The Influence of books’ textual features and caregivers’ extratextual talk on children’s science learning in the context of shared book reading

Abstract

During early childhood, reading books with one’s caregiver (shared book reading) is a valuable means of supporting learning. Yet there are gaps in our understanding of the influence of shared book reading on young children’s science learning. The current research bridges this gap by examining the pedagogical quality of science books in preschool-aged children’s environments and investigating how such books influence children’s learning and caregivers’ extratextual talk during shared book reading. In Study 1, we coded 60 science books that were readily available in young children’s environments within the United States. The books were designed for English speakers and recommended for preschool-aged children. We coded the books for the extent to which they were Coherent (elaborated on facts presented by providing details, examples, comparisons, etc.) and included Embedded Questions. However, many books were low in Cohesion and Embedded Questions and thus were of low pedagogical quality. In Study 2, we tested 38 4- to 5-year-old children (55% female, 76% white) and their caregivers in the Southeastern United States. We assessed the influence of books’ levels of Cohesion and Embedded Questions and of caregivers’ Elaborative extratextual talk during shared book reading on children’s science learning. Children learned more from books high in Cohesion, irrespective of levels of Embedded Questions and caregivers’ Elaborative talk. Additionally, children learned more from books high in Embedded Questions when caregivers used more Elaborative talk. This research highlights the importance that books’ textual features and social interactions during shared book reading have in promoting early science learning.

Early learning before formal schooling, including learning in STEM domains (Science, Technology, Engineering, and Math), relates to children’s later academic achievement (Duncan et al., 2007; Hirsh-Pasek, 2009; Verdine et al., 2017). Young children gain knowledge and interest in STEM domains through various types of activities, such as during play (Bulunuz, 2013; Jirout & Newcombe, 2015), museum visits (e.g., Haden, 2010; Pagano et al., 2020), and shared book reading (Ganea et al., 2011; Kelemen et al., 2014). In this research, we focus on shared book reading as a means of influencing children’s early STEM learning in the domain of science. Shared book reading is the practice of reading books with a knowledgeable other, such as a caregiver. It is common in homes (Bus et al., 1995) and preschools (Hindman et al., 2008, 2012). There are a variety of young children’s expository (factual) books on science topics, such as living creatures, the human body, geography, nature, plants, and physics (Pentimonti et al., 2010). Reading science books to preschool or early elementary school-aged children facilitates learning of science concepts, such as balance (Larsen et al., 2020), camouflage (Ganea et al., 2011; Strouse & Ganea, 2021), nutrition (Gripshover & Markman, 2013), natural selection (Kelemen et al., 2014), and characteristics of animals and physics (Mantzicopoulos & Patrick, 2010). These findings indicate that shared book reading with science books can be valuable for early science learning. Yet, there remain gaps in our understanding of whether young children’s science expository (factual) books align with research on reading and learning and how the books’ design and shared book reading interactions affect early science learning.

To address these gaps, we build from and combine insights from distinct lines of research on (a) shared book reading with young children; and (b) text comprehension and learning, primarily with adult and school-aged child readers. By combining insights from these two lines of research, we gain a deeper understanding of young children’s comprehension and factual learning from expository science books. We also further understanding of the potential of expository science books to be a valuable source for children’s early science learning.

Young children’s learning during shared book reading interactions

Research on shared book reading with preschool-aged children and their caregivers identified variations in caregivers’ extratextual talk while reading, mainly with fictional narrations, and found that such variations affected children’s language and literacy skills (Breit-Smith et al., 2017; Haden et al., 1996; Hindman et al., 2008; Kang et al., 2009; van Kleeck et al., 1997). One important style variation is the extent that caregivers elaborate on books’ concepts. Elaborative talk includes making inferences, retrieving earlier learned information, drawing comparisons between textual elements or between the text and the child’s life, defining concepts, and providing explanations (Blewitt et al., 2009; Haden et al., 1996; van Kleeck et al., 1997). Caregivers who use more elaborative talk during shared book reading tend to have children who score higher on measures of language and literacy, such as vocabulary, story grammar understanding, and story comprehension (Breit-Smith et al., 2017; Dunst et al., 2012; Haden et al., 1996; Hindman et al., 2008; Reese & Cox, 1999). Elaborative talk is thought to promote language and literacy skills because it is cognitively demanding and engages young children in strategies that help to promote their comprehension and build on their prior knowledge (Hindman et al., 2008; Price et al., 2009; Smolkin & Donovan, 2003).

Although most research focused on fictional narrations, some studies investigated similarities and differences between fictional narrations and expository books, including science books (Anderson et al., 2004; Pellegrini et al., 1990; Price et al., 2009; Robertson & Reese, 2017; Torr & Clugston, 1999). Such work found that caregivers tended to use more extratextual talk with expository books compared to fictional narrations and their speech was more elaborative (Anderson et al., 2004; Price et al., 2009). Additionally, caregivers’ extratextual talk with expository books predicted children’s language and literacy skills, similar to relations found with fictional narrations (Robertson & Reese, 2017).

In sum, extratextual talk is a variable influencing children’s learning from books, with expository books having added benefits of eliciting more elaborative extratextual talk from caregivers. However, there are aspects of this research that would benefit from further inquiry. One is that past research has not focused on how caregivers’ extratextual talk during expository book reading influences children’s knowledge acquisition, including science knowledge. Instead, the focus has been on language and literacy skills. Research that has investigated learning science from books during early childhood has had an experimenter read the books, without using or measuring extratextual talk (e.g., Ganea et al., 2011; Heisey & Kucan, 2010). Elaborative extratextual talk would likely be as important for children’s knowledge acquisition as it is for their language and literacy skills (Breit-Smith et al., 2017; e.g., Hindman et al., 2008) because it could help to facilitate a deeper understanding of the text. Another aspect that would benefit from further inquiry is to identify which features of preschool-aged children’s books support knowledge acquisition, shape the nature of extratextual talk, and interact with such talk to influence knowledge acquisition. To address these gaps, we turn to research primarily with adults and school-aged child readers on text comprehension to gain insights into the types of textual features of books that likely support early factual learning in the context of shared book reading.

Textual Features of Books Influencing Learning

Research on textual features of books identified cohesion and embedded questions as important textual features that can support factual learning, including science learning. Cohesion is the extent to which book text links ideas and concepts (Best et al., 2005; Kintsch, 1994, 1998; van den Broek et al., 1999). Texts are cohesive when they contain introductory and summary paragraphs, include semantic overlap between adjacent paragraphs and/or between adjacent sentences, draw connections to earlier text, as well as define and explain concepts. Embedded questions are queries inserted within the text that prompt a response.

Cohesion is an important textual feature because it promotes individual readers’ formation of a coherent mental representation of the textual information (i.e., understanding of text that connects to readers’ prior knowledge). It supports comprehension by reducing the need for readers to draw certain types of inferences and by decreasing the amount of active processing required to understand the text (Kintsch, 1998; van den Broek & Helder, 2017). For example, when text is coherent the reader does not need to draw as many inferences between elements of the text or between the text and their prior knowledge as the text is directly linking ideas and concepts across sentences and paragraphs and includes definitions and examples. This reduced need to make inferences also makes active processing more manageable as the reader does not need to hold as much information in their working memory to form inferences (van den Broek et al., 1999). Additionally, their increased comprehension of high cohesive texts reduces the need for readers to engage in deliberate strategies such as rereading and reflection to understand the text (van den Broek et al., 2011; van den Broek & Helder, 2017). Active processing is important for one’s comprehension and memory of text; but when a reader has low prior knowledge, they are unable to draw as many connections to their prior knowledge. It can then become difficult to hold the textual information in memory and actively make inferences (Linderholm et al., 2000; McNamara et al., 1996).

Across multiple studies, research showed that increasing text cohesion supported comprehension and learning among adults and school-aged children (Beck et al., 1991; Britton & Gülgöz, 1991; Gilabert et al., 2005; Linderholm et al., 2000), including learning from science texts (cf. van den Broek, 2010 for review). A few studies have been conducted with younger children in the context of listening to fictional narrations. Such research found patterns similar to those observed with older children and adults, whereby increasing the texts’ causal connections, or implementing a hierarchical goal structure increased 4–6-year-old children’s recall of story events and inferences made (Lynch et al., 2008; Lynch & van den Broek, 2007). Although there is limited research on the impact of text cohesion in reading with young children and no known research has been conducted within the context of expository books, it is logical to assume that young children would benefit from expository books high in cohesion. Young children have low prior knowledge, have difficulty in making inferences and integrating information (Bauer & San Souci, 2010; Cain et al., 2001; Filiatrault-Veilleux et al., 2016), have low working memory capacity (Cowan, 2014), and typically do not engage in deliberate strategies (Coffman et al., 2008; van Kleeck et al., 2006). Each of these skills and processes supports one’s comprehension of texts that are low in cohesion (Linderholm et al., 2000; McNamara et al., 1996; van den Broek et al., 1999, 2011; van den Broek & Helder, 2017). Cohesion is thus likely an important element in supporting early factual learning from books.

