The interplay of prominence-lending factors and language production in children

Several factors influence how adult speakers produce an utterance and which element they choose to start with. For example, German-speaking adults are more likely to begin an utterance with an element if it is (a) in their focus of attention, (b) animate rather than inanimate, and (c) on the left side of a picture rather than the right. However, while such prominence-lending factors affect adult language production, little is known about their effect on child language production. The current study addresses this issue by testing German preschoolers in picture-based language production tasks while measuring gaze patterns via eye-tracking. Specifically, we examined the effect of three factors known to affect adult language production—attentional cueing, animacy, and position in a visual display. To assess how children respond to these manipulations, we zoomed in on two syntactic structures for which children demonstrate varying degrees of proficiency: conjoined noun phrases and transitive sentences. Our results revealed no effects of animacy and position, but significant effects of attention on child language production. Children were more likely to first fixate on a character that had been visually cued and were also more likely to name that character first when producing conjoined noun phrases. Furthermore, initial fixations predicted the order in which children mentioned elements, suggesting a tight link between visual attention and language production. However, visual attention exclusively affected the linearization of nouns in simple conjoined noun phrases but not the selection of starting points in children's production of transitive sentences. These findings show that children's tendency to verbalize what is in their focus of attention is influenced by syntactic complexity. Specifically, even when children's focus of attention was on the patient character, they nonetheless adhered to their preferred active SVO sentence structure. Furthermore, children's sentence planning strategies as reflected by speech-onset-times and the number of disfluencies were likewise immune to prominence-lending factors. Thus, unlike adults in comparable experiments, our findings seem to suggest that children did not show a radically linear incremental word-by-word sentence planning strategy that proceeds from the element in the focus of attention.

If you produce an utterance, you have to start somewhere. This is true regardless of whether you produce a simple utterance like “the cat and the mouse,” or more complex sentences like “the mouse was chased by the cat that had been very hungry for days.” However, how do you decide what to start with? Studies have shown that whether an entity is mentioned first can be influenced by a variety of different factors. These factors are often subsumed under the terms “prominence” or “accessibility.” One central factor that can increase the prominence of an entity and thereby influence the starting point of a sentence is the focus of attention (e.g., Gleitman et al., 2007; Myachykov et al., 2012, 2018). For instance, in a seminal study by Gleitman et al. (2007), English participants were presented with images depicting two characters involved in an action (e.g., a dog and a cat growling at each other). Before watching the pictures, participants' attention was manipulated by means of a very brief visual cue (a small rectangle presented for 60 to 80 ms) that drew the speakers' attention either to the left or to the right character. Critically, when describing the pictures, speakers were more inclined to start their utterances with the character that had been cued previously and was hence in their focus of attention (i.e., speakers were more likely to say “The dog and the cat are growling at each other” when the picture of the dog had been cued, and vice versa for the cat). The same was true for transitive events that involved an agent acting on a patient (e.g., a dog chasing a cat). Thus, when the cue directed participants' attention to the patient location, speakers were more likely to start their sentences with the patient and to assign the grammatical role of the subject to the patient, thereby increasing the number of produced passive voice sentences (e.g., the cat was chased by the dog). These findings were corroborated by various studies (e.g., Myachykov et al., 2012, 2018). However, while attentional cueing with a subliminal (or implicit) cue is already sufficient for modulating structural choices in English speakers, findings from German suggest that the duration of the visual cue has to be extended in order to exert an influence on structural choices (see Esaulova et al., 2021a). Thus, while a very brief cue presented for 60 ms at the position where the patient was to appear in a visual scene did not lead to the production of significantly more passives in adult German speakers' sentence production (Esaulova et al., 2019), the same cue presented for a longer duration (700 ms) led to a significant increase of producing passives (Esaulova et al., 2021a). Results of experiments employing implicit or explicit visual cueing converge on the finding that speakers are more inclined to start an utterance with a referent that is in their focus of attention.

Beyond this rather bottom-up process of attentional orienting, there is evidence that sentential starting points can also be influenced by conceptual factors. One prime factor is the animacy status of a referent (e.g., Esaulova et al., 2019; Franz et al., 2021; McDonald et al., 1993; Van Nice and Dietrich, 2003; Tanaka et al., 2011). For instance, when asked to produce conjoined noun phrases in response to picture stimuli (e.g., depicting a dolphin and a planet), German speakers preferably produced animate items before inanimate ones (i.e., they more likely produced “dolphin and planet” than the reverse, see Franz et al., 2021). Similar effects of animacy have been observed at the sentence level. For instance, when English speakers were asked to reproduce sentences involving an animate and an inanimate entity (e.g., The music soothed the child), they were more likely to incorrectly remember animate referents as sentential subjects than inanimate ones, even if this resulted in the production of passive constructions (e.g., The child was soothed by the music; McDonald et al., 1993). The same was true for speakers of Japanese (Tanaka et al., 2011). Furthermore, Japanese speakers were more likely to assign animate referents earlier positions in a sentence than inanimate referents, demonstrating that animacy affected both syntactic structure and linear word order (see Tanaka et al., 2011). Effects of animacy have also been found to influence the choice of syntactic structure in event-description tasks (e.g., Esaulova et al., 2019; Van Nice and Dietrich, 2003; Van de Velde et al., 2014). For instance, Esaulova et al. (2019) found that German speakers were more likely to start their utterance with the patient and produce passives when the patient was animate rather than inanimate, demonstrating that animacy can render a referent more prominent, resulting in its promotion to grammatical subject.

In addition to conceptual factors like animacy, yet another factor that can influence the starting point of an utterance is the position of a character in a visual display. For instance, when asked to produce conjoined noun phrases in response to pictures where two entities were presented next to one another, adult German speakers exclusively started their utterance with the left entity and only subsequently named the right entity (Dolscheid et al., 2024). This strong spatial bias is likely shaped by cultural habits, specifically the directionality of a writing system that participants are enculturated with (i.e., reading from left to right in German). In line with this proposal, the left-first bias was significantly attenuated and even partly reversed in a cohort of Arabic participants who had been primarily exposed to a right-to-left writing system (cf. Dolscheid et al., 2024). In addition to conjoined noun phrases, similar spatial biases have also been observed in the context of visually depicted transitive events. For instance, Esaulova et al. (2019) asked German-speaking participants to describe scenes depicting simple transitive events with two characters (e.g., a fisherman filming a clown). Whereas, on half of the trials, the patient was located on the right, on the other half, the patient was located on the left of the agent. The results revealed that participants were more likely to produce passives when the patient character was located on the left rather than on the right, demonstrating that German speakers are more inclined to start an utterance with the left-most character in line with conventions of reading and writing from left to right (for converging evidence, also see Pokhoday et al., 2019). Conversely, participants were faster in initiating verbal descriptions when the patient appeared on the right, rather than on the left of the agent, again suggesting that a left-to-right bias is affecting adult speakers' sentence production (Esaulova et al., 2021b, 2020).

