Infants Assume Questions Serve an Information‐Seeking Function, Link Them to Interrogative Sentences and Differentiate Them From Assertions

ABSTRACT

Eye‐tracking studies tested the understanding of two types of speech acts (questions and assertions) in 14‐, 18‐, and 30‐month‐olds (N = 280; 149 females; ethnicity data collection forbidden, testing in 2021–2024). Experiments involved objects either hidden or visible for a speaker. By 14 months, when the speaker asked questions, infants focused on hidden objects (rs > 0.31). Infants linked novel labels in interrogative sentences to hidden objects by 18 months and novel labels in declarative sentences to visible objects by 14 months (ds > 0.52). Thus, infants assume questions seek information one is lacking, while assertions share information one has access to. Furthermore, infants connect interrogative sentences to questions and declarative sentences to assertions, showing an understanding of communicative form–function relations.

1. Introduction

Questions are fundamental in human communication and learning, present in all languages (Siemund 2001), and without equivalent in non‐human animals. They are key to expressing informational needs, from the mundane (“What is the weather forecast for Thursday?”) to the profound (“Is there a God?”), or the scientific (“How did Wiles prove Fermat's last theorem?”). Attempting to infer such specific needs solely through non‐verbal cues is nearly impossible. Yet, when learners articulate them through questions, it becomes much easier to address their informational requirements, vastly expanding the potential to communicate useful information.

For questions to fulfill their communicative roles, listeners must detect them and understand their information‐seeking function. This article investigates how infants address these challenges. Throughout the paper, we distinguish between interrogatives and questions. We assume that asking a question is a type of speech act used to request information, while an interrogative is a type of linguistic unit, typically ending with a question mark (van Elswyk 2025; Whitcomb 2017).

1.1. Infants' Efficient Information‐Seeking Behaviors

Infants' capacity to process the information‐seeking function of questions has not been studied in details. In contrast, substantial evidence suggests that infants engage in information‐seeking actions, sometimes using communication to elicit information from others. In individual learning contexts, infants consider whether information is available in their environment when planning information searches. Fourteen‐month‐old infants are faster at performing an action (opening a shutter) when it can yield relevant information (about a toy's location) than when it does not (Aguirre, Brun, Couderc, et al. 2022). In addition, toddlers prioritize seeking information about things they do not know (Poli et al. 2024). In short, infants and toddlers' information‐seeking strategies are modulated by whether information is available and whether it is already known. This behavioral flexibility is best explained by postulating that infants seek information in a purposeful manner, forming the goal of accessing information, and planning how to collect it.

Infants' information‐seeking behavior extend to social learning contexts. They actively look to others for information when uncertain (Striano et al. 2006; Vaish et al. 2011). By 12 months, these information‐seeking behaviours are selective: infants preferentially orient toward adults who might have expertise, or were more likely to help them learn, specifically when in need of gaining information (Bazhydai et al. 2020; Bazhydai and Harris 2021; Goupil et al. 2016; Stenberg 2009, 2013; Stenberg and Hagekull 2007). Moreover, between 1 and 2 years of age, children start to use behaviors that can be described as “proto‐interrogative” because they are (i) communicative, and (ii) geared toward eliciting relevant information from social partners (for reviews, see Begus and Southgate 2018; Harris et al. 2017; Lucca 2020; Ronfard et al. 2018). For instance, 12‐month‐old infants persist more in pointing when adults respond by providing them with novel rather than familiar information (Kovács et al. 2014). Sixteen‐month‐old infants also point more for informants that were accurate rather than inaccurate in the past (Begus and Southgate 2012). Thus, infants preferentially point for adults who provided novel or accurate information. In addition, 16‐ to 18‐month‐old children learn better information—labels or objects' functions—provided in response to their pointing gestures, suggesting that they use them to convey what they want to learn (Begus et al. 2014; Lucca and Wilbourn 2018, 2019). From 18 months onward, children also routinely use verbal questions to gain information (Chouinard et al. 2007; Nikolaus et al. 2022). In short, infants and toddlers use communication to elicit information from social partners, first appealing to non‐verbal cues such as pointing, and later, to verbal questions.

Evidently, infants' impressive information‐seeking abilities do not establish that they can process other people's questions. Achieving an action and representing it when performed by others tap into distinct abilities. For instance, showing that infants reach for objects does not demonstrate that they can represent the goal of reaching for objects in others (e.g., Woodward 1998). Thus, evidence for infants' efficient information‐seeking does not suffice, on its own, to establish that infants can recognize and process others' questions.

Question processing involves several key components: representing information search and requests for information in others (the mindreading challenge), connecting interrogatives to their information‐seeking function (the form‐function challenge), and processing the syntax and prosody of interrogatives. This paper focuses on infants' capacity to address the mindreading and form‐function challenges, as previous research has already explored the syntactic and prosodic elements of infants' question processing (Frota et al. 2014; Geffen and Mintz 2015; Moradlou et al. 2021; Nguyen and Legendre 2022; Perkins and Lidz 2021; Seidl et al. 2003).

1.2. The Mindreading Challenge: Representing the Information‐Seeking Function of Questions

Questions serve an information‐seeking function, with speakers using them to acquire novel information. In its most sophisticated forms, representing the information‐seeking function of questions might develop late. For instance, to start searching for a piece of information, learners must be aware of lacking it. Thus, fully representing information‐seeking requires tracking what others know they do not know—that is, forming complex representations of mental states embedded within mental states. Only adults have been found to appeal to such elaborate representations when processing questions, with children as old as 6 years of age showing no evidence of doing so (Aguirre, Brun, Reboul, et al. 2022). By contrast, in its elementary forms, listeners' sensitivity to the information‐seeking function question may involve relatively simple and early developing mental state representations, such as monitoring what others can see. Such simpler strategies are used by children and adults, who assume that a speaker's question pertains to objects hidden from the speaker, even if visible to themselves (Aguirre, Brun, Reboul, et al. 2022; Brown‐Schmidt 2009; Brown‐Schmidt et al. 2008).