In addition to Cohesion, another important textual feature is embedded questions. Decades of research on testing effects demonstrated that having students actively process information both by asking pretest questions before the material is learned and by asking retrieval questions as a form of study enhances student learning (Karpicke et al., 2016; e.g., Karpicke & Grimaldi, 2012; Pressley et al., 1990). In terms of reading specifically, embedding questions within the text has been shown to support adults’ and school-aged child readers’ comprehension and factual learning from text (Callender & McDaniel, 2007; Hamaker, 1986; B. L. Smith et al., 2010). For example, B. L. Smith et al. (2010) found that prompting students approximately every 150 words with a question related to material learned from a science textbook helped the students learn the science material. Additionally, Callender and McDaniel (2007) found that questions embedded within the text were particularly helpful for learning among individuals who tend to perform worse on structure building. Structure building is a measure of the ability to form a coherent mental representation by mapping text read onto one’s existing knowledge (Gernsbacher, 2013). It is thought that embedded questions facilitate reading comprehension and learning because the questions help the reader learn specific information asked and reduces the likelihood that the reader processes irrelevant information (e.g., Callender & McDaniel, 2007). This then helps one form a coherent mental representation of the text as the questions help one to integrate the relevant content. Based on these findings, it is logical to assume that embedded questions are beneficial when texts are low in cohesion and readers struggle to form a coherent mental representation of the text.

Regarding young children, embedded questions, as well as after-text questions, have also been shown to support learning. Such questions support language and literacy skills (Ard & Beverly, 2004; van Kleeck et al., 2006), and science learning (Heisey & Kucan, 2010; Leung, 2008), as well as support caregivers’ extratextual talk (Troseth et al., 2020). For example, in Heisey and Kucan (2010), 1^st and 2^nd-grade students were read science texts during read-alouds, with texts that either contained questions embedded within them or questions asked afterward. They found that students in the embedded question group performed better on posttest science questions. In terms of preschool-aged children, Leung (2008) had 3- to 4-year-old children listen to science books during read-alouds over 4 weeks, with half the children retelling the stories after being read them. Children in the retelling group increased their science vocabulary relative to children who did not retell the stories. In terms of extratextual talk, eBooks that include dialogic questions engaged caregivers in more on-topic extratextual talk with their preschool-aged children and in more question asking, even on pages without questions (Troseth et al., 2020).

In sum, textual elements of cohesion and embedded questions are important features that support comprehension and learning from the text. However, open questions remain as to how these features facilitate young children’s text comprehension and learning from science books during shared book reading. Specifically, it is unknown whether cohesion influences preschool-aged children’s science learning and whether and how books’ cohesion interacts with embedded questions to support young children’s science learning. Additionally, it is unknown how cohesion and embedded questions influence elaborative talk and how they interact with elaborative talk to support children’s text comprehension and learning. One important barrier to addressing these questions is that it is unknown the extent to which cohesion and embedded questions are represented within young children’s books. With regards to cohesion specifically, cohesion has not been operationalized within preschool-aged children’s expository books. Unlike textbooks or fictional narrations on which research on cohesion has been conducted, young children’s expository books do not include paragraphs with introductions, conclusions, and transitions. Nor do they include a storyline with a sequence of events that cohesively builds throughout the story. Instead, they typically have a general theme and present multiple facts on different topics in a non-sequential order.

Current Research

In the current research, we bridge gaps in understanding how young children learn science information during shared book reading by considering both the role of books’ textual features and caregivers’ elaborative extratextual talk. In Study 1, we operationalized Coherence and Embedded Questions within preschool-aged children’s science books and then conducted a corpus analysis to assess the prevalence of these features across the books. As reviewed above, young children’s expository books are not designed in the same way as school-aged children’s and adults’ books. Thus, we needed to define these textual features based on how young children’s science books naturally vary within their environment. In Study 2, we conducted an experiment to test how variations in levels of Coherence and Embedded Questions, as operationalized in Study 1, influence children’s learning during shared book reading. We selected 4 science books from the corpus analysis and evaluated whether books’ textual features affect children’s recall of science facts, influence the nature of caregivers’ elaborative extratextual talk, and interact with elaborative talk to influence children’s recall of the books’ science facts.

Study 1

In Study 1, we conducted a corpus analysis on readily available early childhood science books. The goals were to evaluate how textual features of Cohesion and Embedded Questions are instantiated within such books and to examine the prevalence of such supportive textual features. For young children’s science books to be a valuable resource for learning, the books’ textual features should include elements that are supportive of such learning. Prior research has conducted corpus analyses, primarily with elementary school-aged children’s science books, investigating such elements as alignment with science literacy goals and school curriculum (Ford, 2006; Newton et al., 2002; Schroeder et al., 2009; Smolkin et al., 2009), the accuracy of information (Rice, 2002; Trundle et al., 2008), as well as gender and race representation (Kelly, 2018). These are all important textual features to consider when evaluating science books. However, it remains unknown whether young children’s science books are designed in alignment with cognitive science research on ways that book text generally supports reading and learning. We thus sampled early childhood science books and designed a coding scheme that would capture how supportive textual features of Cohesion and Embedded Questions are operationalized within such books. We then conducted a corpus analysis to understand the prevalence of these textual features within the books.

Method

Corpus

The corpus included 60 science books designed for preschool-aged children (see Appendix A a for list of books). It included books on the following science topics: animals (15), plants and earth (13), human body (8), vehicles (8), weather (6), space (5), and science other (5). The books were written by 45 different authors. Books were selected if they met the following criteria: (1) designed for English speakers within the United States; (2) listed on children’s bestseller book lists (based on the New York Times and Amazon bestseller lists) and/or currently sold at bookstores; (3) available at local metropolitan public libraries where the research was conducted; and (4) recommended for preschool-aged children. It was important that the books were readily available through both bookstores and local libraries as that is indicative that the books are available to children across socioeconomic statuses. To create the corpus, a research assistant transcribed each sentence in the book and a different research assistant checked the accuracy. Two additional books were excluded for being outliers on our measures of Cohesion. Those books were more than 3.5 standard deviations from the mean, and had a significant influence on the analysis, such that each book’s inclusion caused significant changes in how the other books clustered.

Operationalization of Cohesion and Embedded Questions

We used a sampling of 10 books (17%) of diverse topics and lengths within our corpus to identify how Cohesion and Embedded Questions were represented within children’s science books. We then applied the operationalization to the rest of the corpus. In terms of Cohesion, we identified that children’s science books varied in how often they switched topics and elaborated on concepts. Books that switched topics would, for instance, change subjects and switch between mentioning different types of animals (e.g., anteater, hummingbird) or mentioning different features of animals (e.g., fastest animal, longest animal). Books also varied in the extent to which they elaborated on the main points of each topic versus extended the topic. Elaborating on the topic included: 1) providing details, examples, or definitions; 2) providing analogies or antonyms to the facts; 3) previewing information; or 4) making connections and referencing earlier learned facts. Extending topics included providing new information related to the topic, but not reinforcing the main ideas. The sentences that are categorized as elaborative are the types of sentences that help to make text cohesive (Best et al., 2005; Kintsch, 1998). This conceptualization of elaboration is also similar to causal connections described in the narrative literature in which children remember more information when there are more casual connections (this is an example of a causal connection described in Lynch et al. (2008), “Rat liked to copy Cat.” is causally connected to the sentence “Rat built a house that was just like Cat’s”).

An example of how sentences were coded as elaborative versus extension is provided in Table 1. For example, one of the topics is “Insect Builders.” Within this topic, the introductory sentence is “Some insects are good builders.” The elaboration sentences provided examples of how the insects were good builders (e.g., “Ants build tunnels”; “Bees build honeycomb of wax that comes from their bodies.”). In contrast, extension sentences provide new information that does not directly support the concept that insects are builders (e.g., “These insects live in large communities where each helps the other.”). Thus, when a book includes more elaboration sentences, the sentences likely strengthen the main points. This may result in increasing children’s learning, similar to how children learn more narrative information with more causal connections (Lynch et al., 2008). When the books include more extension sentences as well as more topics, then one may learn more different types of information, increasing the size of their semantic network, but with less activation strength for information in the network

Table 1.

Sample Coding of Cohesion and Embedded Questions

There are many types of insects in the world.	Introductory	Statement	Features of Insects
Some insects are ants.	Elaboration	Statement	Features of Insects
Other insects are crickets and mosquitoes.	Elaboration	Statement	Features of Insects
Here’s how you know if it’s an insect.	Extension	Statement	Features of Insects
Count each of its legs.	Elaboration	Interactive Prompt	Features of Insects
Count the number of body parts each one has.	Elaboration	Interactive Prompt	Features of Insects
There are types of insects who are great builders.	Introductory	Statement	Insect Builders
Ants make tunnels from dirt.	Elaboration	Statement	Insect Builders
Wasps make nests out of paper.	Elaboration	Statement	Insect Builders
These builder insects live in big communities.	Extension	Statement	Insect Builders

Note. Sample excerpts paraphrased from Bugs are Insects by Anne Rockwell.