Taken together, a range of different factors appear to influence language production and the selection of starting points in adult speakers. Such factors are often conceived of as increasing an entity's “accessibility” or “prominence.” While these terms are often used interchangeably, the term prominence covers a broader scope, encompassing phenomena from phonetics over morpho-syntax to discourse and text linguistics as well as the hearer and the speaker perspective. Here we draw on a definition of prominence in morphosyntax by Himmelmann and Primus (2015) (also see Von Heusinger and Schumacher, 2019). According to this definition, prominence entails that some entities stand out more than others and are often mentioned early in a sentence and take up “higher” grammatical roles. Linguistic prominence is connected to human attention in “that there is a correspondence between prominence-related linguistic structures and the psychological notion of attention” (Himmelmann and Primus, 2015). In line with this proposal, there is evidence that attentional orienting can also influence the starting point of a sentence or an utterance. As outlined above, an entity that is in the speaker's focus of attention is more likely to be mentioned first and to be assigned a more prominent grammatical role (such as the grammatical subject, see e.g., Gleitman et al., 2007; Esaulova et al., 2021a). Critically, however, while such effects have been studied quite extensively in adults, hardly anything is known about their impact on children. Specifically, it is unknown whether and how attention comes to influence child language production, including the selection of linguistic starting points. Similar gaps pertain to other prominence-related cues known to influence adult language production but whose effect is unclear for children. In the present study, the central goal is to fill this gap by investigating whether and how increasing the prominence of an entity affects child language production and whether these effects are similar to what has been observed for adult speakers. For this, we make use of eye-tracking to provide new insights into children's language production strategies. Specifically, we examine whether the prominence-lending factors attentional cueing, animacy, and position influence child language production, and, if so, which aspects of language production are affected. In the following, we first review the existing research and open issues for each of these factors before outlining our research goals in more detail.

Effects of attention on child language production

Up to now, hardly any research has investigated links between attention and language production in children. In one exception to this, Ibbotson et al. (2013) studied attention-language links in English preschool children. Children were asked to describe transitive events which were enacted with puppets (e.g., a cat hitting a dog), while the experimenter's eye-gaze was directed to one of the puppets. Critically, when attention was drawn to the patient character (i.e., when the experimenter explicitly looked at the patient), children were also more likely to start their utterances with the patient, resulting in a greater rate of passive structures (Ibbotson et al., 2013). Yet, while these findings seem to provide some first evidence for a link between attention and language production in children, several questions remain open: for one, attention was manipulated in terms of a socially meaningful cue (i.e., eye gaze shifts by an interlocutor). However, it is unknown whether similar effects can be observed when attention is shifted in the absence of social interaction, for instance, in terms of a brief, meaningless cue, as has been demonstrated for adults (see e.g., Gleitman et al., 2007). Second, although the increase in children's passive production is taken to reflect attentional orienting, the study did not provide any direct measure of children's allocation of attention (e.g., by measuring children's eye gaze patterns). A similar limitation relates to a more recent study by Stanford and Delage (2021) who conducted an event-description study with three groups of French-speaking school children, a group of 20 typically-developing children (mean age 8 years) and two groups of neuro-divergent children. In one of the tested conditions, Stanford and Delage manipulated the perceptual salience of the patient of the depicted transitive event by presenting it in color while the agent in the scene was presented in black and white. However, although this perceptual manipulation resulted in a slight increase in the production of passives in the group of the typically-developing children compared to a control condition in which both depicted characters were colored, the authors did not provide a direct measure of the effectiveness of their manipulation in directing children's attention (e.g., eye gaze patterns). Furthermore, since all other event scenes in the experiment were fully colored, this raises the question whether children's attention was not in fact drawn to the exceptionally presented black-and-white agent character in the pictures of this experimental condition. In sum, it remains an open question whether and how shifts in visual attention come to affect children's language production. Although previous work has shown that preschool children are more prone than older children and adults to attend to visually salient features (e.g., Helo et al., 2014, 2017), it remains unclear whether cueing with a subtle, meaningless cue can affect children's attentional orienting and language production, gaps we seek to address in the current study.

Effects of animacy on child language production

Unlike for attentional orienting, which has hardly been studied in children, more work has been devoted to investigating the role of animacy in child language production. As these studies show, effects of animacy already appear influential during language acquisition (e.g., Drenhaus and Féry, 2008; Harris, 1978; Lempert, 1989, 1990; Prat-Sala et al., 2000). For instance, when asked to produce conjoined noun phrases in response to picture stimuli, German children preferably named animate items before inanimate ones (e.g., Dolscheid et al., 2024; Franz et al., 2021). Beyond the phrase level, animacy also appears to influence children's choice of sentence structure. For instance, children were better able to produce and comprehend passives when they were taught passive constructions with animate patients, such as “The baby is being picked up by the girl” than when they were taught passives with inanimate patients, such as “The flower is being picked up by the girl” (Lempert, 1989, 1990). Similarly, when English-speaking children were asked to describe pictures of transitive events, they were more likely to produce passives when the patient outranked the agent in terms of animacy (e.g., as in “A man was hit by a swing,” see e.g., Harris, 1978). As these findings show, an increase in the production of non-canonical utterances like passives was most pronounced when an inanimate agent acted upon an animate patient (e.g., Dewart, 1979; Harris, 1978; Prat-Sala et al., 2000). However, research on children (e.g., Lempert, 1989) and adults (Esaulova et al., 2019, 2021a) has also found that an animate (vs. an inanimate) patient was sufficient to increase the production of passives, even when the agent was always animate. Taken together, these findings show that the animacy status of a referent can affect child language production, shaping both starting points and the choice of grammatical structure (such as passivization). Based on these findings, we also anticipate to observe animacy effects in the present study.

Effects of position on child language production

Similar to attention, little is known about how the position of an element in a visual display affects children's descriptions of pictures. As outlined above, for adult speakers, effects of referent position on language production have consistently been observed in event-description tasks where a patient positioned left of the agent in an action scene coaxed adult speakers to produce significantly more passives compared to a condition where the patient appears on the right side of the action (Esaulova et al., 2019, 2020, 2021a,b). Also, in cases where speakers produced the canonical active SVO structure to describe such scenes, a left-positioned patient led to a significant increase in speech-onset times compared to conditions with a left-positioned agent, indicating planning costs related to a mismatch between the preferred active SVO structure and the expectation to have the agent on the left side of the action scene (Esaulova et al., 2021b). In contrast, there is some first evidence to suggest that spatial biases in language production are not yet present in preliterate children. For instance, unlike literate German adults who almost exclusively started to produce a conjoined noun phrase by first mentioning the left entity, the same bias was not (yet) observed in preschool children (Dolscheid et al., 2024). Since these German preschool children were not yet able to read and write, it appears likely that spatial biases in language production emerge as a function of literacy (i.e., by learning to read and write in a specific directionality). However, to date no research has targeted the question of whether and when such effects emerge in children and which literacy skills are necessary for this prominence-lending factor to become operative. It could be that early precursors of literacy such as picture-book reading might be sufficient to induce such biases. However, although joint picture-book reading might in principle already accustom preschool children with the left-to-right reading direction of German, eye-tracking research has shown that preschool children rarely focus on text during this activity (Evans and Saint-Aubin, 2005; Roy-Charland et al., 2007). Studies have found that even preliterate children sometimes show spatial biases in other domains, e.g., in associating smaller numbers with the left side of space, and vice versa for bigger numbers (see e.g., Patro and Haman, 2012). It has also been argued that spatial biases are due to hemispheric asymmetries and, hence, should be present early in development, irrespective of cultural factors such as reading and writing habits (e.g., Chatterjee et al., 1999; Rugani and de Hevia, 2017). Based on these assumptions, it could be that spatial biases reveal themselves in language production in that the left-positioned element is mentioned first. Alternatively, it might be that spatial biases only become visible when more subtle measures are considered. For instance, it is possible that spatial biases are observable in children's ways of inspecting pictures, even if these biases do not necessarily resurface in overt patterns of language production. Finally, if spatial biases are dependent on learning to read and write, such biases should not yet affect preliterate children. In the present study, the use of eye-tracking technology is ideally suited to tap into these open issues and to study the potential effects of spatial biases on children's viewing behavior and language production.