The capacity to monitor an agent's visual perspective, evidenced in children's and adults' interpretation of questions, emerges during the first year of life. From 3 months of age, infants consider what agents can see when processing their actions (e.g., Choi et al. 2018; Luo and Baillargeon 2007; see also Kampis et al. 2015 for evidence about the neural processes supporting infants' representation of what is hidden from others). Soon after their first birthday, infants also consider what others have seen or experienced when communicating. Twelve‐month‐old infants point more when adults are ignorant than when they are knowledgeable (Liszkowski et al. 2007; O'Neill 1996). By 14 months, infants disambiguate ambiguous requests of an experimenter based on what she experienced (Liebal et al. 2009; Moll et al. 2008; Saylor and Ganea 2007). In short, infants track other people's visual perspectives early on; they consider that people's access to information about the world can be incomplete and apply this knowledge in communicative contexts.

Moreover, evidence suggests that outside of the communication domain, infants start to represent others' information‐seeking behaviors during the second year of life. At 14 months, infants look longer when a character takes a longer route to see something she has already seen than when she makes the same detour to gain novel information (Varga et al. 2021). Thus, infants assume that agents make efforts to collect new information—a key component of the capacity to process questions. In short, infants track what others can see from an early age, and they have some ability to represent information search by 14 months. Study 1 builds on these results to investigate whether infants are sensitive to the information‐seeking function of questions.

1.3. The Form‐Function Challenge: Connecting Interrogatives to Questions

Questions raise a second challenge. To process them, one needs to detect them. Linguistic cues can simplify this challenge. Information requests are often made using interrogatives (e.g., “Is it raining?” or “What is a platypus?”). Although cues that set interrogatives apart vary across languages, often, they are characterized by a specific syntax and prosody (Siemund 2001).

Thus, when a speaker uses an interrogative, it is a cue that she might intend to seek information. Importantly, interrogatives do not always convey genuine information‐seeking questions. They can also express, for instance, indirect requests (“Can you pass me the salt?”), threats (“Who do you think you are?”), test questions (e.g., a teacher to her pupil: “What is the capital of France?”) or advice (“Why don't you consult your doctor?”). Thus, the relation between interrogatives and questions is not systematic, and discovering it is far from trivial.

Yet, children show some sensitivity to the link between interrogatives and questions. When an adult uses interrogatives to ask where an object is, 2.5‐ to 6‐year‐old children assume that she refers to an object whose location she does not know (Aguirre, Brun, Reboul, et al. 2022; Nurmsoo and Bloom 2008). We build on these results to investigate infants' sensitivity to the relation between interrogatives and questions. Infants discriminate interrogatives from declaratives based on prosodic and syntactic cues, such as intonation by 6 months (Frota et al. 2014) or word order by 12 months (Geffen and Mintz 2015). Thus, infants detect interrogatives from an early age. Yet, it is unclear whether infants assign specific communicative functions to interrogatives, such as asking questions. Study 2 addresses this issue.

1.4. Experimental Rationale

Experiments were adapted from Aguirre, Brun, Reboul, et al. (2022). In the test condition of Study 1, two unfamiliar objects are moved while an actress is absent. Upon returning, the actress can see one object but not the other. She asks questions about the location of an object, using a novel label (e.g., “Where is the tralet?”). The test phase assesses which unfamiliar object participants mapped to the novel label by prompting them to look at the “tralet.” If infants are sensitive to the information‐seeking function of questions, they should map the novel label to the object hidden from the speaker. A control condition checks that infants' behaviors are not driven by the visual properties of the stimuli alone. In the control condition, while the actress asks about the location of the “tralet,” participants are prompted to look at the “lagui” during the test phase.

The youngest participants in Studies 1 and 2 were 14 months old. At this age, infants can learn novel labels after hearing them just a few times (Paulus and Fikkert 2014; Pomiechowska and Csibra 2022; Woodward et al. 1994). Eighteen‐month‐olds were also tested as this age marks the mastery of sophisticated communicative uses of questions (Chouinard et al. 2007; Perkins and Lidz 2021). Study 1 also includes 30‐month‐olds. By then, toddlers routinely use interrogative sentences to request information themselves (Chouinard et al. 2007; Ronfard et al. 2018). The studies used interrogative “where” questions about an object's location, one of the wh‐ questions children react to appropriately from an early age (Moradlou et al. 2021; Seidl et al. 2003). Data, stimuli samples, and analysis scripts of all studies are accessible in an open repository (URL: https://osf.io/4u2zr/?view_only=98e647d7f8a24e83b94c14a92ca5468e).

1.5. Ethical Considerations

The research reported in this manuscript was approved by an independent committee for biomedical research (CPP Sud‐Est II, IRB: 00009118). The parents of all participants gave their informed written consent before their inclusion in any of the studies.

2. Study 1

2.1. Method

2.1.1. Participants

In Study 1, 14‐, 18‐, and 30‐month‐olds were assigned randomly to the test or the control condition (N = 168, 91 females; n = 28 per condition for each age group). Sample sizes were set by conducting analyses with G*power (v.3.1; Faul et al. 2007), assuming effect sizes identical to those observed in comparable studies (d = 0.67; Aguirre, Brun, Reboul, et al. 2022). These analyses revealed that a sample size of 28 participants per condition was sufficient to achieve a power of 0.90 (α = 0.05) when assessing performance during the test phase against chance with two‐tailed t‐tests.

Demographic information and exclusions are detailed in Tables S1 and S2. Data collection occurred in France, where collecting information on ethnicity is forbidden. Recruitment procedures and recruitment pool were the same for all studies (cf. the Supporting Information). Data collection happened between June 2021 and March 2024, from Monday to Saturday, between 8:30 and 5:30 p.m.