Embedded Questions was operationalized as book text that prompts a response from the child related to the book’s facts. This includes instances when the book poses a question (e.g., “what shape is water?”) or provides a request (e.g., “count the sides on each snowflake”).

Coding and Data Reduction

Coders rated the levels of Cohesion and Embedded Questions for each book. First, coders identified the number of topics per book and assigned each sentence to a topic. All on-task sentences, which are sentences related to facts within the books, were coded for topic. Sentences not considered on-task were non-factual sentences (e.g., “I love the summer”; “Listen close…cuck-la-ree!”), activities to be conducted outside the books (e.g., an experiment, building instructions), or section headings.

For Cohesion, we calculated the number of elaboration sentences per topic. To identify elaboration, coders first identified an introductory sentence per topic that conveyed the main point of the topic. Then starting with the introductory sentence, coders rated whether the other sentences within the topic were extensions or elaborations. Extension sentences provided new information from the introductory sentence and all other sentences within the topic. Elaboration sentences build on either the introductory sentence or an extension sentence. There could be multiple elaboration sentences per introductory or extension sentence within a topic if, for example, the text elaborated by providing multiple examples or made an analogy and provided further detail. Books with fewer topics and more elaboration sentences were higher in Cohesion than books with many topics and few elaboration sentences. For Embedded Questions, we calculated the number of questions or interactive prompt sentences out of the total number of on-task sentences. Embedded Questions coding was separate from the Coherence coding, and thus a sentence could be coded as providing an embedded question irrespective of whether it was coded as an introduction, extension, or elaboration. The Embedded Questions category was coded out of the number of on-task sentences rather than topic, as topic is related to Cohesion and Embedded Questions is independent of Cohesion.

Two research assistants served as coders, and each rated about half the books. Seventeen books were double-coded (28%). The first author resolved discrepancies. For Coherence coding (ratings of whether the sentences were introductory/extension or elaboration), there was 89% agreement, with Cohen’s Kappa at .78. For Embedded Questions coding (rating of whether the sentences provided questions or interactive prompts or were statements), there was 99% agreement, with Cohen’s Kappa at .98. For the number of topics, we conducted a correlation between each coders’ ratings of topics per book and found a correlation of r = .96. The average absolute value difference in agreement between the two coders’ ratings of the number of topics counted per book was less than 1 topic (0.35). The coding manual, data, and R scripts are available on Open Science Framework (Miller-Goldwater et al., 2022). This study was not preregistered.

Results and Discussion

The primary goal of the analyses was to understand how the 60 books clustered based on Cohesion and Embedded Questions. As shown in Table 2 (All Books column), books on average provided approximately 3 elaboration sentences per topic (Cohesion measure), and less than 1 question or interactive prompt per topic (Embedded Questions measure). There was also large variability in books’ levels of Cohesion and Embedded Questions. In terms of Cohesion, books ranged from including 0 to 15 elaboration sentences per topic. In terms of Embedded Questions, books ranged from including 0 to .35 questions or interactive prompts per on-task sentence.

Table 2.

Results from cluster analysis

Cluster Characterization		High Cohesion-High Embedded Questions	High Cohesion-Low Embedded Questions	Low Cohesion-High Embedded Questions	Low Cohesion-Low Embedded Questions	All Books
Quantity of books	number	9	10	15	26	60
	%	15	17	25	43	100
Cohesion	m	5.23	8.34	2.02	1.79	3.46
	sd	1.95	2.98	1.37	1.46	3.08
	z	0.58	1.59	−0.47	−0.54	0.00
Embedded Questions	m	0.25	0.04	0.13	0.01	0.08
	sd	0.05	0.05	0.05	0.02	0.09
	z	1.80	−0.46	0.49	−0.73	0.00
Within cluster SS	-	5.93	10.57	6.68	6.75	29.93

For the cluster analysis, we normalized the measures of Cohesion and Embedded Questions by using z-scores and running a k-means cluster analysis in R using the factoextra package. This method groups data such that the sum of squares between a predetermined number of clusters and the cluster center is minimized (Hartigan & Wong, 1979). We determined the number of clusters by observing dendrogram, scree, and silhouette plots. For the dendrogram plot, we used average linkage and saw that all books grouped into 4 discrete clusters. The scree plot showed that adding more than 4 clusters would have a minimal impact on improving the within-group sum of squares. Lastly, the silhouette plot showed that the average silhouette width was .43. We decided that 4 was an appropriate number of clusters for our data. In Table 2, we present descriptive statistics of the results from the k-means clustering method for the 4 clusters. In Figure 1, we present each book by its level of Cohesion and Embedded Questions and their cluster category. As seen, the data clustered into categories characterized by: 1) High Cohesion-High Embedded Questions; 2) High Cohesion-Low Embedded Questions; 3) Low Cohesion-High Embedded Questions; 4) Low Cohesion-Low Embedded Questions. The High Cohesion-High Embedded Questions category included a few books that were average cohesion (z ~ 0), but most books were higher in Cohesion. There was most variability in the High Cohesion-Low Embedded Questions category, with a few books scoring very high in Cohesion. The largest cluster was the Low Cohesion-Low Embedded Questions cluster, which included 43% of the books. Most books were categorized into clusters with an element of Low Cohesion (68%) or with an element of Low Embedded Questions (60%).

Figure 1.

Books’ clustering by Cohesion and Embedded Questions. EQ = Embedded Questions. Large shapes represent the center point of each cluster. Small shapes represent individual books.

Overall, Study 1 identified a means of characterizing early childhood science books based on their potential pedagogical quality. Prior research focused on the element of Cohesion with texts for adults and school-aged children. We successfully developed a scheme that captures Cohesion in young children’s science books, while also considering the element of Embedded Questions. Our findings are significant in identifying that, whereas early childhood science books can potentially be a valuable means of supporting early science learning (e.g., Kelemen et al., 2014; Mantzicopoulos & Patrick, 2010), many books were characterized by elements that are likely not optimally supportive of learning. This brings us to Study 2 in which we assessed whether and how children learn from books that vary based on the textual features of Cohesion and Embedded Questions during shared book reading.

Study 2

In Study 2, we conducted a behavioral analysis to examine how books varying in levels of Cohesion and Embedded Questions influenced children’s learning during shared book reading. Across two sessions, caregivers read four books to their 4- to 5-year-old children that were crossed on the books’ levels (high, low) of Cohesion and Embedded Questions. We selected four science books from Study 1’s corpus analysis (Stimuli section describes selection criteria). During each session, caregivers read two books to their children. After reading both books, the children participated in an immediate fact memory task assessing their memory for the books’ science facts. The memory task included Theme questions that assessed generalization of facts across topics and Specific-Fact questions that assessed learning of facts on one topic. Additionally, in Session 1 children participated in a standardized vocabulary task, and in Session 2 in a 1-week delayed fact memory task assessing memory of science facts from Session 1.

For this study, we had 4 research questions. Our first question was how textual features influence children’s recall of the books’ science facts. For Cohesion, we hypothesized that children would benefit from books that are high in Cohesion due to children’s limited prior knowledge, low working memory capacity, and minimal deliberate strategy use (e.g., Coffman et al., 2008; Cowan, 2014). This is because Cohesion supports each of these reader characteristics (e.g., Linderholm et al., 2000; van den Broek et al., 1999, 2011). Additionally, Cohesion would facilitate children’s performance on the Theme questions as books high in Cohesion draw connections between textual elements, emphasizing how the facts generalize across topics. In terms of Embedded Questions, we hypothesized that children will benefit from Embedded Questions, as found in prior research with elementary school-aged children (Heisey & Kucan, 2010). Additionally, Embedded Questions would interact with Cohesion. Research on adults has shown that Embedded Questions are particularly helpful for readers who have difficulty forming a coherent mental representation of the text (Callender & McDaniel, 2007). Since texts low in Cohesion make it more difficult to form a coherent representation, including Embedded Questions within the text should support learning.

Our second question was whether the textual features influence caregivers’ elaborative extratextual talk. We predict caregivers will use more elaborative extratextual talk when books are low in Cohesion. Just like the adult reader who engages in active processing when texts are low in Cohesion (van den Broek & Helder, 2017), caregivers may engage in Elaborative talk to facilitate their children’s formation of a coherent textual understanding. Relatedly, caregivers may have the awareness that the text is harder for their children to process, similar to findings showing that caregivers tend to use more elaborative extratextual talk with expository compared to narrative texts (Anderson et al., 2004; Price et al., 2009; Robertson & Reese, 2017).

Our third and fourth questions were whether Cohesion and Embedded Questions interact with caregivers’ Elaborative talk to support children’s recall at an immediate test and after a 1-week delay. We predict the same factors that influence children’s recall will influence their performance at the 1-week delay. We also predict children will recall more when caregivers use more Elaborative talk, as found in research on language and literacy skills (e.g., Breit-Smith et al., 2017; Dunst et al., 2012). Lastly, we predict that Elaborative talk will be most supportive of recall when the books are low in Cohesion as caregivers’ Elaborative talk may be needed to help their child form a coherent mental representation of the text.