Goals of the present study

In the present study, the overarching goal is to examine whether and how increasing the prominence of an entity affects child language production. To this end, we examine how children respond to manipulations of attentional cueing, animacy, and the position of a referent. Since these factors range from basic attentional (cueing) to conceptual (animacy) and culturally mediated visual preferences (position), investigating these factors together provides new insights into whether and how these different prominence-related cues influence child language production. Furthermore, this synoptic approach can shed light on whether the different factors exert similar or different effects on child language production. To study child language production, we ask German preliterate preschool children to describe pictures while their gaze patterns are monitored via eye-tracking. Importantly, measuring children's eye-gaze patterns can serve as a manipulation check to verify whether our visual manipulations are indeed effective (e.g., whether attentional cueing leads to a greater proportion of first looks to a cued character). In addition to eye-gaze, we also analyze children's verbal responses. Specifically, we elicit two syntactic structures for which children show different levels of expertise: conjoined noun phrases and transitive sentences. Whereas, children master the production of conjoined noun phrases from very early on (e.g., Hartung, 2012; Lust and Mervis, 1980), it takes them substantially more time to acquire the syntax of transitive sentences, especially non-canonical structures such as passives (e.g., Marchman et al., 1991). For instance, even though German children have been found to produce verbal passives before the age of 4 years (Abbot-Smith and Behrens, 2006; Eisenbeiss, 1993; Fritzenschaft, 1994), the production of passives remains dispreferred and difficult for children (Marchman et al., 1991). Therefore, in addition to transitive sentences that require grammatical role assignment and that may pose challenges in the context of non-canonical sentence structures (such as passives), we also examine potential effects of prominence at the syntactically simpler phrase level by asking children to produce conjoined noun phrases. It is possible that the effects of prominence influence children's language production patterns, regardless of the structure produced (i.e., such effects might reveal themselves for both relatively simple structures, such as conjoined noun phrases, and more complex structures, like transitive sentences). Alternatively, it is possible that some prominence-lending factors do not yet influence child language production, neither at the phrase nor at the sentence level. On yet another option, it could be that the effects of prominence-related factors differ depending on the syntactic structure that has to be produced. For instance, it is possible that effects of prominence reveal themselves for structures well mastered by children but not necessarily for more demanding structures that require grammatical role assignment. In sum, by examining syntactic structures of different complexity, our study will speak to the question of whether potential effects of prominence differ depending on language production affordances and syntactic complexity.

By zooming into child language production patterns, we will also gain some first insights into children's sentence planning strategies. Adult language production typically proceeds in a linear fashion (e.g., Levelt, 1989). That is, speakers usually produce multi-word utterances without having prepared them fully ahead of articulation. However, the extent to which adult speakers plan ahead can vary. Speakers may build an utterance word by word or phrase by phrase, preparing only a small part of the message at each step, which is often called “linear” or “lexical” incrementality (e.g., Gleitman et al., 2007). On the other hand, speakers may plan ahead more broadly, organizing the conceptual content and syntactic structure of a larger preverbal message before starting to speak (hierarchical or relational incrementality, see e.g., Konopka, 2025). While sentence planning strategies may vary depending on the context and task affordances (e.g., Allum and Wheeldon, 2009), at least under certain circumstances, language production seems to follow a radically linear incremental planning strategy where the increment that is encoded first is also produced first and becomes the starting point of the utterance without any further relational planning. Thus, as outlined above, increasing the patient's prominence by virtue of different cues, often results in an increased tendency to start a sentence with the patient, taken to reflect a (radically) linear incremental word-by-word planning strategy that is employed by adult speakers. However, at the moment, it is an open question which strategy is used by children. If children follow such a linear incremental planning strategy, then they also should be more inclined to start an utterance with the most prominent element (e.g., the element that is in their focus of attention). This should become visible in children's order of mention when producing conjoined noun phrases, i.e., the prominent noun should be mentioned first. In sentence production, effects of linear incremental planning might become visible in various ways. On the one hand, it could be that children are more inclined to produce passive sentences in response to an increase in patient prominence, akin to adults. However, given that passives are more demanding, it might also be that children start their sentences with the patient, but then correct themselves to resort to an alternative event description which may be easier to realize (e.g., SVO actives, starting with the agent). These different options would reflect effects of patient-prominence in child language production and could be interpreted in terms of a linear sentence planning strategy in children. By contrast, it might be conceivable that children display no such effects of patient prominence in language production, indicating that children rather follow a more hierarchical linear planning strategy in that they do not immediately start their utterances with the first increment that becomes available. In sum, by examining effects of prominence on child language production, our study will shed some first light on children's sentence planning strategies in response to prominence.

Beyond examining children's order of mention and sentence structure, we will consider additional measures of sentence production to gain more detailed insights into children's planning strategies. For instance, Charest et al. (2016) found that increasing the patient's accessibility by means of semantic priming resulted in higher processing costs in English children's sentence production. Specifically, children were slower in initiating an active sentence, and they also showed more hesitations and produced shorter utterances when producing active sentences that followed patient-related primes. Building on these findings, in the present study, we also include additional measures of sentence production affordances, measuring speech-onset times, number of disfluencies, and mean length of utterance (MLU). If prominence-related factors were already effective at the sentence level in children, we would expect that increasing the prominence of the patient would lead to higher planning costs of active sentences, with longer speech-onset times, a greater number of disfluencies, and shorter sentences (i.e., MLUs) (cf. Charest et al., 2016). Including these more fine-grained measures will thus allow us to investigate whether potential effects of prominence reveal themselves beyond sentence structure, thereby providing additional insight into children's production strategies.

In sum, by including various measures of sentence production and by examining structures of different complexity, our study will address the question of how effects of prominence reveal themselves in child language production and whether these effects differ depending on language production affordances and syntactic complexity.

We administered two language production tasks that were accompanied by eye-tracking: a noun-pair naming task and an event-description task. These two tasks were designed to be as similar as possible but diverging in the syntactic structure that had to be produced by children to describe the presented stimuli. In the noun-pair naming task, we elicited children's production of conjoined noun phrases, and in the event-description task, we elicited the production of transitive sentences. To manipulate children's allocation of visual attention in these two tasks, we employed a visual cue that was presented for 700 ms, a cue duration that has been found to effectively influence language production in adult speakers of German (Esaulova et al., 2021a).