2.1.2. Materials and Procedure

2.1.2.1. Apparatus

In all studies, the participants sat on their caregivers' lap in a soundproof booth, approximately 60 cm away from the eye tracker's monitor on which the stimuli were presented (Tobii Pro Spectrum 150, sampling frequency: 60 Hz, screen diagonal: 23.8″, resolution: 1920 × 1080 pixels). Caregivers were instructed to close their eyes during the whole experiment. Psychopy v.3.0.4 (Peirce et al. 2019), Python v.2.7, and Tobii Pro SDK package v.1.7 were used for stimuli presentation and data collection. A five‐point calibration method, repeated until complete, was used. The participants typically required only one calibration. The experiment started after the calibration phase. Before each experimental video, a colorful transition animation showing rotating arrows was played at the center of the screen. Transition animations were played until the participants looked at the screen for at least 1 s.

2.1.2.2. Familiarization Phase

Study 1's participants first watched twice a 50 s familiarization video (Video S1). The beginning of the familiarization video showed four empty opaque cardboard boxes on a table: two larger central boxes, and two smaller peripheral boxes. The boxes had no frontside—the participants could see their contents at all times. One of the peripheral boxes had an opaque backside, making it impossible to see its contents for anyone facing the participant across the table. The other three boxes had no backsides.

An actress entered the screen from behind curtains while holding two unfamiliar objects (a purple wooden structure and a red plastic object). She placed one unfamiliar object in each of the two central boxes (Figure 1A). Next, the actress left behind the curtains. During her absence, a hand moved the objects to the peripheral boxes (Figure 1B). Each object was moved from the central box where it was initially located to the outer box on the same side of the table. Next, the actress returned from behind the curtains. She pretended to be searching for an object (first, looking toward the center of the table, second, looking across the entire table from side to side, and finally, looking back toward the center of the table; Figure 1C,D). The actress could not see the object in the box with the opaque backside—the hidden object—whereas she could see the other one—the visible object. Next, the actress looked toward the center of the table, flipped her palms up and out to the side, and asked a series of questions using the nonsense word “tralet”: “Where is the tralet? … Where did the tralet go? … Where is the tralet?” (Figure 1E). Since the actress used an unfamiliar label, her question was ambiguous. Yet, it was possible to resolve the question's ambiguity by assuming that the actress sought the object she could not see.

FIGURE 1.

Key events from the familiarization and test phases of Study 1. The arrows, text, and speech bubbles shown in the figure serve illustration purposes. They did not appear in the videos shown to the participants.

During the familiarization phase, the following factors were counterbalanced across participants: the side of each unfamiliar object (right or left); the side of the box with the opaque backside (right or left); the side toward which the actress looked first when searching for the object across the table (right or left); and the identity of the hidden object (purple or red object).

2.1.2.3. Test Phase

After watching the familiarization video twice, the participants saw four test sequences. In each of them, the two unfamiliar objects appeared, each on a different side of the screen. A voice‐over uttered prompt sentences containing an unfamiliar label to invite participants to look at one object. The label used in the prompts was the same as in the familiarization phase in the test condition (“tralet”; Figure 1F) and a different one in the control condition (“lagui”; Figure 1G). There was a 4 s silence period after each prompt. The objects switched sides between sequences. The prompts were: “Where is the [novel label]” and “Did you see the [novel label]?” alternating across the four sequences.

During the test phase, the identity of the hidden object (purple or red object) was counterbalanced across subjects. The side of the screen on which each unfamiliar object appeared across test sequences (right or left) was counterbalanced within subjects.

2.1.3. Preprocessing and Coding

2.1.3.1. Test Phase

Data were preprocessed using R (v.4.1.0) and the package eyetrackingR (v.0.1.8). Planned analyses focused on data from the test phase, in the 367–2000 ms window after the unfamiliar label's onset, during the silence following prompt questions (e.g., “Did you see the [novel label]?”), in line with Aguirre, Brun, Reboul, et al. (2022). Using this time window is standard in studies of infant word learning (Swingley and Aslin 2000). It accounts for label processing and eye movement initiation (367 ms onset), while providing sufficient response time (2000 ms endpoint). Furthermore, eye movements beyond 2000 ms post word onset are less likely to reflect the processing of the unfamiliar label (Swingley and Fernald 2002).

To analyze the eye‐tracking data from the test phase, the screen was divided centrally into two rectangular areas of interest (AOIs) of 960 × 1080 pixels (i.e., the left and right sides of the entire screen). For each participant, we computed the proportion of looking time to the hidden object for each 20 ms time interval by summing up the time spent looking at the AOI occupied by the hidden object and dividing it by the total time spent looking at the two AOIs over that time bin. Following this, for each participant, we computed the average of these proportions across all test sequences for each time bin, and then, across all time bins. We excluded the data from test sequences in which we missed more than 50% of the data. Participants who contributed data for less than two test sequences were excluded from the analysis. Each participant retained in the analysis contributed an average of 3.71 (SD = 0.59) out of 4 test sequences.

2.1.3.2. Familiarization Phase

In addition to planned analyses performed on data from the test phase, exploratory analyses examined participants' gaze responses during the familiarization videos. These analyses investigated if participants identified the referent of the novel label while the actress uttered questions during the familiarization phase. They focused on two time periods. First, a baseline measure assessed participants' tendency to look at each object before any question from the experimenter. This baseline measure was collected for 1500 ms while the actress was absent, starting from the moment the hand displacing the objects inside boxes disappeared from the screen. Second, a post‐labeling measure assessed participants' tendency to look at the hidden object when the actress asked questions. This post‐labeling measure was collected between 367 and 2000 ms after the onset of the unfamiliar label, thus, right after the actress asked questions (e.g., “Where is the tralet?”).

