2022). Indeed, if you discover that the same friend attended a different party, it becomes irrelevant whether their explicit statement (i.e., that they woke up feeling very sick) is truthful; their behavior is likely to still be perceived as an attempt to deceive – particularly by making the implicit message misleading. Uncovering the truthfulness of the explicit versus implicit content in these two scenarios, thus, seems to result in different social consequences. In the examination of commitment, it could be hypothesized that commitment is processed either independently for each stream of meaning (explicit and implicit) or, alternatively, more holistically. Considering commitment as independently processed, the revelation of the actual events concerning a certain content (explicit or implicit) should only influence our commitment to that specific meaning (i.e., the explicit or implicit meaning, respectively). Using the example in (1), if the understood implicit meaning turns out to be incorrect, such as if Josh actually goes to the gym, independent processing would affect commitment to the implicit meaning only, without altering commitment to the explicit meaning, namely that his ankle was indeed hurting. Conversely, if commitment is processed holistically, the truth value (i.e., the SoA) of one type of content is expected to influence the commitment patterns to the other meaning as well. In the above example, this would mean that the commitment to the explicit meaning (hurting ankle) would also change.
The present study seeks to explore commitment to explicit versus implicit meanings. Specifically, we ask whether commitment is processed holistically or independently for each stream of meaning. To do so, we manipulated the revelation of the actual events (SoA) of the two streams of meaning: the explicit, ‘what is said’ and the implicit, ‘what is meant’, using all possible combinations of SoAs. We assessed two measures: commitment using an explicit commitment question (as in Braun et al., Reference Braun, Kuperwasser and Shetreet2024) and general trust using a trust question (as in Braun et al., Reference Braun, Kuperwasser and Shetreet2024; Yuan & Lyu, Reference Yuan and Lyu2022). Notably, while others used trust questions as a measure of commitment (e.g., Bonalumi et al., Reference Bonalumi, Scott-Phillips, Tacha and Heintz2020, Reference Bonalumi, Mahr, Marie and Pouscoulous2023b; Mazzarella et al., Reference Mazzarella, Reinecke, Noveck and Mercier2018; Yuan & Lyu, Reference Yuan and Lyu2022), we treated commitment and trust as distinct. This distinction follows from the different constructs each of them indexes (Braun et al., Reference Braun, Kuperwasser and Shetreet2024). Commitment refers to the speaker’s stance toward a particular utterance, specifically the extent to which the speaker is perceived as standing behind the proposition conveyed. Trust, in contrast, reflects a broader judgment of the speaker’s reliability or credibility across contexts. As such, trust may be expected to be shaped more holistically by both the explicit and implicit content. Although both relate to truth assessment, they need not depend on the same underlying cognitive processes (see Braun et al., Reference Braun, Kuperwasser and Shetreet2024). These two measures were used to examine whether commitment and trust in each stream of meaning are influenced by the other stream.
2. Experiment 1
The first experiment aimed to explore how the explicit meaning of a message affects the implicit meaning, and vice versa. We used relevance inferences to assess perceived commitment to both the explicit and implicit meanings. Crucially, we manipulated the truth value of both the explicit (affirmed or voided) and the implicit (affirmed or voided) contents. This manipulation yielded four distinct combinations of SoAs (see Figure 1).
Paradigm run-through and examples of the different combinations of State of Affairs presented to participants.
Figure 1. Long description
A horizontal process flow at the top consists of five grey rounded rectangles connected by right-pointing arrows. The sequence is: General Context, Exchange with inference, State of Affairs for the explicit and implicit content, Commitment question for the explicit or implicit meaning, and Trust question.
Below the second box, Exchange with inference, is a screenshot of a mobile chat interface. A contact named Josh is online. A green message bubble asks, Are we going to the gym? A white reply from Josh says, My ankle is hurting.
Below the third box, State of Affairs, is a two-by-two matrix. The columns are labeled State of Affairs of the explicit content- Affirmed and State of Affairs of the explicit content- Voided. The rows are labeled State of Affairs of the implicit content- Affirmed (affirms speaker expected inference) and State of Affairs of the implicit content- Voided (voids observer’s expected inference).
The four cells contain the following outcomes:
1. Top-left: Josh’s ankle was indeed hurting and he meant to say no, he could not go to the gym.
2. Top-right: Josh’s ankle was not hurting and he meant to say no, he could not go to the gym.
3. Bottom-left: Josh’s ankle was indeed hurting and he meant to say yes, he was able to go to the gym.
4. Bottom-right: Josh’s ankle was not hurting and he meant to say yes, he was able to go to the gym.
Explicit and implicit contents are expected to influence evaluations, with lower ratings for voided meanings. The effects may be independent or integrative. Furthermore, the presence of a single voided meaning may disproportionately influence speaker’s evaluations. For example, if the implicit content is false, it may dominate the evaluative process such that the truth value of the explicit content becomes relatively less consequential. In this case, the evaluative weight assigned to one falsehood may exceed the additive impact of two falsehoods. Therefore, our analyses included both main effects of the SoA of the explicit content and that of the implicit content, as well as their interaction. The interaction could inform us whether a single falsehood qualitatively shifts overall judgments rather than contributing linearly to them.
2.1. Methods
2.1.1. Participants
Ninety-one participants (47 females, 45 males; M_age = 30.48) were included in the final sample. One additional participant was excluded due to incorrect responses to comprehension questions. The target sample size was determined via a power analysis based on data from a separate pilot study (N = 12; not included in the main analysis), conducted using G*Power 3.1 (Faul et al., Reference Faul, Erdfelde, Buchner and Lang2009).Footnote 1 A linear mixed-effects model was used to analyze participants’ responses in relation to their commitment to the implicit meaning. The model comprised of the fixed effects of SoA of the explicit content (affirmed/voided), SoA of the implicit content (affirmed/voided), their interactions and random intercepts for subject and item. Our sample size calculation was based on the size effect of the interaction and used the F-test family with linear multiple regression. Based on an assumed effect size of 0.023 and a desired statistical power of 80%, the required total sample size was estimated to be 344 trials. Importantly, the planned mixed-effects regression analysis treats each trial as an independent observation within participants, rather than aggregating responses. As such, the G*Power analysis outcome pertains to the number of trials, not participants. Because in each run of an experiment, participants answer four commitment questions regarding the explicit content and four regarding the implicit content, this translated to 86 participants (344/4 = 86). Anticipating that some trials would be excluded due to participants potentially inferring interpretations that diverged from the pre-tested dominant readings, we aimed to recruit approximately 90 participants. The final sample, following exclusions, comprised 91 participants.