Participants

Thirty-three monolingual German-speaking preliterate preschool children (mean age: 5;00; range 4;00 to 5;11 years, 16 female, 17 male) were recruited and tested in various daycare centers in Germany. All tested children had normal or corrected vision as well as normal hearing. Furthermore, no child displayed any evidence for developmental language impairments as reported by caretakers and pre-school teachers. Children were also quite homogeneous in terms of socio-economic background, coming from middle-class families with 81% of mothers having obtained a higher level of education with at least 12 years of schooling. While 33 children participated in the testing session, not all data could be used for analysis. This was due to problems with the sound recording or with the inability of some children to comply with the eye-tracking affordances, such as staying seated and focused on the screen during the task. For the noun-pair naming task, we report data of the 32 children who completed this task (mean age: 5;00; range 4;00 to 5;11 years, 15 female, 17 male). For the event-description task, we were able to analyze data from 28 children (mean age: 5;00; range 4;00 to 5;11 years, 15 female, 13 male). The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Commission of Cologne University's Faculty of Medicine (number of approval: 16–134). Prior to testing, all caretakers gave informed written consent for their children to participate in the study.

Experimental tasks

Noun-pair naming task

In the noun-pair naming task, children were presented with pictures displaying two entities (e.g., a clown and a dwarf, see Figure 1a, and were asked to name the two depicted entities by producing a conjoined noun phrase. We prepared a set of 24 black-and-white pictures that contained stimuli in three conditions: a baseline condition and two experimental conditions, each of which was designed to allow for a pairwise comparison against the baseline condition. In the baseline condition, the stimuli depicted two animate characters that were portrayed with stereotypical features so that they could be easily identified by the tested children (e.g., a clown and a dwarf, see Figure 1a). In the cueing condition, children's allocation of attention was manipulated by means of a brief meaningless visual cue (a red dot) presented for 700 ms in the place where the left animate character was about to appear (cueing condition, e.g., a cowboy and a knight, see Figure 1b). Furthermore, we manipulated the animacy status of the left character in the experimental inanimate condition by presenting an inanimate entity (e.g., a chair next to a clown, see Figure 1c). Noun pairs were composed of 16 nouns depicting animate characters (e.g., clown, pirate, etc.). In the inanimate condition, eight nouns depicting inanimate objects (e.g., carpet, wardrobe) were presented left of an animate character. The depicted noun pairs were matched in terms of gender (only masculine nouns were used), syllable length, visual characteristics (color, size, complexity), and lemma frequency taken from the childlex corpus (Schroeder et al., 2015; mean normalized lemma frequency per million for the youngest age group 6–8 years, animate nouns: 116.1, inanimate nouns 103). We additionally conducted a pilot study with German preschool children to ensure that they were able to identify and name the depicted entities. Two experimental lists were created, consisting of 24 items presented in pseudo-randomized order (i.e., 24 noun pairs, with eight pairs in the baseline condition, eight pairs in the cueing condition, and eight pairs in the inanimate condition). Each of the 16 animate characters appeared on the right as well as on the left side of the presented pictures. Each noun pair appeared only once per list. Lemma frequencies of nouns were matched between the two lists (all p > 0.05). Also, within each list, the lemma frequencies of nouns appearing on the left or right side of the pictures were matched within and across conditions (all p > 0.05). All materials can be found at https://osf.io/hqvz5/.

To familiarize children with the task, children were presented with one picture depicting a pair of animals beside each other (a horse and a cow) and a picture displaying an inanimate and an animate element (a donkey and a trolley), and were asked to name what they saw.

Event-description task

To ensure a direct comparison, the event-description task was designed to be as similar as possible to the noun-pair naming task. The same noun pairs as in the noun-pair naming task were presented, however, with the difference that the characters were involved in a transitive event with an agent acting upon a patient (e.g., a clown pushing a dwarf, see Figure 2a). Pictures showed eight different actions, describable by verbs known to children (e.g., washing, pushing, etc.). We designed three experimental conditions that allowed for a pairwise comparison against a baseline condition, in which an animate agent acted upon an animate patient depicted on the left side of the picture (see Figure 2a). Like in the noun-pair naming task, children's allocation of attention was manipulated by means of a brief visual cue presented for 700 ms in the place where the animate patient character was about to appear (cueing condition, see Figure 2b). In addition, we manipulated the animacy status of the patient, which was either an animate character in the baseline condition (see Figure 2a) or an inanimate entity in the inanimate condition (see Figure 2c). Finally, we manipulated the position of the animate patient. While the patient of the action was positioned left of the agent in the baseline condition, it appeared on the right side of the agent in the condition ‘patient right' (see Figure 2d). Four pseudo-random experimental lists were created, consisting of 32 presented items (i.e., 8 pictures in the baseline condition and in each of the three experimental conditions (cueing, inanimate, and patient right). Each of the 16 animate characters appeared as agent as well as patient in the depicted events. The lemma frequencies of agents, patients and verbs were matched between these four lists (all p > 0.05). Also, within each list, the lemma frequencies of agent and patient were matched within and across conditions (all p > 0.05), and the lemma frequency of the verbs was matched across conditions (all p > 0.05). Each picture was only presented once per list. All materials can be found at https://osf.io/hqvz5/.

To familiarize children with this task and to ensure that they experienced no problems in recognizing and naming the depicted events, children were presented with 8 practice trials showing two cartoon characters (a bear and a tiger) involved in the actions displayed by the experimental stimuli. Children were asked to describe the depicted actions (Look, what are you seeing here?). All children were able to recognize and describe the depicted actions. Neither during the familiarization phase nor during the experiment were the children prompted to produce a particular syntactic structure.

Act-out task

In addition to the two eye-tracking experiments, we assessed whether children were already able to understand passives by testing them in an act-out task with toy figurines. Children were presented with several toy figurines (e.g., a lion and a tiger) and were asked to act out a sentence like the lion is being carried by the tiger, spoken by the experimenter. We subsequently presented five active and five passive sentences for acting out. Sentences were presented in one of two versions in which the agent and patient of the action were reversed.

Procedure

Children were tested individually in a quiet room in their respective daycare centers. In a first step, the experimenter engaged with the child so that the s/he could become familiar with the experimental setting and the eye-tracker (Eyelink Portable Duo). The child was then comfortably seated within a 60 cm distance from the laptop screen on which the experiment was presented, while the experimenter could follow their eye-gaze on a host computer and control the experiment. Children were instructed to describe each picture appearing on the screen. Prior to the two eye-tracking experiments, a 5-point calibration was performed to ensure the validity of eye gaze patterns. If necessary, calibration was repeated during the experiment. Before each trial, the children were asked to fixate on a small ladybug that appeared centrally on the screen to ensure that children's fixations were centered on the screen before a given trial. Once the experimenter ascertained that the child was fixating on the ladybug, she started the presentation of the next stimulus picture.

To avoid that children had to sit for long stretches of time calmly in front of the eye-tracker, the three tasks were administered in a fixed order. Children were first tested in the noun-pair naming task, followed by the act-out task, and finally, the event-description task was administered. For the noun-pair naming task and the event-description task, children were successively assigned to one of the experimental lists.