Method

Participants

There were 38 children (M age = 4.81 years, range = 4.15 to 5.57, 21 females). Each child participated alongside their caregiver (37 females) for two sessions. The sessions were approximately one week apart (M = 7.16 days, range = 5–12 days). The data was collected in the Southeastern United States. Based on parental self-report, the child sample was Asian (3%), Black (8%), White (76%), and mixed race (13%); 8% identified as Latinx. Additionally, 95% of participants had at least one caregiver with a 4-year college degree or higher. Caregivers gave informed consent for themselves and their child to participate. An additional four dyads participated but were excluded for: not completing the second session (2), experimenter error (1), being distracted and not trying to answer any of the test questions (1). Participants were recruited through a university research subject pool of families interested in research participation. As compensation for participation, children received a book and a small toy, and families were given a $10 gift card. The protocol was reviewed and approved by the university’s institutional review board.

Stimuli

Books.

Four books on animal facts were selected from Study 1’s corpus analysis. We selected the following books: Biggest, Strongest, Fastest by Steven Jenkins, Bugs are Insects by Anne Rockwell, What Lives in a Shell?¹ by Kathleen Weidner Zoehfeld, and Whose Food is This? by Nancy Kelly Ann (books listed alphabetically not by Cohesion/Embedded Questions levels). Each book fell into a different quadrant of High/Low levels of Cohesion and Embedded Questions. We selected the four books such that they were close together in value to the other books within its pair (e.g., High Cohesion-High Embedded Questions and High Cohesion-Low Embedded Questions books represent the High Cohesion pair). Table 3 presents descriptive statistics.

Table 3.

Descriptive Statistics of Books used in Study 2

Book	High Cohesion-High Embedded Questions	High Cohesion-Low Embedded Questions	Low Cohesion-High Embedded Questions	Low Cohesion-Low Embedded Questions
Cohesiona	8	6.5	2.56	1.73
Embedded Questionsb	0.17	0.04	0.15	0
Word Count	666	650	650	641
Content Word AoAc	283.11	272.44	259.27	257.62

Notes.

^a/bVariables operationalized from Study 1. Cohesion is operationalized as the number of elaborative sentences out of the number of topics. Embedded Questions is operationalized as the number of questions/interactive prompts out of the number of on-task sentences

^cAoA = Age of Acquisition, calculated using Coh-Metrix. The possible range is 100–700 (McNamara et al., 2014).

In addition to representing the four quadrants, these books were selected for a variety of reasons. First, the books were normed for preschool-aged children based on publishers’ ratings and their concepts were of similar difficulty, as reflected by the books’ content words having similar spoken word age of acquisition ratings (see Table 2). We calculated age of acquisition through the computerized natural language processing program Coh-Metrix ratings (McNamara et al., 2014). Second, the books were on the same science topic of animal facts, a common topic written for this age group. Specifically, they were on the topics about (listed based on the alphabetical order of books): 1) record-holder animals; 2) defining features of insects; 3) the function of shells for aquatic animals; and 4) the food animals eat. Third, the books were of similar word length (Table 3). Fourth, all the books featured bright, full-color images with the images highly aligning with the books’ facts.

We ensured that the books were novel to the participants through a questionnaire asking caregivers if their child had prior exposure to any of these books. Five caregivers indicated that their child was exposed to 1 (3 participants) or 2 (2 participants) of the books. These books were removed from the participants’ data for analyses. Additionally, we ensured that the facts tested were novel to children of this age range through pilot testing (described in Pilot testing stimuli).

Tests.

There were two types of tests: science fact memory and verbal comprehension. The science fact memory tests assessed children’s recall of facts from each of the 4 books at an immediate test and from two books after a 1-week delay. There were 24 unique questions (6 questions per book). The questions were presented in an open-ended format. Half were Theme questions relating to multiple facts within the book (e.g., “Why do animals need a shell?”; “Some animals are big and fast. Why is being big and fast helpful?”) and the other half were Specific-Fact questions relating to one topic from the book (e.g., “What food do hummingbirds eat?”; “How many legs does a crab have?”).

Participants read two out of the four books per session. There was an immediate recall test at the end of each session and a 1-week delayed test at Session 2 only. The immediate recall test included 12 questions (six from each of the two books read during that session). The 1-week delayed test assessed recall of the Theme questions from Session 1 and included six questions (three from each of two books read at Session 1). We only asked Theme questions to avoid overloading the child with questions. The Theme questions were the same as those answered in Session 1.

For the verbal comprehension test, we used the standardized Comprehension-Knowledge subtest of the Woodcock-Johnson-III. This test was included as a control measure in light of the known strong relation between caregivers’ elaborative talk and children’s language skills (e.g., Currie & Cain, 2015; Demir‐Lira et al., 2019). This task involves responding to picture vocabulary, synonym, antonym, and analogy questions.

Pilot testing stimuli.

We conducted pilot testing on the books’ facts to ensure that the participants in this age range did not have prior knowledge of the facts being tested from the books and that the books did not differ based on participants’ abilities to spontaneously answer the questions. For the pilot testing, we had 8 4- to 5-year-old children answer the science fact memory task questions without reading the books. The mean score for each book was: High Cohesion-High Embedded Questions = .08 (4/48), High Cohesion-Low Embedded Questions = .06 (3/48), Low Cohesion- High Embedded Questions = .04 (2/48), and Low Cohesion-Low Embedded Questions = .08 (4/48). This indicated that participants in this age group had low prior knowledge of the science facts and that the questions asked across books did not differ based on their difficulty. However, after conducting Study 2, we learned that children had high prior knowledge of one Specific-Fact question asked (“How many legs does a spider have?”; despite low performance on this fact in open-ended pilot testing). We identified this by analyzing the transcripts from the shared book reading interactions and noticing that many participants spontaneously said that spiders have 8 legs before their caregiver read that fact within the book. Subsequently, this question was removed from analyses. There was no evidence of prior knowledge in the transcripts related to the other questions.

Procedure

Shared book reading period.

The sessions were conducted in a laboratory. Both sessions began with shared book reading in which caregivers read two of the four books. Book assignment was counterbalanced within and across sessions, with the constraint that per session, participants received one high and one Low Cohesion book. Before reading, caregivers were instructed to read the books to their child as they would normally at home. They were told that, after reading, their child would be tested on information from the books. During the shared book reading portion, dyads sat together on a couch and were offered snacks. The experimenter left the room to increase the participants’ comfort in the setting. Two video recorders angled slightly behind the dyad on both sides of the couch and one audio recorder taped the session. The shared book reading period lasted approximately 20–30 minutes.

Test period.

After the shared book reading period in each session, the test phase began. During this phase, the experimenter sat with the child at a desk on one side of the partitioned testing space. The caregiver sat at a desk on the other side and wore noise-canceling headphones. The caregiver completed questionnaires while the child was being tested. The headphones prevented caregivers from hearing questions asked to the child, avoiding potential influence on the child’s second session.

In both sessions, children completed the science fact memory task on books read during that session. There was always a short memory activity before the science fact memory task and a short activity after the task. These short memory activities lasted approximately 5–10 minutes each and were used to test further questions that are beyond the scope of the current paper. After the final memory activity in Session 1, children completed the verbal comprehension test; and in Session 2, children completed the science fact 1-week delayed recall test. Stickers were used throughout the session as reinforcements between tasks.

For the science fact memory test, children responded to questions about the first book read during the session followed by the second book. The test began with the experimenter showing the child the cover of the book and telling the child that they would be asked questions about this book. The child was told that if they do not know the answer, they should give their best guess. During the task, the experimenter orally asked the child each question in one of three predetermined random orders. If the child provided a vague answer, the experimenter asked the child if they had another answer to the question. While asking the question, the experimenter marked the child’s responses. The same procedures were used for the science fact 1-week delayed task in Session 2, with the only difference being that the experimenter asked questions about the book read during Session 1. In terms of the verbal comprehension test, the experimenter administered the task using the Woodcock-Johnson’s test booklet and followed the standardized procedures. The test period was video recorded.

Data Reduction and Analyses

Transcripts.

All speech from the shared book reading phase was transcribed. The transcribers had a template of the transcript of the phrases from each of the four books. The transcribers listened to the audio recording of the conversations and transcribed each of the extratextual talk phrases that the caregiver and child spoke at the point of the book where the words were spoken. A phrase was defined as a unit of speech that could stand on its own. A given sentence could be divided into multiple phrases (e.g., this sentence has two phrases “See the ladybug has six legs, and it has 3 body parts.”). The final document had each phrase on one line, with indications of whether the phrase was book text, caregivers’ speech, or child’s speech. While transcribing, if the speech was difficult to understand, the transcriber watched the video to see if they could decipher the speech. If the speech was still too difficult to decipher, the transcriber marked the speech as not codable. For each dyad, one research assistant transcribed all four of the books. A second research assistant checked the transcripts and made corrections.