Participants were recruited via Prolific. Eligibility criteria included being between 20 and 40 years of age, monolingual English speakers and having no diagnosis of autism. Participants also reported no literacy difficulties and had a Prolific approval rate exceeding 95%. Additionally, recruited participants were residents of either the US or the UK. Informed consent was obtained from all participants prior to their involvement in the study.
2.1.2. Materials and procedure
For the communicative exchanges, we selected conversational inferences in which a question was followed by a response that could be interpreted as either ‘yes’ or ‘no’, with a strong bias toward one of the two interpretations. To confirm the presence of the intended bias, a pretest was conducted with 15 participants. These participants did not participate in the main experiment. They were presented with 20 different inferences and chose their understood meaning, between ‘yes’ or ‘no’. Additionally, they were required to rank their level of confidence in their understanding, between 1 and 100. We selected four ‘yes’-response stories and four ‘no’-response stories (eight experimental trials in total). The final set of stories included only those dialogues where the same interpretation was identified by over 90% of participants and where confidence in these interpretations exceeded 60%. The final materials, analysis and design can be accessed on OSF link: https://osf.io/unmy6/overview.
The experimental paradigm is available in Figure 1. Participants were first given a general context that only presents the interlocutors, which was then followed by an exchange between them, formatted as a text message conversation. Each exchange featured an utterance designed to prompt a relevance inference, favoring either a ‘yes’ or ‘no’ interpretation. Participants were then instructed to indicate their interpretation by selecting either a ‘yes’ or ‘no’ response. This ‘inference comprehension question’ was included solely to verify that participants arrived at the intended interpretation and was not part of our outcome measures of commitment or trust. Following this, they were shown the speaker’s actual intended meaning – that is, the corresponding SoA, in one of the four SoA combinations: (1) Explicit content affirmed & Implicit content affirmed; (2) Explicit content affirmed & Implicit content voided; (3) Explicit content voided & Implicit content affirmed and (4) Explicit content voided & Implicit content voided (see Figure 1).
Participants then answered one commitment question which pertained either to the explicit meaning (e.g., ‘To what extent do you think Josh was committed to conveying his ankle is hurting?’) or the implicit meaning (e.g., ‘To what extent do you think Josh was committed to saying no?’). In other words, participants were asked about the commitment toward the explicit and implicit meaning but never within the same trial (i.e., never for the same exchange). This design was intended to prevent direct comparisons between levels of commitment to different meanings of the same utterance. Responses to the commitment question were given on a 1–100 scale ranging from ‘not committed’ to ‘fully committed’. Following this, participants answered an additional trust question adapted from Yuan and Lyu (Reference Yuan and Lyu2022): an explicit trust question asking them to rate, on a scale from 1 to 100, how reliable they perceived the speaker to be (e.g., ‘Is Josh a reliable person?’). This general trust, or reliability, question was presented at the end of every trial. The two commitment questions and the trust question served as our three outcome measures.
Each participant completed eight target trials, with two trials per condition, counterbalanced across participants and lists. Half of these trials strongly favored an inferred ‘yes’ interpretation, while the other half favored an inferred ‘no’. In addition, commitment questions were split evenly: half targeted the explicit meaning, and half targeted the implicit meaning. To maintain participants’ attention, general yes/no memory questions were presented after each exchange.
2.2. Results
One participant was excluded from the analysis for responding incorrectly to more than two memory questions. In addition, 20 trials (out of 728) were removed because participants’ responses to the inference comprehension question did not align with the pre-identified dominant interpretation.
Three mixed-effects models were fitted, each addressing a separate dependent variable: commitment to the explicit meaning, commitment to the implicit meaning and general trust. Note that the questions probing commitment were specifically tailored to examine commitment to either the explicit or implicit meaning. In contrast, the general trust question was a consistent, holistic measure that remained unchanged throughout. This is because the trust question examines the general reliability of the speaker and does not directly pertain to a specific meaning. The full model comprised of the fixed effects of SoA of the explicit content (affirmed/voided), SoA of the implicit content (affirmed/voided), their interaction and random intercepts for subjects and items (see Figure 2).
The top panel (A) illustrates patterns of commitment to implicit meaning (left) and explicit meaning (right). Note: Results indicate that the SoA of the implicit content significantly influenced commitment to the explicit meaning, in addition to its expected effect on commitment to the implicit meaning. The SoA of the explicit content significantly influenced commitment to the explicit meaning only. The bottom panel (B) presents trust ratings across all trials. All SoAs, as well as their interaction, significantly influenced trust ratings. Boxes represent the interquartile range (25th to 75th percentile), whiskers indicate the spread of the data, dots represent the mean and error bars indicate standard errors of the mean.
Figure 2. Long description
Panel A contains two side-by-side box plots. Both have a Y-axis for Commitment rating from 0 to 100. The X-axis for both includes four conditions: Explicit affirmed/Implicit affirmed, Explicit voided/Implicit affirmed, Explicit affirmed/Implicit voided, and Explicit voided/Implicit voided.
In the left plot, Commitment to implicit meaning, the mean rating is highest for the first two conditions where implicit is affirmed (around 70 and 65) and drops significantly for the two conditions where implicit is voided (around 40).
In the right plot, Commitment to explicit meaning, the mean is highest when both are affirmed (above 80). It remains relatively high (around 75) when only one is voided, but drops to approximately 55 when both explicit and implicit are voided.
Panel B, Trust ratings, is located below Panel A. It uses the same X-axis conditions and a Y-axis for Trust rating from 0 to 100. Trust shows a clear step-down trend from left to right: highest when both are affirmed (mean near 70), lower when explicit is voided (mean near 45), lower still when implicit is voided (mean near 40), and lowest when both are voided (mean near 30).