In both production experiments, in addition to analyzing children's verbal responses, we recorded their eye-gaze to the entities depicted. For this, we analyzed whether the first fixation after presentation of the target picture went to the left or the right depicted entity. To measure the location of children's first fixations, we used Eyelink Data Viewer to determine two rectangular interest areas of the same size surrounding the left and the right entity.

Children's utterances were recorded with a Behringer XM8500 Dynamic Vocal Microphone positioned in front of the children. All utterances were transcribed and coded (e.g., regarding order of mention, sentence structure, etc.) by two research assistants who mutually checked their transcriptions and coding. Transcriptions and coding were additionally checked by the second author of this study. When coders differed in terms of their transcriptions and/or coding, these cases were discussed among the whole research team. Utterances for which no consensus as to what had been said by the child could be reached were deleted from further analyses.

Coding of responses

Responses in the noun-pair naming task were coded as left-first and right-first depending on which depicted entity was named first. Utterances where none (I don't know) or only one entity was named were excluded from further analyses, as were utterances that could not be related to the depicted entities (e.g. a man and another man) or utterances where children incorrectly changed the animacy status of the depicted entities (e.g., by producing bakery instead of cook). In total, 75 out of 768 produced utterances in the noun-pair naming task had to be removed from analysis (10%).

Responses in the event-description task were coded according to whether the agent or the patient of the action was mentioned first, and they were also analyzed with respect to their syntactic structure, i.e., whether an active or a passive clause was produced. Utterances were analyzed when they minimally consisted of two noun phrases naming the depicted entities or of one noun phrase and a verb describing the event. Utterances where children falsely changed the animacy status of the depicted entities were not further analyzed. In total, children produced 873 analysable utterances out of 896 reactions (97.4%), indicating that they were well able to perform the task.

As an indicator for the costs related to lexical access and syntactic planning, we measured speech-onset times from the presentation of the picture to be described until the onset of speech (identified using the Praat software; Boersma and Weenink, 2017). Filled pauses at the beginning of an utterance (e.g. uhm, uh) were ignored in this respect, i.e., speech-onset times were calculated from picture presentation until the start of the first word produced by the child. For trials in which the child spoke to the experimenter while the target picture was presented, no speech-onset time was measured. In addition, we counted the number of disfluencies as another measure of costs associated with utterance planning. As disfluencies, we counted silent pauses of at least 500 ms duration, filled pauses, repetitions of speech sounds, words or phrases as well as revisions of these elements. In addition, we calculated mean length of utterance (MLU in words) for each child per condition as another indicator of processing costs associated with sentence planning.

In the act-out task, children's actions with toy figurines were video-recorded and analyzed with respect to whether the child's acting-out matched the sentence presented by the experimenter, i.e., specifically, whether agent and patient roles were correctly assigned. For further data analysis, the child was awarded one point if s/he succeeded in correctly acting out the requested action and no point if s/he failed to do this. The points achieved for acting out sentences were added up with a maximum of 5 points that could be reached for each sentence type, active sentences and passive sentences. We set a criterion and assumed that a child had acquired the passive construction if s/he correctly acted out at least 4 of the 5 passive sentences. The same criterion was set for active sentences.

Statistical analyses

For statistical analyses, we computed mixed-effects models in R (R Core Team, 2021, Version 4.4.1), using the lmerTest package (Kuznetsova et al., 2017, Version 3.1.3) based on the lme4 package (Bates et al., 2015, Version 1.1.27.1). We started with the maximally specified model (Barr et al., 2013) and successively removed random slopes if the maximal model failed to converge. For all binary-coded dependent variables, GLMMs with a logit link function were computed (e.g., for first fixation to the left vs. the right character). For speech-onset times, we conducted linear mixed effects models. The number of disfluencies and MLUs were analyzed in terms of poisson regression (i.e., GLMMs with a poisson link function). Where appropriate, pairwise comparisons were conducted using the emmeans package (Lenth et al., 2017, Version 1.7.2), which adjusts p -values using the Tukey method for multiple comparisons. In line with open science practices, the data and analyses are openly available at OSF (https://osf.io/hqvz5/).

Noun-pair naming task

We conducted a generalized linear mixed effects logistic regression model on first looks (left vs. right) with condition as fixed effect (treatment coded comparing cueing and inanimacy to the baseline condition) and participants and items as random intercepts. The model revealed that visual cueing was highly effective in modulating children's attention in the noun-pair naming task, as reflected by a significant increase in first fixations to cued characters compared to the baseline without cueing, z = 8.25, p < 0.0001 (cueing: 93% first fixations to cued characters vs. 54% first fixations to these characters in the baseline condition, see Figure 3). Children were also more likely to start their utterances with the cued character compared to baseline, z = 3.32, p < 0.001 (69% first mention of cued characters vs. 55% first mention of these characters in the baseline condition, see Figure 4 for an overview). Children's first fixations on a character also correlated significantly with their propensity to first mention that character, demonstrating that children were more inclined to produce an utterance with an entity that was in their focus of attention, r = 0.51, p < 0.01 (see Figure 5). Additionally, we examined whether children's first looks were predictive of their order of mention. We found that children's order of mention depended on where they looked first, as revealed by a significant influence of first fixation on children's propensity to first mention a character, χ2 = 46.57, p < 0.001. However, generalized linear mixed effects logistic regression models revealed no significant effect of animacy, neither on children's first fixations, z = −0.80, p = 0.42, nor on children's left-first utterances, z = 0.02, p = 0.99 (55% first mention of inanimate entities compared to 55% first mention of animate entities in the baseline condition), indicating that differences in animacy between the two depicted entities did not influence eye-gaze and order of mention. An overview of the data is presented in Table 1. Furthermore, since the models revealed no significant intercept, neither on children's first fixations, z = 0.15, p = 0.88, nor on children's order of mention, z = 0.91, p = 0.36, this suggests that children were not more likely to first fixate or name the character depicted on the left. We additionally examined children's speech onset times (SOTs) and number of disfluencies. A mixed-effects linear regression with condition as fixed effect and participant and item as random intercepts revealed no significant effect of animacy, t = −0.82, p = 0.42, and no significant effect of cueing, t = −0.16, p = 0.87 on children's SOTs. Furthermore, a mixed-effects poisson regression with condition as fixed effect and participant and item as random intercepts on children's number of disfluencies revealed no significant effect of animacy, z = −1.73, p = 0.08 and no significant effect of cueing, z = −0.80, p = 0.43.

Event-description task

In the event-description task, we conducted a generalized linear mixed effects logistic regression model on first looks (patient vs. agent) with condition as fixed effect (treatment coded comparing cueing, inanimacy, and position to baseline) and participants and items as random intercepts. The model revealed that children were more likely to first fixate the patient when it had been cued compared to baseline, z = 8.18, p < 0.0001. In addition, children were more likely to first fixate on the patient when it was animate rather than inanimate, z = −2.022, p < 0.05. By contrast, there was no effect of position on children's first fixations, z = 0.57, p = 0.57. Conversely, there was no significant effect of the intercept, further suggesting that no spatial bias was present in children's first fixations, z = −0.59, p = 0.55 (see Figure 6).