Extratextual talk coding.

Each of the caregivers’ phrases was coded as to whether it was elaborative and whether it provided unique on-task information. Examples of the coding are provided in Table 4. Elaborative talk was coded similarly to how the books’ phrases were coded as elaborative in the Cohesion measure from Study 1. Specifically, elaborative talk was coded as phrases that provided further details, examples, and definitions; made analogies or antonyms to the child’s own life, known concepts, or facts from earlier in the text; retrieved facts from earlier in the text; and pretested facts by asking questions about facts that would soon be presented. Unique on-task extratextual talk was coded as any unique phrase that was about the books’ text or pictures. This included elaborative phrases plus phrases that provided new information that did not directly support understanding of the book facts (i.e, e.g., book phrase: “shells come in many shapes, sizes, and colors,” parent phrase: “if you put your ear on a shell you can hear the ocean”), describing the pictures, repeating the book facts on the same page, and commenting on the fact or pictures. Phrases that were on-task but were not coded as unique on-task talk repeated an extratextual talk statement, affirmed or corrected an extratextual talk statement (e.g., “yes”; “good job”; “it does not have six legs”), or was a direct response to a question (e.g., if the question was “how many body parts does an insect have? and the response “3”). These phrases were not included in analyses because the caregiver was not providing new factual information beyond the information already stated. The Elaborative talk measure used in analyses was the proportion of caregivers’ elaborative phrases out of their new on-task phrases. This measure captures caregivers, who in referencing the book, provided information beyond the book that we expected would facilitate their children’s learning of the science facts.

Table 4.

Caregivers’ Extratextual Talk Coding Examples

Book Phrases	Examples of Caregivers’ Elaborative Phrases
“Snails use one foot to creep along.”	“Is the snail fast or slow?”
“Insects have hard skeletons outside their body.”	“We are squishy and soft on the outside”
“An ostrich doesnť have teeth, so it eats pebbles.”	“Like rocks”
“An anteater can eat up to 30000 insects in one day.”	“Iťs like half as many people as went to see soccer play when we went that day.”
“Just like you get too big for your clothes, a crab grows too big for its shell.”	“So thaťs different from how the snail's shell and the turtle's shell work right?”
	Examples of Caregivers’ On-Task Non- Elaborative Phases
“Hermit crabs have tough claws in the front of their body.”	“So the hermit crab is so interesting.”
“Some insects have long antennae.”	“Look long ones”
“Whose food is this, crawling up a tiny hill?”	“Who eats this?”
“It uses its sense of smell to find anthills.”	“Thaťs pretty cool.”
“The largest land animal is the African elephant.”	“Have you seen an African elephant before?”

Notes. The excerpts of the books’ texts were paraphrased from Biggest, Strongest, Fastest by Steven Jenkins, Bugs are Insects by Anne Rockwell, What Lives in a Shell? by Kathleen Weidner Zoehfeld, and were quoted from Whose Food is This? by Nancy Kelly Ann.

There were two primary extratextual talk coders who each coded 50% of the transcripts. Coders coded one book at a time for half the participants before moving on to the next book. Coding one book at a time helped the coder focus on that specific book’s facts. Coders primarily coded the extratextual talk from each book based on the transcripts but would watch the video when it was unclear which coding category the phrase fit into. A third coder conducted the reliability coding. The reliability coder coded 25% of each participant’s transcripts. For the OnTask coding per book (On-Task phases/Total phrases), the agreement between the original and reliability coders ranged from 94%−95%, with Cohen’s Kappa ranging between .88 and .90. For the Elaborative talk agreement (Elaborative phrases/On-Task phrases), the agreement ranged from .92-.96, with Cohen’s Kappa ranging between .83 and .91. Both transcribers and coders were unaware of the child’s performance on the science fact memory and verbal comprehension tests.

Test Scoring.

Both the science fact recall and the verbal comprehension tasks were scored during the study by the experimenter. The experimenter marked the child’s responses and scored whether the response was correct or incorrect. A research assistant later checked the responses and entered the data. A second research assistant checked the data entry. For the science fact recall test, the measure was a proportion score of the number of correctly answered items out of the total number of items. For the verbal comprehension test, we standardized the raw scores by age using the Woodcock-Johnson III’s scoring system. The coding manual and R scripts are available on Open Science Framework (Miller-Goldwater et al., 2022). The data is available upon request as specified on the consent form signed by the participants. This study was not preregistered.

Results and Discussion

The primary goals of Study 2 were to test the effects of books’ supportive textual features and caregivers’ elaborative talk on children’s recall of book facts. We had four primary research questions: do books’ supportive textual features influence children’s recall; do books’ supportive textual features influence caregivers’ elaborative talk; and do books’ supportive textual features and caregivers’ elaborative talk interact to influence children’s immediate and 1-week delayed recall.

We analyzed the data using mixed-effect linear regression models in the lme4 package in R (Bates et al., 2015), using Type III Wald F-test with Kenward-Roger degrees of freedom. These models can be used to analyze categorical and continuous data together and can handle missing data (see Stimuli section for data removed due to participant’s previous exposure to books). For each linear mixed effect model, we initially modeled the maximal random effect structure, which included modeling the random slopes of each measure and interaction between measures, as well as each model’s random intercept. However, modeling the maximal random effect structure can lead to convergence issues due to problems such as overfitting (Brauer & Curtin, 2018). When this occurred, we removed the random slopes of the interaction terms and then the random slope of each measure. Below, we report all random effects modeled. If a random effect is not reported, it was removed due to convergence issues.

Prior to analyses, we examined the distribution of data and removed participants from analyses if their data was more than 2.5 standard deviations from the mean on a given measure. This resulted in the removal of participants from analyses involving recall task performance (1) and Elaborative talk (1). Correlations among all measures are presented in Appendix Table B.1 and full results from the analyses from Questions 1–4 are presented in Appendix Tables B.2–B.5.

Question 1: Textual Features on Children’s Immediate Recall

We first assessed whether books’ levels of Cohesion and Embedded Questions influenced children’s recall of the book facts based on Question Type while controlling for Verbal Comprehension. This analysis included 37 participants (1 recall score outlier was removed). The dependent variable was proportion correct with predictor variables of Cohesion (Low, High, dummy coded), Embedded Questions (Low, High, dummy coded), and Question Type (Specific-Fact, Theme, dummy coded), with interactions between these 3 variables, and the control variable of Verbal Comprehension (mean-centered). We modeled the by-subject random slopes of Cohesion, Embedded Questions, and Question Type, and by-subject random intercept. The data are plotted in Figure 2. We found significant effects of Cohesion (b = .13, F_{1, 170.16} = 5.42, p = .021), Question Type F (b = −.14, F_{1, 125.05} = 4.59, p = .034), and Verbal Comprehension (b = .01, F_1,34.50= 23.77, p <.001), with Embedded Questions approaching significance (b = .11, F_1,1,151.40 = 3.42, p =.066). These effects were subsumed by interactions of Cohesion x Embedded Questions (b = −.22, F_1,140.45 = 7.27, p =.008) and Cohesion x Question Type (b = .27, F_1,136.94 = 11.23, p = .001). No other interactions reached significance (ps > .389).

Figure 2.

Children’s proportion correct on science fact immediate recall test based on the books’ Cohesion and Embedded Questions, as well as Question Type. Error bars are plotted as 95% confidence intervals of the mean.

To follow up on the interaction between Cohesion x Embedded Questions, we split the data by Cohesion, collapsing across Question Type, and ran linear mixed-effects models on the effect of Embedded Questions. We modeled the by-subject random intercepts for each model. For the High Cohesion model, there was no significant effect of Embedded Questions (b = −.05, F_1,34.08 = .85, p = .362); however, for the Low Cohesion model, there was a significant effect (b = .15, F_1,35.48 = 10.84, p = .002). As indicated by these results and shown in Figure 2, children had better recall of the high Cohesion than low Cohesion books. When Cohesion was high, Embedded Questions did not significantly influence recall. However, when Cohesion was low, children recalled more if Embedded Questions was high. The lowest recall occurred for the book that was low in both Cohesion and Embedded Questions. This suggests that including supportive textual features of Cohesion and Embedded Questions helps children learn more science facts.

To follow up on the interaction between Cohesion x Question type, we split the data by Question Type, collapsing across Embedded Questions, and ran linear mixed-effects models on the effect of Cohesion. We modeled the by-subject random intercepts for each model. The follow-up analyses showed that there was an effect of Cohesion for the Theme questions (b = .32, F_{1, 36.00} = 66.05, p <.001), but no significant effect for the Specific-Fact questions (b = .04, F_1,36.00 = 1.18, p = .284). As indicated by these results and suggested in Figure 2, children performed better on the Theme questions when Cohesion was high. However, their performance on the Specific-Fact questions did not differ based on Cohesion. These findings suggest that books high in Cohesion facilitate children’s learning of the main points from the science books, facts that can generalize across contexts.