Across all plots, blue boxes represent affirmed implicit content and red boxes represent voided implicit content. Individual data points are scattered behind the box plots, and error bars indicate standard errors of the mean.
In the analysis of commitment to the implicit meaning, only the SoA of the implicit content emerged as a significant predictor of commitment level (β = 22.33, p < 0.001). We observed no effect of the SoA of the explicit content (β = −0.69, p = 0.846) and no significant interaction (β = 8.28, p = 0.102). The fixed effects accounted for 19% of the variance in the commitment rating (R2m = .19).
Conversely, in the analysis of commitment to the explicit meaning, both the SoA of the explicit content and the SOA of the implicit content emerged as significant predictors of commitment levels (β = 20.04, p < 0.001 and β = 18.97, p < 0.001, respectively). However, we did not observe a significant interaction (β = −8.46, p = 0.083). The fixed effects accounted for 15% of the variance in the commitment rating (R2m = .15).
For general trust, both SoAs, as well as the interaction, were significant predictors (β = 11.30, p < .001 for the SoA of the explicit content, β = 16.40, p < .001 for the SoA of the implicit content, and β = 12.02, p < .001 for the interaction). To unpack this interaction, we conducted simple-effects comparisons in both directions. First, we examined the effect of explicit content separately for each level of implicit content. When the implicit content was voided, trust ratings were higher when the explicit content was affirmed than when it was voided (β = 11.30, t(606) = 4.41, p < .001). Similarly, when the implicit content was affirmed, trust ratings were higher when the explicit content was affirmed than when it was voided (β = 23.33, t(605) = 9.12, p < .001). Second, we examined the complementary simple effects of implicit content. Trust ratings were higher when the implicit content was affirmed than voided, both when the explicit content was voided (β = 16.40, t(606) = 6.39, p < .001), and when the explicit content was affirmed (β = 28.42, t(607) = 11.04, p < .001). Thus, both explicit and implicit content affected trust at both levels of the other factor. The fixed effects accounted for 23% of the variance in the commitment rating (R2m = .23).
We performed an additional exploratory examination, comparing the overall commitment scores for the explicit meaning and the implicit meaning (Figure 3). This was done to explore similarities with previous findings that compared commitment to the explicit meaning with commitment to the implicit meaning. In other words, we aimed to demonstrate that our results align with the trends observed in previous studies. To this end, we used the average response for each participant across trials in which the implicit meaning was correctly understood (i.e., the inference comprehension question matched the strongly favored interpretation), and restricted the analysis to conditions in which both the explicit and implicit states of affairs were either affirmed or voided. The latter restriction was introduced to prevent any confounding influence of mixed-truth conditions on the interpretation of the results. Similar to previous findings, commitment ratings to the explicit meaning were significantly higher than commitment to the implicit meaning (t(90) = 5.247, p < .001).
Average commitment ratings as a function of whether the question targeted the explicit meaning or the implicit meaning (for the conditions in which the two meanings either affirmed or voided). Note: Commitment ratings were significantly higher for the explicit meaning than for the implicit meaning. Boxes represent the interquartile range (25th to 75th percentile), whiskers indicate the spread of the data, dots represent the mean and error bars indicate standard errors of the mean.
Figure 3. Long description
The Y-axis is labeled Commitment rating and ranges from 0 to 100 in increments of 20. The X-axis contains two categories: Implicit on the left and Explicit on the right.
* Implicit category: A light green box represents the interquartile range from approximately 42 to 71. A central black dot indicates a mean of approximately 56 with small vertical error bars. Whiskers extend from near 3 to 100. Light green data points are scattered vertically across the entire range.
* Explicit category: A light purple box represents the interquartile range from approximately 56 to 87. A central black dot indicates a mean of approximately 70 with small vertical error bars. Whiskers extend from approximately 14 to 100. Light purple data points are scattered vertically across the range, showing a higher density in the upper half compared to the implicit category.
2.3. Discussion
Experiment 1 showed that the SoA of the implicit content significantly influenced commitment to both the explicit (or the verbatim) meaning and the implicit (or gist) meaning. In contrast, the SoA of the explicit content influenced commitment to the explicit meaning only. In other words, the truth value of the implicit content affected commitment holistically, but the truth value of the explicit content only affected commitment locally, to the explicit meaning. We additionally examined trust, which was a metric that examined the general reliability of the speaker and did not directly pertain to a specific meaning. The SoAs of both the explicit content and the implicit content, as well as their interaction, had a significant impact on trust evaluations.
We note that the phrasing of the commitment question targeting the explicit meaning (e.g., ‘To what extent do you think Josh was committed to conveying that his ankle was hurting?’) may itself account for the observed asymmetry. Specifically, the use of the word conveying may have led participants to focus on the inferred or implicit meaning, potentially contributing on its own to the observed ‘implicit dominance’. In addition, the phrasing used in the revelation of the SoA may also have biased attention toward the implicit content. In particular, the revelation explicitly stated what the speaker meant to say, for example, ‘Josh meant to say yes, he was able to go to the gym’, whereas the explicit content was presented more simply, as in ‘Josh’s ankle was hurting’. Although this phrasing was intended to examine the role of the SoA of what was meant, the asymmetry itself, and the greater emphasis placed on speaker meaning, may have made the implicit content more salient and therefore more influential in judgments of commitment. To address these concerns, we conducted a second experiment.
3. Experiment 2
Experiment 2 followed the same design as Experiment 1, with two main modifications designed to minimize the salience of the implicit meaning. First, the question targeting the explicit meaning was rephrased as ‘To what extent do you think X committed himself to the claim that Y?’. For example, ‘To what extent do you think Josh committed himself to the claim that his ankle was hurting?’. Second, the state-of-affairs revelation no longer explicitly stated what the speaker meant to convey in the inferred meaning. For instance, instead of presenting ‘Josh meant to say yes, he was able to go to the gym’, we simply stated ‘Josh was able to go to the gym’.