The preprocessing procedures for these analyses mirrored those applied to the test phase data, with the following exceptions. Data from Study 1's familiarization phase were pooled across the test and control conditions, given familiarization videos were identical in both. Two rectangular AOIs of 457 × 514 pixels centered on the unfamiliar objects were used. We excluded the data from baseline and post‐labeling familiarization sequences if (i) more than 50% of the data was missing, or (ii) infants did not look for at least 100 ms at one of the AOI (Wass et al. 2013). Participants were included if they provided data for at least one post‐labeling sequence. Included participants (n = 127) contributed, on average, 1.75 baseline sequences out of 2 (SD = 0.43), and 2.41 post‐labeling sequences out of 6 (SD = 1.42). Nine included participants did not contribute any data in the baseline phase, and did not enter cross‐phases comparison analyses (baseline vs. post‐labeling).

2.1.4. Analyses

Statistical procedures were the same for all studies. Unless specified otherwise, the analyses reported in the main paper were planned. Significance tests were two‐tailed. Analyses were implemented using R (v.4.1.0) on Rstudio (v.1.4.1717) with the following packages: afex (v.1.0.1), bruceR (v.0.8.3), dplyr (v.1.1.0), emmeans (v.1.7.2), rcompanion (v.2.4.1), rstatix (v.0.7.0), tidyr (v.1.3.0), and tidyverse (v.2.0.0). Data from the familiarization phase were analyzed using nonparametric tests because they departed from normality.

2.2. Results

2.2.1. Test Phase

First, data from Study 1's test phase were analyzed by running a full‐factorial ANOVA on mean proportions of looking time to the hidden object with condition (control vs. test) and Age (14‐, 18‐ and 30‐month‐olds) as between‐subject factors. The ANOVA revealed a statistically significant main effect of condition (F _1,162 = 10.71, p = 0.001, ${\eta }_{\mathrm{p}}^2$ = 0.062, 95% CI [0.01, 0.14]). Thus, in Study 1's test phase, participants' mean proportion of looking time to the hidden object was higher in the test than in the control condition. The ANOVA also showed a statistically significant interaction between age and condition, indicating that the effect of condition differed across age groups (F _2,162 = 3.25, p = 0.041, ${\eta }_{\mathrm{p}}^2$ = 0.04, 95% CI [0, 0.10]). The ANOVA revealed no other statistically significant effect. Subsequently, the data of each age group were analyzed separately.

Condition (test vs. control) had no significant effect on 14‐month‐olds' average proportion of looking time to the hidden object in Study 1's test phase (t(50.96) = 0.03, p = 0.974, d = −0.01, 95% CI [−0.52, 0.54], Welch two‐sample t‐test). As Figure 2 shows, 14‐month‐olds' average proportion of looking time to the hidden object did not differ from the value predicted by chance in any of the conditions (detailed statistics are reported in Table S3). By contrast, 18‐ and 30‐month‐olds' average proportions of looking time to the hidden object were significantly higher in the test condition than in the control condition (18‐month‐olds: t(53.83) = 3.30, p = 0.002, d = −0.88, 95% CI [−1.48, −0.39]; 30‐month‐olds: t(53.61) = 2.26, p = 0.028, d = −0.61, 95% CI [−1.29, −0.07], Welch two‐sample t‐tests). At 18 and 30 months, participants' average proportion of looking time to the hidden object was higher than predicted by chance specifically in the test condition, not in the control condition (Figure 2, Table S3).

FIGURE 2.

Proportion of looking time to the hidden object per condition and age group in Study 1's test phase. The dotted black line represents chance (0.5). Red dots and error bars indicate means and standard errors. Box plots represent data quartiles. Gray dots represent individual data points, and colored areas represent data density. Comparisons to chance level by one‐sample t‐tests, and comparisons between conditions by Welch two‐sample t‐tests. *p < 0.05, **p < 0.01, ns: not significant.

2.2.2. Familiarization Phase

In Study 1's familiarization phase, the median proportions of looking time at the hidden object exceeded chance only post‐labeling, not during baseline, in all groups (Figure 3, see detailed statistics in Table S4). Moreover, participants' median proportion of looking times at the hidden object increased significantly post‐labeling compared to baseline, in all groups (14‐month‐olds: W = 130, p = 0.011, r = 0.44, 95% CI [0.12, 0.72]; 18‐month‐olds: W = 245, p = 0.015, r = 0.38, 95% CI [0.08, 0.63]; 30‐month‐olds: W = 299, p = 0.013, r = 0.37, 95% CI [0.09, 0.58], Wilcoxon tests for matched pairs). Thus, during Study 1's familiarization phase, participants had no baseline preference for looking at any of the two unfamiliar objects. However, when the actress asked about an unknown object's location, participants looked more at the object hidden from her, than at an object she could see.

FIGURE 3.

Proportion of looking time to the hidden object per phase and age group in Study 1's familiarization phase. The dotted black line represents chance (0.5). Red dots and error bars indicate means and standard errors. Box plots represent data quartiles. Gray dots represent individual data points, and colored areas represent data density. Comparisons to chance levels by one‐sample Wilcoxon tests, and comparisons between measurement phases by Wilcoxon tests for matched pairs. *p < 0.05, **p < 0.01, ns: not significant.

2.3. Discussion

During Study 1's familiarization phase, participants showed no baseline preference for either visible or hidden objects. In contrast, when an unfamiliar word was used in a question, 14‐, 18‐ and 30‐month‐old participants looked more at the hidden than at the visible object, demonstrating sensitivity to the information‐seeking function of questions. By 18 months, infants had also memorized the referent of the unfamiliar label embedded within the question. In the test condition of Study 1, 18‐ and 30‐month‐olds mapped the novel label to the hidden rather than the visible object. Although we did not collect a baseline measurement before Study 1's test phase, the contrast between test and control conditions provides compelling evidence that participants' gaze patterns were influenced by the testing label, as this was the sole variable differing between conditions. Hence, participants' behavior in Study 1's test condition cannot be explained by mere visual properties of stimuli.