Regarding language production, children predominantly produced active sentences, beginning their utterances with the agent (95% of the cases). In total, children produced 42 patient-first utterances, with only 10 passive sentences produced by four children, and 32 alternative structures. With two exceptions, the alternative structures all consisted of utterances where children started by mentioning the patient first but then restarted to produce a canonical active sentence with the agent as subject (such as The dwarf…, the clown is pushing the dwarf.). In the baseline condition, children produced 8 patient-first utterances, in the inanimate condition, 7 patient-first utterances, in the cueing condition, they produced 14 patient-first utterances, and in the condition in which the patient was located on the right, children produced 13 patient-first utterances. However, the very low number of patient-first utterances did not allow for a reasonable statistical comparison.

We additionally conducted generalized linear mixed effects models to assess whether animacy, cueing, and position had an impact on children's speech-onset times (SOTs), mean length of utterance (MLU), or number of disfluencies during their production of (agent-first) active sentences. For analyzing children's SOTs, we took into account active utterances in which the agent was mentioned first (resulting in 609 analyzable SOTs). We then removed outliers that were two or more standard deviations above or below children's average response times. The final set of responses consisted of 575 trials. A linear mixed-effects regression on log-transformed SOTs, with condition as a fixed effect and participants and items as random intercepts, revealed no significant effects of inanimacy, t = 0.71, p = 0.39, cueing, t = 0.37, p = 0.72, and position, t = 0.39, p = 0.70, compared to baseline. To analyze children's MLUs, we conducted a generalized mixed-effects poisson regression, with condition as a fixed effect and participant as a random effect. Compared to baseline, the model revealed no significant effect of inanimacy, z = −0.71, p = 0.48, no effect of cueing, z = −0.61, p = 0.54, and no effect of position, z = −0.45, p = 0.65. Regarding, children's number of disfluencies, a mixed-effects poisson regression with condition as a fixed effect and participant as a random effect revealed no significant effect of inanimacy, z = 0.95, p = 0.34, no effect of cueing, z = −0.82, p = 0.41, and no effect of position, z = −0.43, p = 0.67 compared to baseline. An overview of the data is presented in Table 2.

To draw a direct comparison between the two different tasks (the noun-pair naming task and the event-description task), we also conducted a mixed-effects linear regression model on children's SOTs with task as a fixed effect and participants as random effects. The model revealed a significant effect of task, t = −6.22, p < 0.0001, demonstrating that it took children significantly longer to initiate a verbal response in the event-description task (average SOT: 3.49 sec) than in the noun-pair naming task (average SOT: 2.47 sec). We also conducted a mixed effects poisson regression on children's number of disfluencies with task as a fixed effect and participants as random effects. The model revealed a significant effect of task, z = −19.12, p < 0.0001, showing that children produced significantly more disfluencies in the event-description task (average number of disfluencies: 2.36) than in the noun-pair naming task (average number of disfluencies: 0.94). To rule out that differences in disfluencies were simply due to the fact that children produced more words in the event-description task than in the noun-pair naming task, we also compared children's number of disfluencies at the onset of their utterances. A mixed effects poisson regression on children's number of disfluencies at utterance onset with task as fixed effect and participants as random intercept revealed a significant effect of task, z = −3,15, p < 0.01, showing that children produced significantly more disfluencies at the onset of an utterance in the event-description task (average number of disfluencies at onset: 0.48) than in the noun-pair naming task (average number of disfluencies at onset: 0.20).

Act-out task

While 33 children took part in the act-out task, here we report data for the 28 children whose data was analyzed for the event-description task. The data of all 33 children can be found at OSF (https://osf.io/hqvz5/). All of the 28 children whose data could be analyzed for the event-description task reached the criterion for the comprehension of active sentences (4 out of 5 points), with 26 of the 28 children (93%) performing error free. With respect to passive comprehension, 22 children (79%) reached our criterion for passive comprehension and scored 4 or 5 points in acting out passive sentences. The remaining six children scored between 2 and 3 points. The data indicates that the huge majority of the tested children were well able to understand the passive construction.

Our study examined whether prominence-lending factors already affected child language production by specifically focusing on attentional orienting, animacy, and position, factors that have been found to affect German adult language production (Esaulova et al., 2019, 2021a). We found that children's language production was differently affected by these factors. Furthermore, by studying the effects of prominence both at the phrase and the sentence level, we found that the influence of prominence differed depending on the syntactic structure that had to be produced (i.e., we observed effects of attention at the phrase level but not during the production of transitive sentences). Taken together, our results suggest that prominence-related factors come to influence child language production in a rather nuanced way. Before we discuss the broader implications of our work, we first summarize and discuss the key findings for each of the factors examined.

Effects of attention on child language production

By investigating the effect of attentional cueing on child language production, we found that a subtle, meaningless visual cue was highly effective in orienting children's attention. Thus, children were more likely to first fixate a character when it had been preceded by a brief attentional cue. This was true for both the noun-pair naming task as well as the event-description task, indicating that attentional manipulations were overall effective in our study. These findings demonstrate that children's allocation of attention can be modulated by a brief visual cue, corroborating previous findings in adults (e.g., Esaulova et al., 2019, 2021a; Gleitman et al., 2007). Critically, our results also provide novel insights concerning the role of attentional focus for child language production: by manipulating children's visual attention in an experimentally controlled manner, we found that children were more inclined to start an utterance with an entity that was in their focus of attention (in the noun-pair naming task). Unlike previous studies (e.g., Ibbotson et al., 2013; Stanford and Delage, 2021), our task allowed us to directly tap into links between visual attention and language production. Furthermore, since we manipulated attention in terms of purely visual means, devoid of any semantic or social content, our observations cannot be attributed to socially meaningful cues, such as an interlocutor's eye gaze, but rather suggest that attention by itself—beyond joint attention—can influence language production in children. We also found that children's first looks at a character were predictive of their order of mention during the production of conjoined noun phrases. That is, overall, children were more likely to first mention a character when they had also initially looked at this character. Taken together, our findings provide some of the first evidence that children's language production patterns are already susceptible to bottom-up attentional orienting.