3.1. Methods
3.1.1. Participants
A total of 93 participants (48 females, 45 males; M_age = 32) were included in the final sample. Three additional participants were excluded due to incorrect responses on memory questions (see below). The target sample size followed that of Experiment 1, and the inclusion criteria matched those of the previous study, with the added requirement that participants had not taken part in earlier experiments in this series.
3.1.2. Materials and procedure
Experiment 2 followed the same procedure and used the same materials as Experiment 1, differing only in the wording of the commitment question and the state-of-affairs revelation, as described above: First, the explicit commitment question was rephrased as ‘To what extent do you think X committed himself to the claim that Y?’. Second, the state-of-affairs no longer explicitly stated the inferred meaning; for example, ‘Josh meant to say he was able to go to the gym’ was replaced with ‘Josh was able to go to the gym’.
3.2. Results
Three participants were excluded from the analysis for responding incorrectly to more than two memory questions. In addition, 41 trials (out of 744) were removed because participants’ responses to the inference comprehension question did not align with the pre-identified dominant interpretation. We employed the same three models as in Experiment 1. See Figure 4 for a summary of the results.
The top panel (A) illustrates patterns of commitment to implicit meaning (left) and explicit meaning (right). Note: Results indicate that the SoA of the implicit content significantly influenced commitment to the explicit meaning, in addition to its expected effect on commitment to the implicit meaning. The SoA of the explicit content significantly influenced commitment to the explicit meaning only. An interaction was observed between the two SoAs when examining commitment to the implicit meaning. The bottom panel (B) presents trust ratings across all trials. All SoAs, as well as their interaction, significantly influenced trust ratings. Boxes represent the interquartile range (25th to 75th percentile), whiskers indicate the spread of the data, dots represent the mean and error bars indicate standard errors of the mean.
Figure 4. Long description
Panel A contains two box plots. The left plot, Commitment to implicit meaning, has a y-axis from 0 to 100. The x-axis lists four conditions: Explicit affirmed/Implicit affirmed (highest commitment, mean near 80), Explicit voided/Implicit affirmed (mean near 58), Explicit affirmed/Implicit voided (mean near 52), and Explicit voided/Implicit voided (lowest commitment, mean near 48). The right plot, Commitment to explicit meaning, shows the highest commitment for Explicit affirmed/Implicit affirmed (mean near 85), followed by Explicit voided/Implicit affirmed and Explicit affirmed/Implicit voided (both near 70), and the lowest for Explicit voided/Implicit voided (mean near 58).
Panel B, Trust ratings, shows a downward trend across the conditions. The Explicit affirmed/Implicit affirmed condition has the highest trust (mean near 75). Trust drops significantly for Explicit voided/Implicit affirmed and Explicit affirmed/Implicit voided (both means near 48), and reaches the lowest point for Explicit voided/Implicit voided (mean near 32).
In all plots, blue boxes represent Implicit affirmed conditions and red boxes represent Implicit voided conditions. Each box shows the interquartile range with a central black dot for the mean, error bars for standard error, and light grey dots representing individual data points spread along the whiskers.
As in Experiment 1, the analysis of commitment to the implicit meaning showed that the SoA of the implicit content emerged as a significant predictor of commitment level (β = 9.27, p = 0.043), but there was no effect of the SoA of the explicit content (β = 2.98, p = 0.506). Unlike Experiment 1, we also observed a significant interaction (β = 18.28, p < 0.01). To unpack this interaction, we conducted simple-effects comparisons in both directions. First, we examined the effect of explicit content separately for each level of implicit content. When the implicit content was voided, commitment to the implicit meaning did not differ depending on whether the explicit content was affirmed or voided (β = 2.98, t(352) = 0.67, p = .506). However, when the implicit content was affirmed, commitment to the implicit meaning was higher when the explicit content was also affirmed than when it was voided (β = 21.27, t(352) = 4.61, p < .001). Thus, the interaction reflects the fact that explicit content influences implicit commitment only when the implicit content is affirmed, and not when it is voided. Next, we examined the complementary simple effects of implicit content. As expected from the main effect, the implicit content affected commitment to the implicit meaning both when the explicit content was voided (β = 9.27, t(352) = 2.03, p = .043), and when the explicit content was affirmed (β = 27.55, t(352) = 6.08, p < .001). The fixed effects accounted for 14% of the variance in the commitment rating (R2m = 0.14).
Like in Experiment 1, the analysis of commitment to the explicit meaning revealed that both the SoA of the explicit content and the SOA of the implicit content emerged as significant predictors of commitment levels (β = 10.15, p < 0.01 and β = 12.19, p = 0.001, respectively). We did not observe a significant interaction (β = 6.05, p = 0.247). The fixed effects accounted for 13% of the variance in the commitment rating (R2m = 0.13).
Replicating the results of Experiment 1, when analyzing general trust, both SoAs, as well as the interaction, were significant predictors (β = 18.13, p < 0.001 for the SoA of the explicit content; β = 18.22, p < 0.001 for the SoA of the implicit content; and β = 7.8, p = 0.041 for the interaction). To unpack this interaction, we again conducted simple-effects comparisons in both directions. First, we examined the effect of explicit content separately for each level of implicit content. When the implicit content was voided, trust ratings were higher when the explicit content was affirmed than when it was voided (β = 18.13, t(603) = 6.80, p < .001). Similarly, when the implicit content was affirmed, trust ratings were higher when the explicit content was also affirmed than when it was voided (β = 25.93, t(608) = 9.51, p < .001). Second, we examined the complementary simple effects of implicit content. Trust ratings were higher when the implicit content was affirmed than voided, both when the explicit content was voided (β = 18.22, t(605) = 6.77, p < .001), and when the explicit content was affirmed (β = 26.02, t(605) = 9.65, p < .001). Thus, as in Experiment 1, both explicit and implicit content affected trust at both levels of the other factor. The fixed effects accounted for 24% of the variance in the commitment rating (R2m = 0.24).