Moreover, our results suggest that infants processed the speaker's questions as genuinely information‐seeking. Speakers may use interrogative sentences to request information they already possess, for example, asking test questions to verify knowledge or pedagogical questions to teach (Bascandziev et al. 2025; Grosse and Tomasello 2012; Yu et al. 2019). It is unlikely that infants interpreted interrogative sentences this way in our experiment. Had they done so, participants would have directed their gaze to the visible object following the onset of the label—the exact opposite of what we found. In short, in Study 1, infants assumed that questions requested information that the speaker did not have access to by 14 months. By 18 months, participants also relied on this assumption to process the novel label embedded within questions and linked it to the object hidden from the speaker.

In Study 1's control condition, participants could have potentially shown a “mutual exclusivity” response (Markman and Wachtel 1988), associating “tralet” with the hidden object during familiarization, and excluding this object as referent for “lagui” during the test phase. We discuss in the Supporting Information why this did not happen, although 18‐month‐olds showed a non‐significant trend in this direction (p = 0.076). In short, Study 1's control condition was not designed to elicit a mutual exclusivity response and did not include many features contributing to this effect (cf. Supporting Information).

Study 1 indicates that infants understand the information‐seeking function of questions. Yet, it does not tell whether infants specifically link interrogative sentences to questions. For instance, in Study 1, it is possible that infants would have processed any sentence as a request for information, irrespective of whether it was interrogative. Study 2 addressed this issue by testing infants' sensitivity to form‐function mapping between interrogatives and questions.

In Study 2, an actress left two unfamiliar objects in a room and exited. The unfamiliar objects were displaced, so only one was visible to the actress upon return. After coming back, the actress uttered either interrogative sentences in the question condition (e.g., “Where is the mogui?”) or declarative sentences in the assertion condition (e.g., “There is the mogui.”). If infants link interrogative sentences to questions, they should be more likely to map the novel label to the object hidden from the actress in the question condition than in the assertion condition. Conversely, if infants expect declarative sentences to convey assertions, they should map the novel label to the object visible to the speaker in the assertion condition.

Study 2 also introduced changes aiming at reducing task demands. First, it started with a warm‐up phase familiarizing participants with the novel label. Second, the unfamiliar object appeared in the same visual context and position across Study 2's familiarization and test phases. Study 1's materials had been validated previously (Aguirre, Brun, Reboul, et al. 2022). A study ran on a sample of adults (n = 40) validated Study 2's material. It established that adults show the expected response pattern when processing Study 2's videos (cf. Supporting Information).

3. Study 2

3.1. Method

3.1.1. Participants

Fourteen‐ and eighteen‐month‐olds were enrolled in Study 2 (N = 112, 58 females; n = 28 per condition for each age group). Study 2 relied on the same recruitment, inclusion, and exclusion procedures as Study 1. Demographic information and exclusions are detailed in Tables S1 and S2.

3.1.2. Materials and Procedure

3.1.2.1. Apparatus

Study 2's apparatus was the same as in Study 1.

3.1.2.2. Warm‐Up Phase

In Study 2, the participants first watched twice a 26 s “warm‐up video” designed to help participants encode and memorize the novel label (Figure 4A, Video S2). At the start of the warm‐up video, infants saw a table with curtains behind it. An actress emerged from behind the curtains, greeted the participant, and said: “Hi! How are you? I brought some toys today. I brought a [novel label, e.g., mogui]. We're going to have lots of fun with the [mogui]. This [mogui] is great! I love [moguis] because [moguis] are so cool. Do you want to see a [mogui]?”. The label warm‐up video was played twice. Next, the familiarization phase started.

FIGURE 4.

Key events from the warm‐up, familiarization, and test phases of Study 2. The arrows, text, and speech bubbles shown in the figure serve illustration purposes. They did not appear in the videos shown to the participants.

3.1.2.3. Familiarization Phase

Study 2's familiarization phase differed across conditions. In both conditions, the participants watched twice a 63 s “familiarization video” (Video S3). At the beginning of the familiarization video, there were two empty opaque cardboard boxes placed side‐by‐side on a table, with one having a backside, and the other not. Both boxes had no frontside, thus, their contents were visible to the participant at all times. The same actress that appeared during the warm‐up phase then emerged from behind curtains carrying two unfamiliar objects—the same ones as in Study 1. She placed each object on top of a box and left (Figure 4B). While she was away, a hand moved the objects from the top to the inside of the boxes (Figure 4C). Next, the actress returned and looked toward the top of the boxes, before lowering herself slightly. From her viewpoint, the actress could see the object inside the box without a backside—the visible object. She could not see the other object—the hidden object (Figure 4D). Next, the actress uttered sentences using the novel label introduced during the warm‐up phase. Utterances differed across conditions. In the question condition, the actress asked three questions: “Where is the [novel label]? … Where is it? … Where is the [novel label]?” (Figure 4E). Her gaze was directed toward the participant before each sentence, and toward the midline between the boxes during the utterance—providing no cues about the object the novel label referred to. In Study 2's question condition, the unfamiliar label could be disambiguated by assuming that the actress referred to the object she could not see—the hidden object.

Study 2's assertion condition was identical to Study 2's question condition, except that during the familiarization phase, the actress uttered declarative sentences instead of interrogative ones: “There is the [novel label] … There it is … There is the [novel label]” (Figure 4F, Video S4). Thus, in Study 2's assertion condition, infants could resolve the ambiguity of the actress' utterances by assuming that she referred to the visible object—whose location was visually accessible to her. Study 2's familiarization phase was followed by a test phase that was identical for both conditions.