Critically, however, our results suggest that the influence of attention differs depending on syntactic complexity: whereas attentional orienting was effective during the production of coordinated noun phrases, a process well mastered by children at an early age, we did not observe an influence of attention on the production of transitive sentences. That is, even though children directed their attention to the visually cued patient character (as demonstrated by their viewing patterns), this increase in visual attention did not lead to concomitant effects on sentence production. One likely explanation for this finding is that children are less proficient and flexible in their production of non-canonical sentences such as passives (e.g., Marchman et al., 1991). Thus, even though the great majority of the children in our study were able to understand passives (as shown by their performance in the act-out task), children almost exclusively produced active sentences in the event-description task, with only very few instances of passives. Furthermore, even in the few cases where children started an utterance with the patient, most of these sentences were not passives but rather took the form of restarts (like A fisher, a farmer is pushing the fisher), again suggesting that children were not easily coaxed into producing passive sentences. Rather, children preferred the production of active sentences, resulting in a bias that seemed to persist even when the patient character was in the child's focus of attention. Critically, the very low number of patient-first sentences overall (including self-corrections) indicates that children generally did not react to attentional orienting by starting sentences with the patient. In conclusion, our results show that attentional cueing does not invariably affect language production but rather that its effects depend on language production affordances. Specifically, the noun-pair naming task and the event-description task, although designed as similar as possible in terms of lexical elements and visual features, differed critically in terms of production affordances and syntactic complexity. While naming characters in the noun-pair naming task was relatively easy for children, producing transitive sentences involving the same characters was more demanding. This difference was also reflected in children's speech-onset times and the number of disfluencies they produced: specifically, it took children much longer to initiate an utterance, and they produced more disfluencies in the event-description task than in the noun-pair naming task. These differences in language production affordances also interacted with the influence of attentional orienting: when the structure to be produced was relatively easy (like conjoined noun phrases without a sentence context), attentional cueing affected the starting point of an utterance. However, once affordances became higher, children's sentence production was not equally shaped by attentional orienting. Based on these findings, we can conclude that attentional orienting can affect children's language production, but only when attention does not conflict with language production affordances. Once attention and production preferences are in conflict, children continue to begin their utterance with the agent as sentence-initial subject, suggesting that children adhere to an agent-first and canonical SVO sentence structure preference rather than opting for a starting point based on whatever attracts their attention. At least at the sentence level, children thus appear rather immune to attentional cueing. Not only did children stick to their preferred sentence structure (i.e., SVO actives) when attention was drawn to the patient, but they also did not show any increase in speech-onset times or number of disfluencies when the patient was in the focus of attention. These findings suggest that attentional focus, although effectively guiding gaze patterns, does not (yet) carry over to influence children's sentence production. Note, that this is in stark contrast to findings for adult German speakers who have been found to produce significantly mores passives in describing event scenes with a visually cued patient in a study employing a similar design with comparable materials (Esaulova et al., 2021a).

Effects of animacy on child language production

Unlike for attentional orienting, we did not observe any effects of animacy on children's viewing and production patterns in the noun-pair naming task. While this appears surprising, it is not unprecedented. Rather, previous studies have also yielded inconsistent findings for the effects of animacy at the phrase level. For instance, McDonald et al. (1993) did not observe any animacy effects on adult speakers' production of conjoined noun phrases (e.g., McDonald et al., 1993). By contrast, other studies have found effects of animacy at the phrase level: specifically, German-speaking children were more likely to first mention an animate than an inanimate entity in response to picture stimuli (e.g., Dolscheid et al., 2024; Franz et al., 2021). This raises the question why we did not observe similar effects in our study. Several factors may have contributed to this discrepancy. Importantly, previous studies show that animacy is not a simple, dichotomous category (animate vs. inanimate) but rather leads to nuanced and often surprising interpretations (e.g., Radanović et al., 2016, also see Westbury, 2023 for discussion). For instance, even adults sometimes rate animals as equally high or even higher in animacy than human beings (Radanović et al., 2016, also see Westbury, 2023 for discussion). Note in this respect that previous studies who found animacy effects on children's production of conjoined noun phrases compared pictures of inanimates vs. animals (such as a bunny, see e.g., Dolscheid et al., 2024; Franz et al., 2021). By contrast, our study compared pictures of inanimate objects to pictures of human characters (e.g. clown, dwarf, etc.). Thus, children in our study may have judged the depicted human characters not as “sufficiently” animate. Furthermore, since children tend to attribute animacy to entities capable of self-induced motion and self-propulsion velocity (Torabian and Grossman, 2023), the use of static images in the present study may have attenuated effects of animacy. Also, stimuli were cartoon-like depictions of characters, hence not very realistic (e.g., unlike photos). Finally, children often encounter depictions of inanimate entities that still bear features of animacy such as talking cars or flying carpets in cartoons, also blurring the line between animate and inanimate entities. Although in the present study, we took care to use pictures of inanimate entities devoid of any animacy features, it could still be that repeated exposure to non-living yet “animate-like” cartoon characters has mitigated the animacy effect at the phrase level.

In contrast to the noun-pair naming task, for which we did not observe any effect of animacy (neither in language production nor in viewing patterns), in the event-description task, children were more likely to first fixate the patient character when it was animate rather than inanimate. This finding points to the interesting possibility that children were more sensitive to the animacy distinction in the context of transitive events (i.e., where they had to figure out who did what to whom). In line with this observation, it has been argued that effects of animacy primarily play a role for grammatical role assignment (i.e., for mapping grammatical roles to thematic roles in the context of transitive sentences), but that animacy should be rather irrelevant for processes of linear ordering like coordination (e.g., McDonald et al., 1993). However, while our findings suggest that children were sensitive to animacy information, these effects did not carry over to sentence production in the context of transitive events. Thus, we found no effect of animacy on sentence structure nor on any other measure related to processing affordances (i.e., speech-onset times, MLU, and disfluencies). Critically, unlike in many other studies, here we manipulated the animacy status of the patient while the agent was always animate. This was based on previous observations in adult speakers of German who produced a greater number of passives in response to animate vs. inanimate patients, even though the animacy of the agent was kept constant (Esaulova et al., 2019, 2021a). By contrast, in a similar set-up, the German children in our study did not produce more patient-first utterances in response to patient animacy. Again, in line with the arguments presented above, this is likely because children are less proficient in their production of non-canonical sentences (e.g., Marchman et al., 1991). Thus, children's strong preference for producing canonical active sentences may have overruled potential effects of animacy in the event-description task.

Effects of position on child language production

While previous work has demonstrated strong spatial biases in adults' picture descriptions (e.g., Dolscheid et al., 2024; Esaulova et al., 2019, 2021a,b), our results did not reveal any effect of referent position on children's language production. Importantly, by including eye-tracking measures, our findings grant new, additional insights into children's viewing patterns. As our findings suggest, preliterate children did not display any spatial biases concerning the position of a referent, neither in the noun-pair naming task nor in the event-description task. Thus, children were about equally likely to first fixate a character when presented on the left vs. on the right side of a picture, suggesting that there were no left-to-right biases in preliterate preschool children's viewing patterns. This result is interesting because it contrasts with some previous demonstrations that spatial biases are already present in children, even if they are preliterate (e.g., Patro and Haman, 2012; Rugani and de Hevia, 2017 for review). For instance, in the domain of number, preliterate children were more likely to associate lower numbers with the left side of space and vice versa for higher numbers (e.g. Patro and Haman, 2012). These findings have been attributed to other spatial patterns entrenched in cultural conventions, such as starting to turn the pages of a book from the leftmost page (for instance, when adults read aloud to children). However, although children may pick up on such conventions from quite early on, here we find that similar spatial biases did not arise in the context of picture-description tasks. Furthermore, our findings are at odds with the proposal that spatial biases may be innate and caused by hemispheric asymmetries in brain organization (e.g., Rugani and de Hevia, 2017). Instead, we replicate previous findings in that the position of an entity is not yet decisive for preliterate children's sequence of producing conjoined noun phrases (see also Dolscheid et al., 2024). Furthermore, we extend these findings by investigating sentence production in response to visually depicted transitive events. Here, too, children's production patterns seemed unaffected by referent position, regardless of the measures examined (ranging from eye-tracking to sentence structure and planning costs such as speech-onset times). In sum, these findings provide quite consistent evidence that preliterate children do not yet show spatial biases that become manifest in sentence production. While future work is needed to examine when exactly children come to develop spatial preferences (akin to adults), for now, our results suggest that learning how to read and write seems a necessary requirement for these biases to emerge.