3.3. Discussion
Overall, the results follow the main pattern observed in Experiment 1. In particular, we again found an implicit dominance pattern, whereby the implicit content, and its eventual SoA, had a stronger influence on commitment judgments, affecting not only commitment to the implicit meaning but also extending to commitment to the explicit meaning. In contrast, the explicit content showed a more limited and conditional influence where it affected commitment to the implicit meaning only in cases where the implicit meaning was affirmed. This pattern held despite changes to the experimental design that reduced the salience of the implicit meaning in both the commitment questions and the revelation of the states of affairs. However, unlike in Experiment 1, we observed a significant interaction between the two SoAs when examining commitment to the implicit meaning. Noteworthy, this interaction appears to be also driven by the stronger influence of the implicit content on commitment. Specifically, the difference between one falsehood and two falsehoods was evident only when the additional falsehood concerned the implicit content. Put differently, once the implicit content was voided, an additional falsehood in the explicit content had little further effect. With regard to trust, we replicated the pattern observed in Experiment 1, such that both the implicit and explicit contents, along with their interaction, influenced trust judgments.
Before further discussing these results, we would like to acknowledge that in certain cases, the explicit content of our stimuli was subjective (e.g., ‘my ankle is hurting’). Subjectivity, where truth is derived from personal experience, may have influenced our results, potentially explaining why commitment to the implicit meaning was not affected by the SoA of the explicit content. Experiment 3 thus aims to replicate the findings from Experiments 1 and 2, as well as examine the effects of subjectivity.
4. Experiment 3
This experiment extends the findings of Experiments 1 and 2 by examining the relationship of commitment to the explicit and implicit meanings through the lens of subjectivity. One fundamental distinction between subjective and objective statements lies in the grounding of their truth value. Subjective statements derive their veracity from the personal experiences or opinions of the speaker, thus depending upon the individual’s mind, or ‘subject’. In contrast, objective statements are independent of the individual’s mental state (Honderich, Reference Honderich2005). In other words, assessing the truthfulness of a statement becomes inherently more complex when relying on the speaker’s subjective perspective. To illustrate, consider our initial interaction, which includes subjective content (in 2a) and contrast it with a similar exchange that includes objective content (in 2b).
(2)
(a)
Dan: ‘Are we going to the gym?’
Josh: ‘My ankle is hurting’
(b)
Dan: ‘Are we going to the gym?’
Josh: ‘I strained my ankle’
As Josh’s statement about his pain in (2a) is inherently subjective, it is challenging to objectively verify or dispute it without direct access to his personal experience. Consequently, it is harder to assess whether a certain SoA, such as claiming Josh’s ankle was not actually hurting, truly contradicts a subjective statement due to the ambiguity of its veracity. This absence of concrete evidence complicates the assessment of truthfulness or falsity, potentially leading to a more ambiguous effect on commitment to the implicit meaning. Therefore, it is plausible that the impact of the SoA of the explicit content may be less pronounced when the explicit statement is subjective. In contrast, when the explicit statement has external criteria for evaluation, as with objective statements, the effect on commitment is likely to be more substantial. Identifying a fallacy in an objective assertion may trigger a more substantial breach in communication, leading to a more holistic effect that also impacts commitment toward the implicit meaning.
4.1. Methods
4.1.1. Participants
A total of 85 participants (45 females, 40 males; M_age = 30.46) were included in the final sample. Five additional participants were excluded due to incorrect responses on memory questions (see below). The target sample size followed that of Experiment 1, and the inclusion criteria matched those of the previous study, with the added requirement that participants had not taken part in earlier experiments.
4.1.2. Materials and procedure
The materials used were similar to those in Experiment 1, with the additional manipulation of subjectivity versus objectivity. To choose the materials, we created 22 pairs of dialogues, each containing a subjective response and an objective response, for the pretest phase. Following this, 14 participants evaluated the subjectivity of the responses on a scale ranging from 1 (highly subjective) to 100 (highly objective). Of the original 22 pairs, the 18 with the highest subjectivity/objectivity rating dichotomy were ultimately selected for inclusion. This assessment resulted in the formation of two distinct conditions of stories: the subjective condition, with an average rating of 20.5, and the objective condition, with an average rating of 75.5.
A subsequent pretest aimed at verifying that the two groups (subjective and objective) did not differ significantly in the strength of inference drawn from the exchanges. A different group of 15 participants was asked to indicate their understanding of the implied inference and their confidence in their interpretations. The final set of stories included only those dialogues where the same interpretation was identified by over 90% of participants and where confidence in these interpretations exceeded 60%. This criterion resulted in the selection of 16 pairs of exchanges, evenly divided between those biased toward a ‘yes’ response and those biased toward a ‘no’ response. Notably, there was no significant difference in the confidence levels reported by participants between the subjective and objective groups, with a p-value of 0.37.
Participants in the main experiment saw 16 target trials, 8 of them with target sentences including explicit content that were rated as highly subjective (e.g., ‘My ankle is hurting’), and 8 that were rated as highly objective (e.g., ‘I strained my ankle’). All other elements of the design were the same as in Experiment 1. The final materials, analysis and design can be accessed on OSF link: https://osf.io/unmy6/overview.
4.2. Results
Five participants were excluded from the analysis for responding incorrectly to more than two memory questions. In addition, 107 (out of 1360) trials were removed because participants’ responses to the inference comprehension question did not align with the pre-identified dominant interpretation. We employed the same analysis as Experiment 1, with the addition of the subjectivity/objectivity of the explicit content as a fixed effect, along with its interactions with the other fixed effects. When the model failed to converge, we removed the random factor that contributed the least to the variance. See Figures 5 and 6 for a summary of the results. Overall, all the results replicate the findings of Experiment 2, with no effects of the subjectivity factor.
The top panel (A) illustrates patterns of commitment to implicit meaning (left) and explicit meaning (right). Note: Results indicate that the SoA of the implicit content significantly influenced commitment to the explicit meaning, in addition to its expected effect on commitment to the implicit meaning. The SoA of the explicit content significantly influenced commitment to the explicit meaning only. An interaction was observed between the two SoAs when examining commitment to the implicit meaning. The bottom panel (B) presents trust ratings across all trials. All SoAs, as well as their interaction, significantly influenced trust ratings. Boxes represent the interquartile range (25th to 75th percentile), whiskers indicate the spread of the data, dots represent the mean and error bars indicate standard errors of the mean.