3.1.2.4. Test Phase

During Study 2's test phase, the participants saw two 20 s test videos (Video S5). Each test video showed the two unfamiliar objects inside the boxes, in the same position as at the end of the familiarization video, while the actress was absent. Test videos started with a 2.5 s still frame showing the whole scene with the same camera angle as in the familiarization phase. Next, the camera zoomed in on the unfamiliar objects (2 s), the video froze, and there was a 3 s baseline measurement phase. Next, a voice‐over uttered two prompt sentences: “Look! A [novel label].” and “Did you see the [novel label]?” (Figure 4G). The post‐labeling measurement phase began after each prompt sentence, 367 ms after the onset of the novel label. Each prompt was followed by 4 s of silence. The participants viewed two consecutive test videos, resulting in two baseline measures (one at the start of each video) and four post‐labeling measures (one after each prompt utterance).

In Study 2, the following factors were counterbalanced across participants: the identity of the visible object (purple vs. red object); the side of the table it occupied (right vs. left)—the identity and side of the hidden object varied accordingly; the specific unfamiliar label used by the actress (“tralet” vs. “mogui”). For each given participant, the visible object's identity, the side of the screen it occupied, and the unfamiliar label used by the actress were kept constant throughout the experiment.

3.1.3. Preprocessing and Coding

3.1.3.1. Test Phase

In Study 2, pre‐processing was just like in Study 1, except that data were also collected during the 3 s baseline measurement phase. Thus, in Study 2's test phase, baseline data were collected immediately before the test to control for visual preferences unrelated to the processing of the novel label. As in Study 1, data from the post‐labeling phase were collected during the silence period following each prompt utterance (e.g., “Look! A [novel label].”), from 367 to 2000 ms after the unfamiliar label onset.

Post‐labeling and baseline sequences in which more than 50% of the data were missing were excluded from the analysis. Participants who contributed data for less than two post‐labeling sequences were excluded from the study. Moreover, 6 participants (out of 112) did not contribute any data during the baseline phase (4 in the question condition and 2 in the assertion condition). These participants did not enter the analyses comparing performance across phases (post‐labeling vs. baseline).

In the question condition, each participant retained in the analysis contributed an average of 3.29 (SD = 0.82) out of 4 post‐labeling sequences and an average of 1.90 (SD = 0.30) out of 2 baseline sequences. In the assertion condition, each participant retained in the analysis contributed an average of 3.18 (SD = 0.83) out of 4 post‐labeling sequences and an average of 1.78 (SD = 0.42) out of 2 baseline sequences.

3.1.3.2. Familiarization Phase

As in Study 1, exploratory analyses examined participants' gaze responses during the familiarization phase of Study 2. Note, however, that our studies were not specifically designed to collect data during the familiarization phase. In Study 2, when the actress was present, her body could be seen through the box without backside. This feature created an incentive for infants to look toward the area around the visible object, because of the slight movements of the actress's body. As a result, data from Study 2's familiarization phase were not informative about infants' capacity to process the referent of the novel label. For completeness, we nonetheless provide these data in the Supporting Information.

3.2. Results

To test whether Study 2's participants reacted differently to questions and assertions, a full‐factorial ANOVA was run on mean proportions of looking time to the hidden object during the test videos. Phase (baseline vs. post‐labeling phase) entered this analysis as a within‐subject factor, and condition (question vs. assertion) and age (14‐ vs. 18‐month‐olds) as between‐subjects factors. This analysis revealed a statistically significant main effect of condition, indicating that infants' proportion of looking time to the hidden object was higher in the question condition than in the assertion condition (F _1,102 = 7.77, p = 0.006, ${\eta }_{\mathrm{p}}^2$ = 0.07, 95% CI = [0.01, 0.18]). The ANOVA also revealed an interaction between condition and phase (F _1,102 = 17.55, p < 0.001, ${\eta }_{\mathrm{p}}^2$ = 0.15, 95% CI = [0.04, 0.27]). Thus, the effect of phase (baseline vs. post‐labeling) on the mean proportion of looking time to the hidden object varied depending on whether the speaker used interrogative or declarative sentences. The ANOVA showed no other statistically significant effect. Next, we analyzed the performance of participants enrolled in each condition per age group.

In Study 2's question condition, the average proportion of looking time to the hidden object was higher post‐labeling than at baseline, only in 18‐month‐old participants (14‐month‐olds: t(23) = −1.47, p = 0.156, d = −0.30, 95% CI [−0.74, 0.06]; 18‐month‐olds: t(27) = −2.70, p = 0.012, d = −0.51, 95% CI [−1, −0.15], t‐tests for matched pairs). Fourteen‐month‐olds' average proportion of looking time to the hidden object did not exceed the chance level in either the baseline or the post‐labeling phases of Study 2's question condition (Figure 5, Table S3). By contrast, 18‐month‐olds' average proportion of looking time to the hidden object was higher than predicted by chance specifically during the post‐labeling phase and not during the baseline phase of Study 2's question condition (Figure 5, Table S3). In short, in Study 2, 18‐month‐old participants linked the novel label embedded in questions to the object hidden from the speaker, whereas 14‐month‐olds' infants did not do so, just like in Study 1.

FIGURE 5.

Proportion of looking time to the hidden object per phase, condition, and age group in Study 2. The dotted black line represents chance (0.5). Red dots and error bars indicate means and standard errors. Box plots show data quartiles. Gray dots show individual data points, and colored areas show data density. Comparisons to chance level are made by one‐sample t‐tests and comparisons between conditions are made by Welch two‐sample t‐tests. *p < 0.05, **p < 0.01, ns, not significant.