Question 2: Textual Features on Caregivers’ Elaborative Talk

We next assessed whether the books’ levels of Cohesion and Embedded Questions influenced the caregivers’ elaborative extratextual talk while controlling for children’s verbal comprehension. This analysis included 37 participants (1 elaborative talk outlier was removed). We used a linear mixed effects model. The dependent variable was proportion Elaborative talk, with the predictor variables of Cohesion (Low, High), Embedded Questions (Low, High), with interactions between these 2 variables, and the control variable of children’s Verbal Comprehension (mean-centered). We modeled the by-subject random slope for the effects of Cohesion and Embedded Questions and the by-subject random intercept. We found a significant effect of Cohesion (b = −.09, F_1,66.87 = 11.50, p = .001). No other effects approached significance (ps > .191). These findings show that overall caregivers provided proportionally more Elaborative talk when the books were low in Cohesion, suggesting that they were making up for the books’ lack of cohesion by providing their own elaborations (Figure 3).

Figure 3.

Proportion Caregivers’ Elaborative Talk based on books’ Cohesion and Embedded Questions. Error bars are plotted as 95% confidence intervals of the mean.

Question 3: Textual Features and Caregivers’ Elaborative Talk on Children’s Immediate Recall

Next, we investigated whether the books’ supportive textual features interacted with caregivers’ elaborative talk to influence children’s recall. This analysis included 36 participants (2 exclusions: 1 recall outlier and 1 elaborative talk outlier). For this analysis, the dependent variable was proportion correct, with the predictor variables of Cohesion (Low, High), Embedded Questions (Low, High), and Elaborative Talk (mean-centered), with the interactions between these 3 variables, while controlling for Question Type (Specific-Fact), and Verbal Comprehension (mean-centered). We treated Question Type as a control variable in this analysis because of multicollinearity issues. We modeled the by-subject random slope of Cohesion and Embedded Questions and the by-item random intercept. We found significant effects of Cohesion (b = .26, F_1,116.26 = 26.60, p < .001) and Verbal Comprehension (b =.01, F_1,33.55 =19.61, p <.001). We also found a significant interaction between Embedded Questions x Elaborative Talk (b = 1.25, F_1,132.26 = 7.64, p = .007). No other effects reached significance (ps > .130).

To follow up on the interaction between Embedded Questions x Elaborative Talk, we ran correlations on the effects of Elaborative talk, collapsing across Cohesion and Question Type, for each level of Embedded Questions. For the high Embedded Questions books, there was a significant correlation, r(34) = .49, p = .002, whereas for the low Embedded Questions books, the correlation was not significant, r(34) = .18, p = .293. As indicated by these results and seen in Figure 4, when Embedded Questions was High and caregivers used more elaborative talk, children recalled more facts from the books. When Embedded Questions was low, the relation between Elaborative talk and children’s recall was not significant. Overall, these results suggest that children recalled more facts from books that were High in Cohesion, irrespective of caregivers’ Elaborative Talk. Embedded Questions was influential when the caregivers elaborated on the book concepts.

Figure 4.

Relation between Proportion Caregivers’ Elaborative Talk and Children’s Proportion Correct on the science fact recall test based on books’ Cohesion and Embedded Questions. EQ = Embedded Questions. Error bars are plotted as 95% confidence intervals of the regression line.

Question 4: 1-week Delayed Recall of Book Facts

In addition to investigating factors influencing children’s recall of science facts from books read during the session, we also investigated whether children retained facts from the books read in Session 1 over a 1-week delay. As a reminder, in Session 2, children were only asked the Theme questions from the books read in Session 1. We compared performance on the Theme questions across the two sessions. We investigated whether the books’ supportive textual features and caregivers’ Elaborative talk interacted with the delay to influence children’s recall. For this analysis, the dependent variable was proportion correct, with the predictor variables of Cohesion (Low, High), Embedded Questions (Low, High), Elaborative Talk (mean-centered), and Delay (No Delay, Delay), with the interaction between these four variables, while controlling for Verbal Comprehension. This analysis included 36 participants (2 exclusions: 1 recall outlier and 1 elaborative talk outlier). We modeled the by-subject random intercept. We found similar results to the model without delayed recall. There was a significant effect of Cohesion (b = .25, F_1,111.56 =8.63, p =.004), Verbal Comprehension (b = .01, F_1,32.68 = 15.62, p <.001), and interaction between Embedded Questions x Elaborative Talk (b = 2.22, F_{1, 120.40} =8.57, p =.004). No other effects, including the effect of and interaction with Delay, reached significance (ps >.085). As indicated by these results and seen in Figure 5, children learned novel science facts through the influence of books’ supportive textual features and interactions with caregivers’ elaborative talk, and they retained the information over the one-week delay.

Figure 5.

Children’s proportion correct on the Theme Questions from the science fact recall tests (immediate, 1-week delay) based on the books’ Cohesion and Embedded Questions, as well as Delay. Note: 1-week delay recall questions were only asked for two of the four books. The data reflects immediate and delayed science fact recall for the two books that were probed at both time points. Error bars are plotted as 95% confidence intervals of the mean.

Overall, in Study 2, we found that the books’ textual features of Cohesion and Embedded Questions influenced children’s early science learning and caregivers’ extratextual talk, as well as interacted with caregivers’ extratextual talk to support such learning. We found that within the context of shared book reading, young children learned more science facts from books high in Cohesion and/or Embedded Questions. Cohesion particularly influenced children’s recall of science facts related to the central theme of the book but did not influence the recall of specific details. In addition to influencing recall, Cohesion affected extratextual talk as caregivers produced proportionally more Elaborative phrases when the books were low in Cohesion. This suggests that caregivers were compensating for the books’ lack of Cohesion with their own cohesive talk. When considering textual features and elaborative extratextual talk together, we found that Cohesion remained a significant predictor of children’s recall, irrespective of caregivers’ Elaborative talk. Embedded Questions interacted with Elaborative talk such that when books were high in Embedded Questions and caregivers used proportionally more Elaborative talk, children recalled more facts from the books. Lastly, we found that there was no significant decay in recall after the delay. These findings are significant in suggesting that reading high-quality books that align with cognitive science research on reading and learning can influence children’s early science learning. They are also significant in highlighting the importance of considering both the role that books’ textual features and the social interactions surrounding shared book reading play when evaluating the impact of early childhood science books.

General Discussion

In the current research we investigated shared book reading of science books as a means of supporting young children’s science knowledge acquisition prior to formal schooling. Past research on shared book reading focused on preschool-aged children and identified that caregivers’ elaborative extratextual talk supported their children’s language and literacy skills (e.g., Breit-Smith et al., 2017; Hindman et al., 2008). Research on textual features of books primarily with adults and school-aged children identified cohesion and embedded questions as important textual features that support comprehension and learning from books, including science books (Heisey & Kucan, 2010; e.g., Linderholm et al., 2000; van den Broek, 2010). The current research combined insights from these distinct lines of research to investigate whether preschool-aged children’s science books are optimally designed in ways to support early science learning. Additionally, we investigated the extent to which the textual features of the books interact with caregivers’ extratextual talk to support young children’s recall of science facts. To address these questions, in Study 1, we conducted a corpus analysis of readily available preschool-aged children’s books on common science topics. We coded the books for the extent to which their textual features were Cohesive and contained Embedded Questions. In Study 2, we conducted a behavioral study in which caregivers read science books, which varied in levels of Cohesion and Embedded Questions, to their preschool-aged children. We investigated whether and how the books’ textual features interacted with caregivers’ Elaborative extratextual talk to support their children’s immediate and 1-week delayed recall of the books’ science facts.

Overall, our results showed that shared book reading of science books can be a valuable source for supporting children’s recall of science facts when the books were high in Cohesion and/or Embedded Questions. Specifically, in Study 1, we found that science books readily available in young children’s environments varied in levels of Cohesion and Embedded Questions. However, despite the potential benefits of these textual features, most books fell into categories characterized by low levels of one or both dimensions. In Study 2, consistent with our hypotheses, we found that the textual features of Cohesion and Embedded Questions influenced children’s science learning and caregivers’ extratextual talk, as well as interacted with caregivers’ extratextual talk to support science learning. These effects were present when controlling for children’s verbal comprehension.

In terms of Cohesion, Cohesion emerged as an especially important textual feature supporting early learning. When books were high in Cohesion, children recalled and retained more science facts. Specifically, Cohesion enhanced the learning of facts related to the central theme of the books, facts that could generalize across contexts. Children learned more from books high in Cohesion irrespective of books’ levels of Embedded Questions as well as the proportion of the caregivers’ Elaborative talk. That is, it did not matter whether the books were high in Embedded Questions or if caregivers used proportionally more Elaborative talk. Children recalled and retained more facts relative to the books low in Cohesion. These findings are consistent with research on text cohesion with adults and school-aged child readers as cohesion reduces the need to have high prior knowledge, lowers demands on working memory, and lessens the need to engage in deliberate reading strategies (McNamara et al., 1996; van den Broek et al., 1999; van den Broek & Helder, 2017). For the young children, when the books were high in Cohesion, the books likely were sufficient in explaining concepts and drawing connections between factual elements to support children’s learning. There may not have been a need for Embedded Questions or caregivers’ elaborative talk to further support learning.