Figure 5. Long description
A multi-panel figure with three box plots. Each plot shares a common x-axis with four conditions: Explicit affirmed/Implicit affirmed, Explicit voided/Implicit affirmed, Explicit affirmed/Implicit voided, and Explicit voided/Implicit voided. The y-axis for all plots ranges from 0 to 100.
Panel A, Left: Commitment to implicit meaning. The Commitment rating is highest in the Explicit affirmed/Implicit affirmed condition (mean approximately 75) and decreases across the subsequent conditions, reaching its lowest point in the Explicit voided/Implicit voided condition (mean approximately 40).
Panel A, Right: Commitment to explicit meaning. The Commitment rating is highest when both are affirmed (mean approximately 85). It shows a gradual decline as content is voided, but remains higher overall than the implicit meaning ratings, with the lowest mean around 60 for the double voided condition.
Panel B: Trust ratings. The Trust rating shows a sharp decline from the Explicit affirmed/Implicit affirmed condition (mean approximately 75) to the other three conditions. The Explicit voided/Implicit affirmed and Explicit affirmed/Implicit voided conditions show similar means around 50, while the Explicit voided/Implicit voided condition shows the lowest trust rating with a mean near 35.
In all plots, individual data points are shown as small dots, boxes represent the interquartile range, whiskers show the data spread, and a central black dot with error bars indicates the mean and standard error of the mean.
Commitment ratings (Top panel) and trust ratings (Bottom panel) as a function of whether the explicit content of the utterance was subjective or objective. Note: Boxes represent the interquartile range (25th to 75th percentile), whiskers indicate the spread of the data, dots represent the mean and error bars indicate standard errors of the mean.
Figure 6. Long description
The figure consists of three panels labeled A and B.
Panel A contains two side-by-side box plots. The Y-axis represents Commitment rating from 0 to 100. The X-axis has two categories: Subjective (pink box) and Objective (green box).
- The left plot, titled Commitment to implicit meaning, shows both categories with a wide interquartile range from approximately 30 to 82. The mean for both is near 57 with small error bars.
- The right plot, titled Commitment to explicit meaning, shows higher ratings. The Subjective box spans from 58 to 93 with a mean near 70. The Objective box spans from 62 to 95 with a mean near 73.
Panel B, located below, is titled Trust ratings. The Y-axis represents Trust rating from 0 to 100.
- The Subjective pink box spans from 28 to 76 with a mean near 51.
- The Objective green box spans from 26 to 77 with a mean near 52.
In all plots, individual data points are shown as a vertical cloud of dots behind the boxes. Whiskers extend to the full range of the data, and a black dot with vertical error bars represents the mean and standard error.
In the analysis of commitment to the implicit meaning, the SoA of the implicit content emerged as a significant predictor of commitment level (β = 24.69, p < 0.001), but there was no effect of the SoA of the explicit content (β = 3.74, p = 0.338). Additionally, we observed a significant two-way interaction between the two SoAs (β = 11.49, p = 0.038). To unpack this interaction, we again conducted simple-effects comparisons in both directions. First, we examined the effect of explicit content separately for each level of implicit content. When the implicit content was voided, commitment to the implicit meaning did not differ depending on whether the explicit content was affirmed or voided (β = 3.62, t(544) = 1.31, p = .19). However, when the implicit content was affirmed, commitment to the implicit meaning was higher when the explicit content was also affirmed than when it was voided (β = 11.33, t(544) = 4.04, p < .001). Thus, once again, the interaction reflects the fact that explicit content influences implicit commitment only when the implicit content is affirmed, and not when it is voided. Second, we examined the complementary simple effects of implicit content. As expected from the main effect, the implicit content affected commitment to the implicit meaning both when the explicit content was voided (β = 24.99, t(546) = 8.95, p < .001), and when the explicit content was affirmed (β = 32.71, t(546) = 11.78, p < .001). The fixed effects accounted for 25% of the variance in the commitment rating (R2m = 0.25). Conversely, in the analysis of commitment to the explicit meaning, both the SoA of the explicit content and the SOA of the implicit content emerged as significant predictors of commitment levels (β = 7.93, p = 0.031 and β = 10.41, p < 0.01, respectively). We did not observe a significant interaction (β = 4.89, p = 0.346). There was no main effect of subjectivity/objectivity (β = −5.97, p = 0.104), nor did subjectivity/objectivity interact with any other fixed effect (all ps > 0.346). The fixed effects accounted for 12.5% of the variance in the commitment rating (R2m = 0.125).
For general trust, both SoAs, as well as their interaction, were significant predictors (β = 16.84, p < 0.001 for the SoA of the explicit content; β = 13.78, p < 0.001 for the SoA of the implicit content; and β = 14.79, p < 0.001 for the interaction of the two SoAs). To unpack this interaction, we again conducted simple-effects comparisons in both directions. First, we examined the effect of explicit content separately for each level of implicit content. When the implicit content was voided, trust ratings were higher when the explicit content was affirmed than when it was voided (β = 13.30, t(1166) = 7.02, p < .001). Similarly, when the implicit content was affirmed, trust ratings were higher when the explicit content was also affirmed than when it was voided (β = 27.30, t(1166) = 14.28, p < .001). Second, we examined the complementary simple effects of implicit content. Trust ratings were higher when the implicit content was affirmed than voided, both when the explicit content was voided (β = 14.90, t(1166) = 7.76, p < .001), and when the explicit content was affirmed (β = 28.90, t(1165) = 15.31, p < .001). Thus, as in Experiments 1 and 2, both explicit and implicit content affected trust at both levels of the other factor. The fixed effects accounted for 27% of the variance in the commitment rating (R2m = 0.27).