Results from Study 2's assertion condition indicate that participants recognized the knowledge‐sharing function of assertion, by 14 months. In Study 2's assertion condition, the average proportion of looking time to the hidden object was significantly lower during post‐labeling than at baseline, for both age groups (14‐month‐olds: t(25) = 2.17, p = 0.040, d = 0.43, 95% CI [−0.05, 0.88]; 18‐month‐olds: t(27) = 2.33, p = 0.028, d = 0.44, 95% CI [0.05, 0.83], t‐test for matched pairs). During the post‐labeling phase of Study 2's assertion condition, the average proportion of looking time to the hidden object was significantly below the value predicted by chance for both age groups (Figure 5, Table S3). Thus, participants from both age groups looked significantly longer toward the visible object than toward the hidden object during the post‐labeling phase of Study 2's assertion condition. This pattern was not observed during the baseline phase, indicating a phase‐specific effect (Figure 5, Table S3).

Exploratory analyses assessed whether participants discriminated assertions from questions in Study 2, looking at each age group separately. These analyses primarily aimed to determine if 14‐month‐olds, despite chance‐level performance in the question condition's post‐labeling phase, still distinguished questions from assertions. To this end, we first examined whether the impact of test phase (baseline vs. post‐labeling) differed between conditions (assertion vs. question).

As expected, during the baseline phase, condition had no significant effect on mean proportion of looking time at the hidden object, for any age group (14‐month‐olds: t(45.02) = 0.49, p = 0.625, d = 0.14, 95% CI [−0.41, 0.78]; 18‐month‐olds: t(53.89) = −0.48, p = 0.633, d = −0.13, 95% CI [−0.65, 0.44], Welch two‐sample t‐tests). By contrast, during the post‐labeling phase, participants' mean proportion of looking time at the hidden object was significantly higher in the question condition than in the assertion condition, for both age groups (14‐month‐olds: t(53.10) = −2.02, p = 0.049, d = −0.54, 95% CI [−1.26, −0.06]; 18‐month‐olds: t(53.46) = −4.35, p < 0.001, d = −1.16, 95% CI [−1.77, −0.64], Welch two‐sample t‐tests). Thus, direct cross‐condition comparisons confirmed that participants discriminated questions from assertions. A second set of exploratory analyses verified this conclusion. These analyses showed a significant interaction between the effects of phase (baseline vs. post‐labeling) and condition (question vs. assertion) on mean proportion of looking time at the hidden object, in each of the age groups (see details in the Supporting Information).

In Study 2, 14‐month‐olds linked the novel label to the visible object in the assertion condition, while there was no evidence that they linked the novel label to the hidden object in the question condition. Thus, an exploratory analysis investigated whether infants were significantly better at processing assertions than at processing questions. For this analysis, we recoded data to focus on infants' proportion of look toward the target object (the hidden object in the question condition, and the visible object in the assertion condition).

Two full‐factorial ANOVAs were run on mean proportions of looking time to the target object during Study 2's test phase (one per age group). Phase (baseline vs. post‐labeling phase) entered these analyses as a within‐subject factor and condition (question vs. assertion) as a between‐subjects factor. Both ANOVAs revealed a significant main effect of the phase, indicating that participants' mean proportion of looking time at the target object was significantly higher post‐labeling than at baseline (14‐month‐olds: F _1,48 = 6.18, p = 0.016, ${\eta }_{\mathrm{p}}^2$ = 0.11, 95% CI [0.00, 0.29]; 18‐month‐olds: F _1,54 = 12.63, p < 0.001, ${\eta }_{\mathrm{p}}^2$ = 0.19, 95% CI [0.04, 0.36]). These ANOVAs revealed no other significant effect. Of note, there were no significant interactions between condition and phase (14‐month‐olds: F _1,48 = 0.01, p = 0.944, ${\eta }_{\mathrm{p}}^2$ < 0.001, 95% CI [0.00, 0.03]; 18‐month‐olds: F _1,54 = 0.08, p = 0.776, ${\eta }_{\mathrm{p}}^2$ = 0.002, 95% CI [0.00, 0.07]). Thus, for both age groups, there was no evidence that infants performed better when processing questions, than when processing assertions.

4. General Discussion

We investigated the early development of the capacity to process the communicative functions of questions and assertions. In Studies 1 and 2, two unfamiliar objects were displaced in the absence of an actress. When the actress returned, she could see only one of the objects, not the other. She used a novel label embedded in questions or assertions. Infants could resolve the ambiguity of the novel label by considering what the speaker could see and the communicative functions of questions and assertions. Questions express a request for information, likely referring to the object hidden from the actress. Assertions imply that the speaker has sufficient information to warrant what she claims, likely referring to the object visible to the actress.

In the familiarization phase of Study 1, when the speaker asked questions about an object's location, infants focused on objects hidden rather than visible for the speaker, by 14 months. Moreover, by 18 months, infants subsequently linked the novel label embedded in questions to the object hidden from the speaker (Studies 1 and 2, test phase). When the speaker made assertions about an object's location, infants subsequently linked the novel label to the object visible to the speaker, by 14 months (Study 2, test phase). These results have implications for the development of theory of mind, information‐seeking, linguistic abilities, and trust in communication.

4.1. The Communicative Function of Questions

We found that infants represent the information‐seeking function of questions by 14 months. In Study 1's familiarization phase, 14‐month‐old infants demonstrated the ability to determine that questions relate to information that a speaker was lacking. Importantly, this competence was demonstrated when infants' ability was probed online, when the speaker asked her questions. Moreover, we found that the ability to process labels embedded in questions improves during the second year of life. Results from the test phases of Studies 1 and 2 indicate that infants memorize new labels embedded in questions by 18 months (but not before).

These results provide information about infants' capacity to represent information‐seeking in others. First, they indicate that infants assume a speaker is unlikely to search for information that is currently available to her. Second, they show that infants represent other people's information‐seeking in a non‐egocentric way, processing questions by considering what might be cognitively useful for someone else (in this case, information about an object's location). Third, they suggest that infants can represent other people's communicative requests for information, not just other people's information‐seeking behaviors. Doing so requires acknowledging that people might use communication as a means to request information. Furthermore, we found that the repertoire of core communicative functions that infants can represent extends beyond questions and also includes assertions.