In contrast, when books were low in Cohesion, children’s learning depended on the books’ Embedded Questions but not on caregivers’ Elaborative talk. Specifically, children learned more from the book low in Cohesion when Embedded Questions was high. This finding is showing that questions support young children’s science learning from books (Heisey & Kucan, 2010; Leung, 2008). It is also consistent with research with adults suggesting that embedding questions within the text are particularly helpful for individuals who have trouble forming a coherent mental representation (Callender & McDaniel, 2007).

Unlike Embedded Questions, caregivers’ elaborative talk did not interact with Cohesion to support children’s learning. This is surprising considering that caregivers used proportionally more Elaborative talk with books low in Cohesion, suggesting that they were implicitly or explicitly compensating for the books’ lack of Cohesion with their own Elaborative talk. Additionally, the proportion of caregivers’ Elaborative talk overall related to children’s recall (Appendix Table B.1) and has been found to relate generally to language and literacy skills (e.g., Robertson & Reese, 2017). One possibility for the lack of relation between books’ Cohesion and caregivers’ Elaborative talk is that the caregivers on average had low prior knowledge of the books’ facts. Thus despite an increased proportion of Elaborative talk, their Elaborative talk did not sufficiently make up for the books’ lack of Cohesion to support their children’s learning. Within research on teaching, it is considered important for teachers to have conceptual knowledge of science concepts to support children’s comprehension when reading early childhood science books (Hoffman et al., 2015).

Unlike for Cohesion, Embedded Questions interacted with Elaborative talk to support children’s recall of the books’ science facts. When books were high in Embedded Questions and caregivers used proportionally more Elaborative talk, children recalled more science facts from the books. Embedded Questions during book reading have been found to keep dyads on task when reading (Troseth et al., 2020). Additionally, although not investigating book reading, research on testing effects suggests that questions are effective in directing learners’ attention to relevant materials (e.g., Little & Bjork, 2016). The Embedded Questions from our books were on central facts. Thus, Embedded Questions likely interacted with the caregivers’ Elaborative talk because it helped to cue the caregivers into elaborating on central facts within the book.

Lastly, we found that children retained general science information that they learned from the books for at least a 1-week delay. That is, we did not find an effect of delay on children’s recall performance for the theme questions (specific-fact questions were not tested at the 1-week delay). These findings are similar to research with elementary school-aged children showing that they could retain information learned over long-term delays from being read science books or participating in science literacy interventions (Kelemen et al., 2014; Mantzicopoulos et al., 2013). Overall, these findings are promising in suggesting that reading high-quality books not only helps the child encode the information learned but also helps them retain that information, likely facilitating the development of their science knowledge base.

In sum, our findings make several contributions to research in the domains of books’ textual features, shared book reading, and early science learning. In terms of textual features, we successfully adapted and applied past research on adults’ and school-aged children’s books (e.g., Callender & McDaniel, 2007; van den Broek, 2010) to characterize the quality of preschool-aged children’s books for supporting early science knowledge. In terms of shared book reading, we demonstrated how the textual features of books and social interactions surrounding book reading jointly influence young children’s early learning. Most prior research focused on how one dimension or the other influenced learning (e.g., Breit-Smith et al., 2017; Heisey & Kucan, 2010). However, learning, including from books, is intrinsically shaped by multilevel factors that interact with each other (Rogoff & Lave, 1984; L. B. Smith & Thelen, 2003).

In terms of early science learning, we extended past research on corpus analyses with children’s science books (Smolkin et al., 2009; Trundle et al., 2008) and showed that many books were not optimally designed to support learning. Despite this, our findings add to the growing body of research demonstrating that book reading can be a valuable source for early science learning (e.g., Ganea et al., 2011; Kelemen et al., 2014) when the books are high in Cohesion and/or Embedded Questions. Additionally, past research focused on what science content children learn from books (e.g., Kelemen et al., 2014) and how caregivers use extratextual talk while reading science books (e.g., Anderson et al., 2004; Robertson & Reese, 2017). We highlighted that to understand science learning from books, research also should consider that both the textual features of early childhood science books and caregivers’ extratextual talk are critical factors affecting early science learning during shared book reading. In the sections to follow, we discuss the limitations of the current research and propose directions for future research.

Limitations and Future Directions

The current research provided novel insights into early childhood science learning during shared book reading. However, this work is not without limitations. One limitation is that we did not causally manipulate the textual features of the books to test their impact within Study 2. Instead, we used published science books based on how the books fell within our corpus analysis. The benefit of this approach is that the books used in the study are reflective of the types of books children may encounter in their naturalistic environments. However, the lack of experimental control warrants recognition that there may be factors beyond Cohesion and Embedded Questions that influenced the results. For instance, although the books were on similar topics of animal facts, we were unable to evaluate whether children may have had differential interests in the specific animal topics. Another limitation is that the caregivers in Study 2 were overwhelmingly female, highly educated, and English speakers living in the United States. This limits the generalizability of our findings and thus the findings cannot be assumed to apply to male caregivers and children from diverse socioeconomic, cultural, and linguistic backgrounds. For example, in terms of gender differences, a few studies found gender differences in caregivers’ extratextual talk styles (Anderson et al., 2004; Vandermaas-Peeler et al., 2012). For instance, Anderson et al. found that male caregivers tended to have more interactions with their children during shared book reading but produced similar numbers of elaborations. Additionally, in terms of socioeconomic status, prior research found that elaborative talk styles tend to vary based on socioeconomic statuses (Heath, 1982; Mol & Neuman, 2014). It thus would be important for future research to both causally manipulate Cohesion and Embedded Questions within early childhood science books as well as to test whether caregivers’ gender and dyads’ socioeconomic, cultural, and linguistic backgrounds relate to shared book reading interactions and children’s science learning from the books.

With regard to other avenues for future research, it will be important to consider other features of books as well as individual differences in cognitive and motivational factors among children and their caregivers that may support early learning. In terms of features of books, it will be valuable to test other textual features beyond Cohesion and Embedded Questions including, but not limited to, picture alignment and book style. Picture alignment is the extent to which the pictures within books are congruent with the text. Research in the context of narrations showed that children recall more story elements from books when the books included illustrations and the pictures were congruent with the text (Greenhoot et al., 2014; Takacs & Bus, 2018). Children’s eye-gaze pattern when looking at congruent illustrations suggests that they spontaneously integrate the illustrations with the text (Takacs & Bus, 2018). Additionally, when narrations include illustrations, caregiver-child dyads make more book references, including references to textual content (Greenhoot et al., 2014). Whereas past research has not focused on expository books, the alignment of text with illustrations is likely an additional feature that supports the coherence of text to facilitate young children’s science learning.

Another important element is book style, whether the book is presented within an expository (factual book without narrative elements) or informational narrative style (factual book with narrative elements). The current research was focused on expository books. Science information is also commonly presented in informational narrations and such book style may have a differential influence on children’s learning in the context of shared book reading. For instance, research with school-aged children found that children comprehended more information from the text that they read when the text was presented in a narrative rather than expository style. Comprehension of expository texts depended on the children’s word knowledge and inferencing skills (Best et al., 2008; Eason et al., 2012). We found congruent effects such that children’s verbal comprehension strongly related to their fact recall with our expository science books. Additionally, research with adults specifically compared informational narrations to expository texts and found no overall differences in learning between the texts, but found that prior knowledge related more strongly to learning from expository than informational narrations books (Wolfe & Mienko, 2007). Taken together, it will be important to understand how children learn science information from informational narrations due to the books’ differential cognitive demands.

In addition to textual features, it is important to consider individual differences in cognitive and motivational factors influencing children’s science learning. For instance, we hypothesized that young children as an age group benefitted from books’ high in Cohesion and/or Embedded Questions because they have limited prior knowledge, difficulty making inferences, low working memory capacity, and tend not to spontaneously engage in deliberate memory strategies (e.g., Cain et al., 2001; Coffman et al., 2008; Cowan, 2014). However, individual differences among these factors may account for some of the variability in children’s science learning from the books. That is, children higher on these factors may readily recall and retain information from books and depend less on supportive textual features. Another important individual difference factor could be children’s intrinsic motivation. A large body of research on school-aged children showed that intrinsic motivation is an important factor contributing to children’s book comprehension as well as learning in STEM domains (e.g., Gottfried et al., 2013; Logan et al., 2011). Children who have a higher interest in book reading, as well as interest in the STEM topics of the books, may have learned more information through their engagement and interest during the shared book reading sessions. In addition to children, there may also be individual differences among caregivers such as their background knowledge or accuracy in describing science information (Mills et al., 2021). In furthering, the understanding of the influence of shared book reading on children’s early science learning, it thus will be important to consider multiple factors influencing learning at the level of the artifacts, the child, and caregiver.