4.3. Discussion
The results of Experiment 3 continue the pattern observed in Experiments 1 and 2, demonstrating an interdependence between commitment to explicit and implicit meanings. Specifically, the implicit content, and its SoA, influenced commitment judgments for both the implicit and the explicit meanings, whereas, like in Experiment 2, the explicit content had a more limited effect: it primarily affected commitment to the explicit meaning, and affected commitment to the implicit meaning only when the implicit content was affirmed. In other words, once again, we see indication of an ‘implicit dominance’. As in Experiment 2, we also found a significant interaction between the SoA of the implicit content and that of the explicit content when examining commitment to the implicit meaning. This interaction followed the same pattern as in Experiment 2: once the implicit content was voided, an additional falsehood in the explicit content had little further effect. Thus, the interaction further supports a pattern of implicit dominance.
Importantly, this pattern held regardless of whether the explicit content was subjective or objective, as indicated by the absence of both a main effect of subjectivity/objectivity and any interactions between this factor and the other fixed effects. Therefore, the manipulation of subjectivity did not significantly influence the results of commitment and trust. The finding that the subjectivity of the explicit content did not significantly alter commitment to the explicit meaning likely indicates that people still perceive speakers as committed to explicit content even when it is relatively subjective (e.g., ‘my ankle is hurting’). Accordingly, if commitment to the explicit meaning was not significantly affected by subjectivity, it is reasonable to assume that subjectivity will not exert a stronger downstream influence on commitment to the implicit meaning. Overall, then, the observed implicit dominance does not appear to stem from the fact that subjective content (e.g., ‘my ankle is hurting’) is less directly accessible without direct access to the speaker’s internal state. Instead, it may be shaped by the question under discussion or by the social associations attached to each state-of-affairs scenario, as we discuss in the General Discussion.
5. General discussion
In three experiments, we demonstrate an interdependency between commitment to the explicit meaning (‘what is said’ or the verbatim) and commitment to the implicit meaning (‘what is meant’ or the gist). Specifically, our results indicate that commitment to explicit and implicit meanings is influenced not only by the truth value of their respective contents, but also reflects a more holistic integration across meanings. This interdependency was not symmetric and reflected an ‘implicit dominance’ wherein the truth value of the implicit content exerted more influence on the processing of commitment. This pattern was observed across all three experiments, despite variation in the SoA provided, the wording of the commitment question, and whether the explicit content was subjective or objective. General trust was influenced by the SoA of both explicit and implicit content (as well as their interaction). Compared to commitment, we did not observe an implicit dominance for trust across the three experiments; rather, trust was equally affected by the truth value of both content types. Overall, across the three experiments, the effects on commitment (‘implicit dominance’) and trust (‘holistic integration’) were consistent, although their magnitude ranged from modest to moderate. In the following, we will discuss how these results are interpreted in the context of commitment and subsequently in relation to general trust.
5.1. Commitment
Our results clearly indicate that commitment to the explicit meaning and commitment to the implicit meaning are integrated: commitment ratings were not only based on the actual SoA of one stream of meaning, but were also affected by the other. This was true even despite the fact that the SoA of each stream of meaning is independent of the SoA of the other stream of meaning. Importantly, this interdependency was not symmetrical. Experiments 1–3 revealed that the SoA of the implicit content influenced commitment to the explicit meaning, in addition to its expected effect on commitment to the implicit meaning. However, the relationship was not reciprocal, as the SoA of the explicit content influenced commitment almost exclusively to the explicit meaning.
In Experiments 2 and 3, the pattern differed slightly while still preserving implicit dominance. Specifically, the interaction between the two SoAs showed that, for commitment to the implicit meaning, the truth value of the explicit content mattered only in a limited case: when it was the only content found to be voided. To illustrate, consider our running example (1) where Dan asks Josh, ‘Are we going to the gym?’ and Josh replies, ‘My ankle is hurting’. If Josh truly was unable to go to the gym, then whether his ankle was in fact hurting had a small effect on perceived commitment. However, if it turned out that Josh was not being truthful about his intention or ability to go to the gym, that is, if the implicit content was voided, this appeared to cross a threshold beyond which any additional falsehood had little further impact. In this sense, implicit dominance was reflected in the fact that only a falsehood in the implicit content was sufficient to push perceived commitment to the implicit meaning past a threshold, after which additional falsehoods no longer had a significant effect.
The pattern of implicit dominance suggests that the implicit meaning is more influential in our representation and attenuation of commitment. This notion aligns with recent evidence suggesting that the ‘gist’ of a message, represented in our case by the implicit meaning, is often better remembered, particularly regarding its truth value (Brainerd & Reyna, Reference Brainerd and Reyna1990; Langdon et al., Reference Langdon, Helgason, Qiu and Effron2024; Reyna et al., Reference Reyna, Corbin, Weldon and Brainerd2016; Siegel et al., Reference Siegel, Zehr, Bacovcin, Schofield and Schwarz2018).
What can explain this persistent asymmetry and the implicit dominance? We would like to consider the communicative goals between interlocutors. Consider for example, our running example (1). If it turns out that Josh’s ankle was not actually hurting (voiding the explicit statement), but that he genuinely did not plan to go to the gym (affirming the implicit meaning), this could be seen as a simple white lie. A white lie involves saying something false while preserving an implied truth, often to avoid conflict, protect feelings, or maintain social harmony (Bok, Reference Bok2011; Terkourafi, Reference Terkourafi and Meibauer2018). In this case, Josh’s explicit statement was false, but the implicit message, that he did not intend to go to the gym, remained true and aligned with the communicative goal. Such white lies are generally considered socially acceptable and even polite. By contrast, the reverse case, true explicit content but false implicit content, is more closely associated with plausible deniability (Mazzarella, Reference Mazzarella2023). For example, if Josh’s ankle is indeed hurting but he is later seen at the gym, he can deny having communicated the inference, claiming that he only stated that his ankle was hurting, without explicitly committing to not going. These differences likely map onto distinct social evaluations: politeness-driven falsehoods are likely to be perceived as acceptable, whereas plausible deniability (often characteristic of strategic or ‘political’ communication) is less socially acceptable and can be perceived as highly deceptive (Meibauer, Reference Meibauer2023; Reins & Wiegmann, Reference Reins and Wiegmann2021; Weissman, Reference Weissman2024; Wiegmann et al., Reference Wiegmann, Willemsen and Meibauer2022). Accordingly, only the latter case is expected to incur greater reputational costs, potentially reducing perceived commitment more broadly. Notably, following this line of reasoning, one might expect trust to erode in the latter case, yet this was not observed. One possibility is that these effects are restricted to commitment to the specific content under consideration and do not generalize to perceptions of the speaker’s overall trustworthiness. Alternatively, opposing forces may be at play in cases of ‘strategic deception’, where the implicit content is false but the explicit content is true. On the one hand, falsehoods concerning the QUD may be especially damaging to trust. On the other hand, this effect may be offset by the fact that the falsehood is conveyed implicitly rather than overtly stated (i.e., compared to white lies which are explicitly false). As a result, these opposing forces may cancel each other out when looked across white lies and strategic deception, yielding no net difference in overall trust. Additionally, because listeners assume cooperative intent (Grice, Reference Grice1975; Levinson, Reference Levinson1989), ‘My ankle is hurting’ is interpreted as implying ‘I’m not going’. Crucially, this inferred meaning directly addresses the question under discussion (Roberts, Reference Roberts2012), and is therefore likely to be perceived as the speaker’s primary communicative intent. Thus, any falsehood that pertains to the speaker’s primary communicative intent, or to the question under discussion, is likely to incur greater overall reputational harm. These two factors, separately and in combination, may help explain the implicit dominance pattern, whereby falsehood in the implicit content reduces perceived commitment more broadly, including commitment to the explicit content, despite its truth.