4.2. The Communicative Function of Assertions

In the assertion condition of Study 2, a speaker made a statement about an object's location. Fourteen‐ and 18‐month‐olds assumed that the speaker referred to an object she could see rather than an object hidden from her. Thus, 14‐ and 18‐month‐old infants demonstrated sensitivity to the knowledge‐sharing function of assertions. When uttering assertions, speakers present themselves as sharing reliable information, grounded in sufficient evidence (Kelp and Simion 2021; Turri 2016). Such a presumption of reliability is not always fulfilled (informants may lie or make mistakes). Yet, it shapes the way assertions are processed. For instance, adults spontaneously interpret bare assertions as grounded in first‐hand evidence (Degen et al. 2019; Mahr and Csibra 2021). In Study 2, we observe a precursor of this presumption of reliability, as infants assumed that a speaker's assertions referred to what she could see. Study 2 also revealed infants' capacity to link specific linguistic forms (such as interrogatives and declaratives) to their prototypical communicative functions.

4.3. Linking Linguistic Forms to Communicative Functions

Infants distinguish interrogatives from declaratives based on their syntactic or prosodic properties (Frota et al. 2014; Geffen and Mintz 2015). Studies 1 and 2 indicate that infants assign different communicative functions to these sentence types and connect them to speakers' perceptual access. In Study 2, infants linked declarative sentences and assertions (by 14 months) and interrogative sentences and questions (by 18 months). Importantly, our study did not require infants to possess comprehensive full‐blown grammatical knowledge of declaratives and interrogatives. More simply, in our study, infants needed to distinguish these sentence types and link them to their prototypical functions.

Moreover, questions can be differentiated from assertions on numerous bases, including prosodic, syntactic, and non‐verbal cues (e.g., the palm‐up gesture used in Study 1, which typically expresses ignorance or request, see Esteve‐Gibert et al. 2017; Harris et al. 2017). Our studies deliberately used several of these cues to give participants as much redundant information as possible. Future research should assess which specific cues elicit infants' expectations about questions and assertions. Moreover, our studies used a limited set of utterances (wh‐ questions and statements, all about locations, in a single language). Thus, investigating whether our results generalize to a broader set of utterance types across diverse languages is an important question for future research.

Linking interrogatives and declaratives to their prototypical functions requires addressing substantive challenges. First, the specific features of interrogatives and declaratives vary cross‐linguistically—infants have to learn them. Second, while interrogatives are often used to ask questions and declaratives to make assertions, they are also used to achieve many other types of communicative goals. For instance, child‐directed speech is rife with interrogatives that are not genuine information‐seeking questions, such as pedagogical questions (Bascandziev et al. 2025; Yu et al. 2019). Despite these challenges, we find that infants manage to use linguistic cues to discriminate questions from assertions. Third, the attitudes conveyed by asking a question or making an assertion are fully abstract—they cannot be reduced to a spatio‐temporal vocabulary—and they are not directly observable—they involve sophisticated intentions that have to be inferred. We find that infants manage to link such abstract, unobservable intentions to linguistic forms.

4.4. Communicative Functions and Trust

Studies 1 and 2 also impact our understanding of children's trust in communication, a central source of information in human learning (Sperber et al. 2010). Young children have a powerful disposition to rely on communicated information from a single, confident individual (Aguirre et al. 2023; Jaswal 2010; Mascaro and Morin 2011, 2015; Mascaro and Sperber 2009, 2019; Vanderbilt et al. 2011). For instance, 24‐month‐olds trust an informant over their memory when searching for a reward, even if the informant is ill‐informed (Mascaro and Kovács 2022). By preschool age, communicative actions like pointing elicit in children the presumption that the person using them is well‐informed. This presumption of competence can hinder children's ability to identify who possesses relevant knowledge (Palmquist and Jaswal 2012).

Our data, however, indicate that even during infancy, communication does not convey a blanket presumption of competence. When processing questions, infants recognized that informants might request information they lack. When processing assertions, infants recognized that individuals are likelier to convey information they have direct access to rather than information beyond their perceptual reach. Thus, even infants demonstrate a nuanced understanding of the contexts in which communication conveys a presumption of competence.

4.5. Infants Use Speakers' Visual Access to Disambiguate Spoken Language and Learn Novel Words

Our studies also indicate that infants consider speakers' visual access when processing language and learning labels. Previous research has shown infants' ability to learn novel labels embedded in carrier sentences, typically using assertions or a mix of speech acts. Studies 1 and 2 indicate that by 18 months, infants can learn novel labels embedded in questions. Furthermore, children consider speakers' visual access when learning novel labels from 2.5 years onwards (Nurmsoo and Bloom 2008). We find this ability has origins during infancy, by 14 months. Thus, at an age of rapid vocabulary expansion, infants use information on what is visible to speakers when disambiguating spoken language and learning the meaning of labels.

4.6. Concluding Remarks

This research explored the early ontogeny of the understanding of questions and assertions. It reveals that infants recognize questions as information‐seeking requests about what a speaker lacks knowledge of, and assertions as knowledge‐sharing statements related to what a speaker can see. In addition, we observed that infants link core types of speech acts to specific linguistic forms, despite formidable learning challenges. These findings indicate that infants possess a sophisticated, non‐egocentric understanding of communication, considering a speaker's knowledge and visual access, that informs their language processing and word learning. These results also open many questions about the scope, nature, development, and human specificity, of the capacity to represent speech acts. They open key avenues for future research about infants' capacity to represent other communicative acts (e.g., commanding), the role of learning in setting the human speech act repertoire, and the developmental sequence in which these representations emerge.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1.

Data Availability Statement

The data and analytic code necessary to reproduce the analyses presented here are publicly accessible. The materials necessary to attempt to replicate the findings presented here are publicly accessible. The analyses presented here were not pre‐registered. Data, code, and sample materials for this research are available in the Supporting Information.