Revisiting the role of prominence in child language production strategies

Our findings suggest that prominence-related factors critically interact with the complexity of the syntactic structure that has to be produced. We found that even though attentional orienting and animacy effectively shaped children's eye-gaze patterns, these factors did not influence children's production of transitive sentences in the event-description task. By contrast, many studies have shown that adults are more likely to start their utterances with the referent that is most prominent and captures their attention, even if this applies to the patient of an event (e.g., Esaulova et al., 2021a; Gleitman et al., 2007; Myachykov et al., 2012). This has been interpreted to reflect a strongly or radically linear incremental sentence planning strategy (see e.g., Gleitman et al., 2007). Specifically, a speaker starts with the encoding of the most accessible concept, which then becomes the first increment in the sentence to be uttered. In case this is the agent of an action, an active SVO structure is produced. In case the patient is the most accessible element, starting the utterance with the patient will result in the production of a passive sentence. The first lemma retrieved from the mental lexicon thus constrains the structure of the sentence as it determines its starting point. While this planning strategy is often employed by adults (e.g., Gleitman et al., 2007 for English speakers, Esaulova et al., 2021a for German speakers), our findings suggest that children's planning of sentences does not proceed in the same way. Thus, even though children's attention was effectively modulated in the event-description task, they did not simply start to produce a sentence with the patient character when it was in the center of their attention. In fact, it appears that children were not even tempted to consider this option. Specifically, although we might have expected that children show higher planning costs when the prominence of the patient increased (e.g., by producing more disfluencies or delays in initiating verbal responses in line with some previous studies, e.g., Charest et al., 2016, also see Esaulova et al., 2021b, for adults), we did not see any of these effects in the present study.

These observations and our finding that children persisted in producing canonical SVO sentences regardless of their visual attention suggests that children may engage in more conceptual and structural pre-planning of the utterance before starting articulation. That is, while adults may already start their utterance with the first accessible increment available, children seem to first carefully evaluate the event and the roles of the elements involved in the scene and link these roles to the grammatical functions of subject and object in their preferred canonical SVO sentence template before starting the utterance. This difference in sentence planning strategy may suggest that children aged 4 to 5 years have not yet reached the level of proficiency to employ a radically linear incremental sentence planning strategy which may be too demanding for them. But why should such a linear incremental sentence planning be more demanding? Although at first glance, planning a sentence in rather small increments (e.g., word by word) might seem a more convenient and even easier strategy, this entails that sentence production and articulation start before the speaker has conceptually and structurally planned the rest of the utterance ahead. This means that while articulating, speakers will have to continue planning what they want to say next, thereby “juggling” multiple different production processes at once. While adults are able to master this process and tend to show more radically linear sentence planning in response to prominence-related cues in a comparable set-up (see e.g., Esaulova et al., 2021a), the concurrency of planning steps involved in articulation and in the conceptual and structural planning of the remaining utterance might still overstretch the language production abilities of the child speaker. Furthermore, in the specific case of transitive events, simply starting the utterance with a small increment could mean that this increment turns out to be the patient rather than the agent of an event, which might require the production of a dispreferred non-canonical structure like a passive or a later revision of the starting point. Unlike adults who seem more apt to flexibly adapt to this challenge, children might be more conservative and thus less likely to employ a radically linear sentence planning strategy in the event-description task. While tentative, support for this assumption and for the notion that children require more pre-planning of who-does-what-to-whom before initiating an utterance comes from the comparison of speech-onset times and utterance-initial disfluencies between the noun-pair naming and the event-description task in our study. Even though the same characters had to be named in the two tasks, the event-description task led to much longer speech-onset times and was accompanied by more utterance-initial disfluencies during production compared to the production of syntactically simpler conjoined noun phrases in the noun-pair naming task, observations that indicate increased syntactic affordances in the event-description task.

In summary, our findings might suggest that children have not yet reached the proficiency in utterance planning and the required flexibility that allows for radically linear incremental sentence production in adult speakers. Thus, the ability for such linear incremental sentence planning that entails the necessity to proceed with conceptual and structural planning while already speaking seems to require and evolve with language production experience. To determine when this ability becomes operative in children, further research is necessary.

Limitations of the current study

An obvious limitation of the current study concerns the question of sample size and statistical power. Specifically, it can be debated whether the absence of statistical effects really implies the evidence of absence of prominence-related effects, especially in the context of the event description task. However, critically, in the noun-pair naming task, we found very clear effects of attentional orienting on both children's eye-gaze and their order of mention, suggesting that the sample size and the statistical power was sufficient to observe effects at least under certain conditions. In addition, the effects of prominence-related factors (cueing and animacy) on eye-gaze patterns in the event-description task again suggest that—in principle—the sample size was sufficient to observe effects, indicating that the current experiments were of sufficient statistical power.

Unlike in previous studies with adults (e.g., Esaulova et al., 2019, 2021a), in the present study, the three experimental factors (animacy, attentional cueing, and position) were examined in isolation. By implication, this means that our results do not speak to potential interaction effects between the various factors. However, since a full factorial design would have required a greater number of trials, the experiment would have taken too much time and exceeded the attention span of 4- to 5-year-old children (and hence was not feasible with children of this age).

Finally, while previous work on adults forms the background of the present study (especially, the comparable set-up in Esaulova et al., 2021a), the focus of our study is exclusively on children. This implies that any comparison between children and adults is limited, as it is only indirect. Note, however, that we adopted a design and materials that were as similar as possible to the study of Esaulova et al. (2021a) while still being appropriate for testing children. Although our results suggest that children seem to differ from adults when it comes to the influence of prominence-related cues on language production and sentence planning strategy, future work is necessary to examine this in more detail. Specifically, more fine-grained time-course analyses of children's eye-gaze patterns might provide a fruitful avenue to investigate the degree to which children employ a more hierarchical sentence planning strategy in the context of transitive sentences. Furthermore, a direct comparison between children and adults would be warranted but would require testing both groups with the identical set-up and materials, a desideratum that is challenged by differences in e.g., attention-span, compliance, and strategic behavior.

We found different effects of prominence-lending factors on child language production, suggesting that not all factors are equally effective in shaping children's verbal responses. Whereas we did not observe any effects of animacy and position in our sample, we observed significant effects of attention. Specifically, children were more likely to first fixate on a character when it had been visually cued, and they were also more likely to first name this character. Furthermore, first fixations were also predictive of children's order of mention, suggesting a tight link between visual attention and language production. However, while we provide some of the first evidence that children's language production patterns are already susceptible to bottom-up attentional orienting, this effect seems restricted to the phrase level (i.e., the production of conjoined noun phrases) rather than influencing grammatical role assignment or sentence planning. Thus, our findings show that children's propensity to verbalize what is in their focus of attention is modified by syntactic complexity. When attentional and production biases compete, as is the case when the patient of a transitive event is visually cued, children do not simply start their sentence production based on what captures their attention but rather resort to their preferred sentence structure (i.e., active SVO sentences). Our findings hence suggest that in planning and producing sentences for describing event scenes, preschool children lack the proficiency to apply a radically linear incremental planning strategy that involves simultaneous processes of articulation while still conceptually and structurally planning the utterance ahead.