Finally, we note that because our stimuli were pre-tested to elicit strong implicatures, our findings reflect a ‘worst-case’ for violations of implicit content. If the same content yielded only a weak implicature, listeners might not reliably derive the hidden meaning, and any subsequent updating of perceived commitment would be smaller. Consistent with this, Weissman and Terkourafi (Reference Weissman and Terkourafi2019) found that false implicatures were often forgiven, whereas several strong generalized or particularized implicatures were more heavily rated as lies when discovered to be false. From a Gricean/Relevance perspective, strong implicatures make an utterance fully relevant and thus firmly commit the speaker, whereas weak implicatures leave room for alternative meanings and reduce harm to commitment (as seen in Braun et al., Reference Braun, Kuperwasser and Shetreet2024). Nevertheless, these studies did not directly examine the balance or interdependence between explicit and implicit content, and future work will be needed to determine whether this prediction holds in such cases.
5.2. Trust
Our results clearly demonstrate that both the explicit and implicit content affect our evaluations of speaker’s trust, as reflected in reliability ratings. This is evident in the way trust ratings vary based on the affirmation or negation of the SoA of both types of content. Furthermore, in all experiments, explicit and implicit content showed an interaction effect. The presence of an interaction suggests that trust judgments may be non-additive. Under this account, the first case in which either the implicit or explicit meaning proves false has a greater impact than subsequent falsehoods, such that the shift from zero to one falsehood is expected to be larger than the shift from one to two. This pattern suggests a threshold-like erosion of trust, rather than a simple additive accumulation of falsehoods. However, trust ratings did continue to decrease when both contents were false, and ratings fell most clearly below the scale midpoint only in this condition. Thus, rather than a strict ‘one strike and you’re out’ pattern, the results suggest diminishing marginal effects of additional falsehoods. Importantly, unlike the interaction observed for commitment, this pattern held regardless of whether the first falsehood concerned the explicit or the implicit content: no comparable implicit dominance emerged for trust, and either type of falsehood was sufficient to cross the threshold. One possible explanation for this threshold-like pattern comes from prior work suggesting that acceptance of information is often the cognitive default, whereas distrust emerges as a secondary process that overrides this default. Importantly, this framework treats trust and distrust as qualitatively distinct states, rather than as opposite ends of a single, linear continuum (Braun et al., Reference Braun, Yeshurun and Shetreet2025; Coliva, Reference Coliva2025; Levine, Reference Levine2019). From this perspective, a single clear falsehood may be sufficient to trigger a categorical shift from trust to distrust, after which additional inconsistencies exert a reduced marginal effect. However, because the QUD in the current experiment is more directly relevant to the implicit content, different forces may be at play, such as relevance to the QUD on the one hand and the negation of explicit content on the other. Future studies will therefore be needed to disentangle these forces and determine whether trust in interpersonal settings follows the same pattern.
Although both commitment and general trust are influenced by the explicit and implicit content, they appear to rely on distinct underlying representations. This is evident from the influence of all aspects of the SoA on general trust, but not on commitment. In other words, general trust appears to be more broadly sensitive than commitment, responding to any verification or violation of what was explicitly or implicitly said. To further explore the distinction between these two concepts, let us return to our running example. When Josh says ‘My ankle is hurting’ while feeling perfectly fine, we can interpret his choice of words as a white lie, such that Josh is conveying that he will not go to the gym by telling a lie to maintain social harmony. As we previously saw, in these cases, commitment to the implicit meaning remains relatively unaffected by the false explicit content, suggesting that a lie unrelated to that of the communication goal (e.g., whether Josh will go to the gym) can be overlooked. However, when assessing a speaker’s overall reliability, any breach of trustworthiness, i.e., any lie, regardless of its relevance, will factor in the evaluation (and possibly the content’s processing; see Braun et al., Reference Braun, Yeshurun and Shetreet2025), as shown by the decrease in trust ratings in such cases. These differences between commitment and trust may lie in the nature of general trust, which reflects a broader, more integrative view of an individual (Braun et al., Reference Braun, Kuperwasser and Shetreet2024).
Taken together, these findings underscore the nuanced relationship between explicit and implicit content in shaping trust evaluations, highlighting that while commitment remains anchored to the specific communicative goal, general trust reflects a more general assessment of a speaker’s reliability.
Acknowledgments
The authors would like to thank the members of the Language in Social Context laboratory.
Funding statement
This work was funded by the Israel Science Foundation (811/23) and the Mind and Language Center, Sagol School of Neuroscience, Tel Aviv University.
Competing interests
The authors declare none.
AI use
AI assistance was limited to grammar and phrasing edits only; no new ideas or conceptual content were generated.
Ethical standard
The experimental protocols were approved by the ethics committee at Tel Aviv University.
Open access