Conclusion

In conclusion, in the current research, we conducted a corpus analysis to investigate the prevalence of high-quality science books in children’s environments and conducted a behavioral experiment to investigate how the textual features of books influence caregivers’ extratextual talk and interact with such talk to support children’s early science learning. Through this investigation, we identified that most science books readily available in young children’s environments were not optimally designed in ways to support early science learning. However, when science books include supportive textual features of Cohesion and Embedded Questions, the books can be a valuable tool for supporting early science learning in the context of shared book reading. This work makes several novel contributions to elucidating the multiple interacting factors that support children’s early science learning from book reading.

Public Significance Statement.

One of the ways young children learn is from books, and children’s books are a potentially good way to support science learning. However, in our research, we found that many science books for young children were not optimally designed to support learning as the books included few supportive features, such as examples and questions for children to answer. Fortunately, we also found that when books include more of these supportive features, and the caregivers reading the books use supportive talk, children learned more science facts from the books and also remember more one week later.

Appendix A

Table A.

Books used in Corpus Analysis from Study 1

Book Name	Book Author
A Butterfly Is Patient	Dianna Hutts Aston
A Tree is a Plant	Clyde Robert Bulla
Air is All Around You	Franklyn M Branley
Amazing Airplane	Tony Mitton and Ant Parker
Apples	Gail Gibbons
Biggest, Strongest, Fastest	Steve Jenkins
Boats are busy	Sara Gilingham
Bugs are Insects	Anne Rockwell
Capacity	Henry Pluckrose
Commotion in the Ocean	Giles Andreae
Cranes	Amanda Askew
Digger, Dozer, Dumper	Hope Vestergaard
Dinosaurs	Kathy Weidner Zoehfeld
Earthquakes	Ellen J. Prager
Eyes, Nose, Fingers, and Toes	Judy Hindley
Flashing Fire Engines	Tony Mitton
Fly Guys Presents Dinosaurs	Tedd Arnold
Four Seasons Make a Year	Anne Rockwell
From Seed to Plant	Gail Gibbons
From Seed to Pumpkin	Wendy Peffer
Here We Are: Notes for Living on Planet Earth	Oliver Jeffers
How Many Teeth?	Paul Showers
How to Hide an Octopus & Other Sea Creatures	Ruth Hellers
I Get Wet	Vicki Cobb
I’m Growing	Aliki
Insect Detective	Steve Voake
Look at Me: How to Attract Attention in the Animal World	Steve Jenkins, Robin Page
My Awesome Summer	Paul Meisel
My Feet	Aliki
My Five Senses	Aliki
My Hands	Aliki
My, Oh My, a Butterfly!	Trish Rabe
On Beyond Bugs: All About Insects	Trish Rabe
Over and Under the Pond	Kate Messiner
Plants feed me	Lizzy Rockwell
Roaring Rockets	Tony Mitton and Ant Parker
Sea Turtles	Gail Gibbons
Seashells by the Seashore	Marianne Berkes
Seed to Plant	Kristin Baird Rattini
Shooting Stars	Martha E. H. Rustad
Sleep Is for Everyone	Paul Showers
Snow Is Falling	Franklyn M Branley
Solar System	Jill McDonald
Sounds All Around	Wendy Peffer
Storms	Mariam Busch Goin
Sunshine Makes the Seasons	Franklyn M Branley
Super Submarines	Tony Mitton and Ant Parker
The Cars and Trucks Book	Todd Parr
The Sun and the Moon	Carolyn Cinami DeCristofano
The Tiny Seed	Eric Carle
The Tooth Book	Theo LeSieg and Dr. Seuss
The Vegetables We Eat	Gail Gibbons
There’s No Place Like Space!	Tish Rabe
Time	Henry Pluckrose
Tornadoes!	Gail Gibbons
Weather National Geographic	Kristin Baird Rattini
What Lives in a Shell	Kathleen Weidner Zoehfeld
What makes a shadow?	Clyde Robert Bulla
Whose Food	Nancy Kelly Allen
Why Should I Recycle?	Jen Green

Appendix B

B.1.

Correlation table of measures

		Specific-Fact Immediate Recall	All Immediate Recall	Theme 1-week Delay Recall	Elaborative Talk	Verbal Comprehension
All Books	Theme Immediate Recall	.38*	.77***	.48**	0.14	.64***
	Specific-Fact Immediate Recall		.84***	.56***	.44**	.38*
	All Immediate Recall			.57***	.40*	.66***
	Theme 1-week Delay Recall				.08	.43**
	Elaborative Talk					.31t
High Cohesion-High Embedded Questions	Theme Immediate Recall	0.19	.78***	.72**	.33t	.65***
	Specific-Fact Immediate Recall		.76***	.35	.29	.24
	All Immediate Recall			.65**	.40*	.58***
	Theme 1-week Delay Recall				.49t	.28
High Cohesion-Low Embedded Questions	Theme Immediate Recall	0.23	.78***	0.41t	−.09	.40
	Specific-Fact Immediate Recall		.79***	.02	.49	.16
	All Immediate Recall			.27	.26	.35
	Theme 1-week Delay Recall				−.09	.51*
Low Cohesion-High Embedded Questions	Theme Immediate Recall	.38*	.80***	.32	.40*	.45**
	Specific-Fact Immediate Recall		.86***	.29	.27	.52**
	All Recall			.38t	.40*	.59***
	Theme 1-week Delay Recall				.36	.40t
Low Cohesion-Low Embedded Questions	Theme Immediate Recall	0.24	.69***	.48t	−.12	.38*
	Specific-Fact Immediate Recall		.87***	.56*	−.17	.32
	All Immediate Recall			.48t	−.19	.43**
	Theme 1-week Delay Recall				−.51t	.18

Notes. The number of participants varied across measures due to 1) some participants being excluded due to having read the book previously and 2) only 50% of participants received 1-week delayed recall questions per book.

^tp = <.1

^*<.05

^**<.01

^***<.001.

B.2.

Linear Mixed Effect Model Table for Study 2 Question 1

Measures	b	F	df residual	p
Cohesion	0.13	5.42	170.16	.021
Embedded Questions	0.11	3.42	151.41	.066
Question Type	−0.14	4.59	125.08	.034
Verbal Comprehension	0.01	23.77	34.50	<.001
Cohesion X Embedded Questions	−0.22	7.27	140.45	.008
Cohesion X Question Type	0.27	11.23	136.94	.001
Embedded Questions X Question Type	0.07	0.75	135.77	.389
Cohesion X Embedded Questions X Question Type	0.03	0.10	137.80	.758

B.3.

Linear Mixed Effect Model Table for Study 2 Question 2

Measures	b	F	df residual	p
Cohesion	−0.09	11.50	66.87	.001
Embedded Questions	0.02	0.46	67.45	.502
Verbal Comprehension	0.00	1.78	34.71	.191
Cohesion x Embedded Questions	−0.03	0.69	35.11	.411

B.4.

Linear Mixed Effect Model Table for Study 2 Question 3

Measures	b	F	df residual	p
Cohesion	0.26	26.60	116.26	<.001
Embedded Questions	0.08	2.33	116.65	.130
Elaborative Talk	−0.42	2.00	57.36	.162
Question Type	0.03	0.96	167.25	.329
Verbal Comprehension	0.01	19.61	33.55	<.001
Cohesion X Embedded Questions	−0.09	1.52	227.51	.219
Cohesion X Elaborative Talk	0.62	2.29	113.11	.133
Embedded Questions X Elaborative Talk	1.25	7.64	132.26	.007
Cohesion X Embedded Questions X Elaborative Talk	−0.53	0.62	179.44	.432

B.5.

Linear Mixed Effect Model Table for Study 2 Question 4

Measures	b	F	df residual	p
Delay	0.06	0.54	90.62	.463
Cohesion	0.25	8.63	111.56	.004
Embedded Questions	0.05	0.30	119.55	.586
Elaborative Talk	−0.90	3.02	119.96	.085
Verbal Comprehension	0.01	15.62	32.68	<.001
Delay X Cohesion	0.08	0.50	90.62	.483
Delay X Embedded Questions	−0.02	0.03	90.62	.875
Delay X Elaborative Talk	−0.44	0.47	90.62	.497
Cohesion X Elaborative Talk	0.20	0.10	115.40	.756
Embedded Questions X Elaborative Talk	2.22	8.57	120.40	.004
Delay X Cohesion X Embedded Questions	−0.15	0.69	90.62	.409
Delay X Cohesion X Elaborative Talk	1.10	1.77	90.62	.187
Delay X Embedded Questions X Elaborative Talk	−0.11	0.01	90.62	.908
Cohesion X Embedded Questions X Elaborative Talk	−0.26	0.06	110.49	.808
Delay X Cohesion X Embedded Questions X Elaborative Talk	−1.04	0.55	90.62	.460